George 

Sidebotham

Heat
Transfer
Modeling
An Inductive Approach
Heat Transfer Modeling
George Sidebotham

Heat Transfer Modeling

An Inductive Approach
George Sidebotham
Mechanical Engineering Department
The Cooper Union for the Advancement
of Science and Art
New York, NY, USA

ISBN 978-3-319-14513-6 ISBN 978-3-319-14514-3 (eBook)

DOI 10.1007/978-3-319-14514-3

Library of Congress Control Number: 2015930196

Springer Cham Heidelberg New York Dordrecht London

# Springer International Publishing Switzerland 2015
This work is subject to copyright. All rights are reserved by the Publisher, whether the whole or part
of the material is concerned, specifically the rights of translation, reprinting, reuse of illustrations,
recitation, broadcasting, reproduction on microfilms or in any other physical way, and transmission or
information storage and retrieval, electronic adaptation, computer software, or by similar or dissimilar
methodology now known or hereafter developed.
The use of general descriptive names, registered names, trademarks, service marks, etc. in this
publication does not imply, even in the absence of a specific statement, that such names are exempt
from the relevant protective laws and regulations and therefore free for general use.
The publisher, the authors and the editors are safe to assume that the advice and information in this book
are believed to be true and accurate at the date of publication. Neither the publisher nor the authors or the
editors give a warranty, express or implied, with respect to the material contained herein or for any errors
or omissions that may have been made.

Printed on acid-free paper

Springer International Publishing AG Switzerland is part of Springer Science+Business Media

(www.springer.com)
My wife, Judy Wirkula, and our growing
family of dogs have inspired me with their
patient and consistent love, and I dedicate
this work to them.
Preface

There are many excellent textbooks on Heat Transfer. Why another?

In the course of teaching Heat Transfer to junior level Chemical and Mechanical
Engineering students over several years and developing reams of notes, it became
difficult for me to conduct classes in parallel with a traditional textbook. The order
of topics and general approach I wanted to emphasize in my courses evolved to a
point that deviated too much from any textbook I could find. So I used a sabbatical
leave to write my notes up in the form of a textbook.
Unlike traditional texts, I wanted to follow the adage that “Heat Transfer takes
time” and consider transient problems first, with steady-state applications being
treated as a limiting case. I also wanted to introduce each topic with readily
recognizable applications and build quantitative methods to analyze them as
needed. This perhaps controversial “just in time” approach means that the order
of development deviates from the accepted norm to some degree. The goal is that by
showing a few specific examples in some detail and then drawing general
conclusions, the reader will be in a position to develop his or her own approach
to situations that arise in his or her practice. This approach is considered to be
inductive, in contrast to a deductive approach in which general principles are taught
first. Heat transfer is an invisible, massless, abstract phenomenon and it may help
some students to visualize its nature with a tangible example first. I credit my
mother, a religious educator, for teaching me the inductive method.
Only a few problem statements are given in each chapter, some of which are
analyzed in detail as part of the development of the analysis methodology (with a
suggestion given for the reader to attempt a solution before proceeding). These are
called workshops, not problems, and form the basis for how I teach my courses. I
use a part of class to lecture and a large part of class time for students to work
together on problems that use those concepts right away. My hope is that instructors
that adopt this textbook will write their own problem statements or draw from the
many resources available to them. Better yet, I have found that students learn a lot
and often enjoy making up their own problem statements.
A simple, recognizable situation, namely a cooling mug of coffee, is introduced
early with an experimental solution and revisited repeatedly. A new modeling
technique is introduced with each iteration, and virtually all traditional first level
heat transfer concepts are thereby developed. The patient reader acquires a toolbox

vii
viii Preface

that can be applied to a broad range of applications. There naturally arises,

however, a certain redundancy, and some topics are treated in multiple places,
but with a different twist each time. This characteristic of an inductive method is
consistent with the truth that “learning is an iterative process”.
All of those variations are centered on the science, the knowing and/or predicting
how quickly a mug of coffee cools. However, it is not engineering in the sense that,
other than informing me how long I’ll need to wait (the science), some human
intervention is made to address a need (the engineering). I find it curious that, in this
modern high tech world, when you buy a cup of coffee, it is usually too hot to drink
when you get it. So you have to wait. Or add a few ice cubes to cool (and dilute)
it. Furthermore, if you get distracted, the coffee becomes tepid.
Then it was brought to my attention that a pair of visionary young entrepreneurs
(Dave and Dave) had engineered a solution and had brought “Coffee Joulies™” to
market. So, with their blessing, I’ve added sections to appropriate chapters that
demonstrate the principles involved. This example provides a good application that
involves phase change, something that is generally not included prominently in
traditional heat transfer texts. A testimonial of mine could be “I like Coffee Joulies
so much, I wrote a book about them!”Another common situation is the opposite,
namely if I put a warm beverage in a refrigerator, how long do I wait before it’s
cold? This problem, and an engineering solution called the Cooper Cooling process
invented by Greg Loibl, a Cooper Union masters student of mine, is treated in
workshops.
While addressing key traditional fundamental concepts in heat transfer practice,
the emphasis is on simplified modeling philosophies, which carry into all walks of
life. If it is determined that simplified analytical solutions are considered to be
insufficient to address a specific need, basic numerical methods are preferred to
complex analytical solutions. The goal is to develop the art of using theory as a
design tool. Theory for theory’s sake can obfuscate sound engineering judgment.
The practicing engineer must navigate a diverse array of sometimes inconsistent
and/or conflicting nomenclature and definitions, so a good foundation in Engineer-
ing Thermodynamics is considered a prerequisite to this textbook. In the so-called
thermal applications, terms like energy, heat, work, and temperature have specific
meanings. Heat transfer is defined as the transport of energy across a system
boundary driven by a spatial temperature gradient. Heat is a transported form of
energy and is always associated with a boundary that has a surface area. Other
forms of energy (internal, kinetic, and potential) are “storable” and are associated
with a region of space that has a volume. Temperature is not a form of energy and is
difficult to succinctly define, yet everybody knows what it is (or thinks they do),
some sort of measure of hotness or coldness. Work is another transported energy
form and is associated with a force acting through distance, or in the case of
“electrical work” of electrons moving across the boundary against an electrical
potential difference.
The word “heat” in common usage frequently means something different than
that formal technical definition. My wife often says, “There’s no heat in this house.”
I know what she means, and I know to respond, “I’ll turn up the thermostat. In the
Preface ix

meantime, put on a sweater.” An hour later, the house is toasty. But in my head,
I might say “You’re right. There is technically no heat in this house. But the air
temperature is sufficiently low relative to your body temperature that the rate at
which you are losing heat is uncomfortable to you. Turning up the thermostat will
send a signal for a fuel valve to open and a spark to ignite a flame in the boiler,
where chemical energy will be converted to thermal energy. Then heat will be
transferred from the hot gases to the water in the boiler, which will create steam that
will flow into the radiators (which should be called convector/radiators) throughout
the house. Then heat will be transferred to the air in the room, raising its internal
energy, and thereby the air temperature to the new set point.” I guess it’s easier to
just say “There’s no heat in this house.”
Modeling involves the creation of a simplified view of a complex system.
In general, mathematical modeling (of anything, not just heat) involves two main
steps: model development and its subsequent solution. They are not always purely
sequential, and feedback from one to the other and an iterative solution is usually
required for a good engineering solution. Modeling becomes a design tool when
that solution leads to a decision that addresses the original problem statement, such
as “what material and what dimensions should that heat sink be to keep the
motherboard from overheating?”
Model development involves a derivation of governing equations from the
application of fundamental laws of nature to a defined system, a specific limited
portion of the universe. The art of this stage of modeling is to choose clear system
boundaries and to make simplifying assumptions that ease the subsequent solution,
while still capturing the “controlling physics.” Frequently, complex systems are
controlled by a simple balance between two or three “things.” A determination of
the extent to which that model captures reality is context-specific and a major part
of engineering decision-making. Creation of the model is not the end, it is the start.
The input is a word problem statement, and the output is a mathematical problem
statement.
The next step is to solve the mathematical problem. This step can be trivial, easy,
hard, or impossible, and there may be multiple approaches. It is good design to invoke
a feedback response, a return to the model development with an eye toward either
making additional simplifications or relaxing others, and ultimately arrive at a
solution that is deemed adequate to address the original word statement. It is common
to lose sight of the original problem statement once a detailed analysis is initiated.
The approach of this book is to start with a word problem statement (often with
specific parameter values) and to develop A modeling strategy in detail. Not THE
modeling strategy. A patient and resourceful reader will develop alternative models
to the same initial problem statement, solve those, and then compare (qualitatively
and quantitatively) to the examples given in this text.
Once they are implemented (and generally not before), models can be tested and
then defined as being simple, simplistic, or complex. The testing criterion depends
on the problem at hand and may involve comparison with experiment, parametric
study (observe the effect of various parameters), or testing assumptions after the
fact. Generally, a simple model is most desirable. It does the job with the least effort
x Preface

(and captures what is most important). A simplistic model misses something

important. However, a legitimate, perhaps counterintuitive modeling strategy is to
start as simple as possible, then determine whether the model is simplistic. If so,
a more detailed model, perhaps guided by the first model, can be developed and
then tested.
However, there appears to be a natural tendency to approach things from the
opposite way, namely to make a very complex model right at the start. That first
instinct can do more harm than good. How complex should a model be? Only as
complex as it needs to be, no more. In plumbing, there appears to be a natural
tendency to tighten fittings as much as you can, thinking that’s better. But you can
damage fittings that way and MAKE them leak. So how tight should they be? Only
as tight as they need to be (perhaps a bit more). In other words, a complex model
may seem to be better than a simple one. However, dealing with the complexity can
be costly in time, effort, and money. Worse, the inherent beautiful simplicity of the
underlying controlling aspects of problem that often (usually?) exist may be missed.
The general philosophical approach taken in this text is to start with as simple a
model as possible (usually a “1-Node Model”), learn about the big picture from it,
and then decide whether a more complicated model is needed to do the job. This
strategy can be used in most (if not all) aspects of life. Numerical methods involve
discretizing time to make a prediction of the future and/or discretizing space
to create multinode models. Mathematically, differential equations are converted
to algebraic equations. This textbook therefore is intended to be an introduction to
numerical methods in general and a start toward computational fluid dynamics
(CFD), an important and growing discipline.
The intention in this text is to favor an inductive over a deductive approach, in
which the study of a specific case leads to a generalization that can be applied to
other situations. Most engineering textbooks use a deductive approach in which
general principles are developed and then applied to specific cases. Neither
approach is inherently better than the other. They are just different tools.
All of the problems presented in the main body are solved using EXCEL
spreadsheets, without writing VBA code. All of the problems could also be
conducted by writing code in other programs. A few of the problems encountered
really should NOT be solved using spreadsheets. A general rule is that if a nested
loop is fundamentally required for a problem, program code is probably a better
approach, otherwise whatever tool is most comfortable. It is really good practice
developing programming skills to reproduce some of the solutions presented as part
of the main text.
This textbook, targeted to upper level engineering students (but accessible to all
life-long learners), is intended to be inexpensive to the student budget and a
nurturing challenge to their intellect. It is dedicated to the Cooper Union students
that I have had the honor and privilege of serving over the years. I feel blessed to
have been given a chance to influence, in some incremental way, hundreds of young
women and men on their way to becoming enlightened and loving human beings
(through the intense development of engineering skills).
Preface xi

NOTE TO STUDENTS
This book is written for you. My hope is that many of you will gain confidence and
apply the techniques developed here to situations that you define. You can have fun
and challenge yourself at the same time.

NOTE TO INSTRUCTORS
My challenge to you is to draw from your experience and apply it using principles
developed in this text. If you have taught Heat Transfer many times before, maybe
you will see some things in a new light. The only missing technical element is the
case of thermal radiation between opaque surfaces, which I never really can get to
in a first course. I do emphasize radiation concepts throughout, however. My advice
is to create in-class and homework-based group activities that use numerical
methods early, even before students may have reached those topics in a straight
reading of the text.

NOTE TO LIFE-LONG LEARNERS

If you’ve already taken a course in Heat Transfer and especially if your practice
involves heat transfer, I hope that this textbook will help you hone your skills and
begin to simplify your models and clarify your thought.

New York, NY, USA George Sidebotham

Contents

Part I Modes of Heat Transfer

1 Thermal Circuits . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3
1.1 RC Circuit Driven by a Temperature Source . . . . . . . . . . . . . . 3
1.2 RC Circuit Driven by a Current Source . . . . . . . . . . . . . . . . . . 9
1.3 RC Circuit Driven by Multiple Sources . . . . . . . . . . . . . . . . . . 10
1.4 Resistors in Series . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 13
1.5 Resistors in Parallel . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 15
1.6 Iterative Solution Methods . . . . . . . . . . . . . . . . . . . . . . . . . . . 17
1.7 Numerical Solution 1: Forward Explicit Formulation
(Euler Method) . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 18
1.7.1 Variable Step Size Explicit Methods . . . . . . . . . . . . . . 19
1.8 Numerical Solution 2: Backward Implicit formulation . . . . . . . 20
1.9 Numerical Solution 3: Central Implicit formulation
(Crank–Nicolson Method) . . . . . . . . . . . . . . . . . . . . . . . . . . . . 21
Workshop 1.1: Numerical Solution of First-Order RC Circuit . . . . . . . 22
Workshop 1.2: Numerical Solution of RC Circuit
with Time Varying Ambient Temperature . . . . . . . . . . . . . . . . . . . . . 28
2 Lumped Capacity Systems and Overall Heat
Transfer Coefficients . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 31
2.1 A Cooling Mug of Coffee . . . . . . . . . . . . . . . . . . . . . . . . . . . . 32
2.1.1 Problem Statement . . . . . . . . . . . . . . . . . . . . . . . . . . . 32
2.1.2 Experimental Solution . . . . . . . . . . . . . . . . . . . . . . . . 32
2.2 Lumped Capacity Model Development . . . . . . . . . . . . . . . . . . 36
2.2.1 Conservation of Energy . . . . . . . . . . . . . . . . . . . . . . . 36
2.2.2 Lumping the Stored Energy into a Single Node . . . . . . 37
2.2.3 Thermodynamic Analysis of Mug Heating Phase . . . . . 39
2.2.4 1-Node Lumped Capacity Model . . . . . . . . . . . . . . . . 41
2.2.5 Overall Heat Transfer Coefficient:
Empirical Value . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 42
2.3 A Melting Cup of Ice/Water . . . . . . . . . . . . . . . . . . . . . . . . . . 47
2.3.1 Problem Statement . . . . . . . . . . . . . . . . . . . . . . . . . . . 47
2.3.2 Experimental Solution . . . . . . . . . . . . . . . . . . . . . . . . 48
xiii
xiv Contents

2.3.3 Theoretical Model Prediction: Energy Analysis . . . . . . 49

2.3.4 Results . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 50
Appendix 1. Raw Data for Cooling Mug of Coffee . . . . . . . . . . . . . . . 52
Appendix 2. Measurement of the Mug-Specific Heat . . . . . . . . . . . . . 54
Workshop 2.1. Cooling Coffee Data . . . . . . . . . . . . . . . . . . . . . . . . . 55
Workshop 2.2. Euler Method Simulation of Cooling
Mug of Coffee . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 56
Workshop 2.3. Floating Thermometer Response Time . . . . . . . . . . . . 58
3 Heat Transfer Modes: Conduction, Convection, and Radiation . . . 61
3.1 Example: Single-Paned Window . . . . . . . . . . . . . . . . . . . . . . . 61
3.2 Heat Transfer Modes and Thermal Channels . . . . . . . . . . . . . . 62
3.3 Fourier’s Law of Conduction . . . . . . . . . . . . . . . . . . . . . . . . . . 67
3.4 Newton’s Law of Cooling . . . . . . . . . . . . . . . . . . . . . . . . . . . . 70
3.5 Stefan–Boltzmann and Kirchhoff’s Radiation Laws . . . . . . . . . 72
3.5.1 Case 1: Surrounded Object . . . . . . . . . . . . . . . . . . . . . 75
3.5.2 Case 2: Radiation Source . . . . . . . . . . . . . . . . . . . . . . 76
3.5.3 Case 3: Radiation Between Infinite Plates . . . . . . . . . . 79
3.5.4 Radiation in the Single-Paned Window Problem . . . . . 81
3.6 Single-Paned Window: Results . . . . . . . . . . . . . . . . . . . . . . . . 83
Appendix. Conductive Thermal Resistances
in Various Geometries . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 86
Workshop 3.1. Double-Paned Window Models . . . . . . . . . . . . . . . . . . 89
Additional Workshop Ideas . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 92

Part II Transient Conduction

4 1-Node Transient Models . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 97

4.1 Lumped Capacity Model for Coffee/Mug Problem . . . . . . . . . . 98
4.1.1 Model 1: Neglect Radiation and Evaporation . . . . . . . . 98
4.1.2 Model 2: Add Thermal Radiation . . . . . . . . . . . . . . . . 103
4.1.3 Model 3: Add Evaporation Effects . . . . . . . . . . . . . . . 104
4.2 Heisler and Gurney–Lurie Charts from a 1-Node Lumped
Capacity Analysis . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 107
4.2.1 Example: Quenching a Solid Rod . . . . . . . . . . . . . . . . 108
4.2.2 Application to Quenching a Solid Rod . . . . . . . . . . . . 112
4.3 Thermal Boundary Layers in Thick Walls . . . . . . . . . . . . . . . . 115
4.3.1 Suddenly Heated Thermally Thick Wall . . . . . . . . . . . 115
4.3.2 Application to Cooking a Hamburger . . . . . . . . . . . . . 119
4.3.3 Suddenly Heated Thermally Thick Cylinder . . . . . . . . 124
4.3.4 Thermally Thick Wall Suddenly Exposed
to a Hot (or Cold) Fluid . . . . . . . . . . . . . . . . . . . . . . . 126
Contents xv

4.4 Cooling Mug of Coffee Revisited: Mug Heating Phases . . . . . . 138

4.4.1 Phase 1: Thermally Thick Mug . . . . . . . . . . . . . . . . . . 139
4.4.2 Phase II: Mug as Lumped Capacity Node . . . . . . . . . . 141
Appendix 1. Detailed Calculation of Individual Coffee
Mug Thermal Resistances . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 144
Appendix 2. Parabolic Fits for Plotting Temperature Profiles . . . . . . . . 149
Workshop 4.1. Cooking 1-Node Hot Dogs . . . . . . . . . . . . . . . . . . . . . 152
Workshop 4.2. Beverage Chilling Methods . . . . . . . . . . . . . . . . . . . . 159
5 Few-Node Transient Models . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 163
5.1 A Generic 2-Node RC Network . . . . . . . . . . . . . . . . . . . . . . . . 164
5.2 Nodal Equations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 166
5.2.1 Explicit Numerical Formulation . . . . . . . . . . . . . . . . . 167
5.3 Coffee Mug: Explicit Numerical Solution
with Constant Coefficients . . . . . . . . . . . . . . . . . . . . . . . . . . . . 169
5.3.1 Spreadsheet Program Description . . . . . . . . . . . . . . . . 170
5.3.2 Model Prediction Results . . . . . . . . . . . . . . . . . . . . . . 174
5.3.3 Numerical Accuracy and Stability Issues . . . . . . . . . . . 179
5.3.4 Implicit Numerical Formulation . . . . . . . . . . . . . . . . . 181
5.4 Coffee Joulies™: A 3-Node Model . . . . . . . . . . . . . . . . . . . . . 183
5.4.1 Incorporating a Node with Phase Change . . . . . . . . . . 186
5.4.2 Nodal Equations . . . . . . . . . . . . . . . . . . . . . . . . . . . . 187
5.4.3 Spreadsheet Implementation . . . . . . . . . . . . . . . . . . . . 189
Appendix 1. Nodal Response Time . . . . . . . . . . . . . . . . . . . . . . . . . . 193
Appendix 2. Detailed Thermal Resistance Formulas
for Coffee Mug Problem . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 195
Appendix 3. Joulie Property Estimation . . . . . . . . . . . . . . . . . . . . . . . 199
Workshop 5.1. A Batch Mode Solar Water Heater . . . . . . . . . . . . . . . 201
Workshop 5.2. Cooper Cooling with a Finite Reservoir . . . . . . . . . . . . 205
Additional Workshop Ideas . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 208
6 Multi-Node Transients . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 211
6.1 Multi-Node Model Development . . . . . . . . . . . . . . . . . . . . . . . 212
6.2 Defining the Grid (i.e., Discretization of Space) . . . . . . . . . . . . 213
6.3 Nodal Equations for Explicit Method (Euler) . . . . . . . . . . . . . . 217
6.3.1 Capacitances . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 220
6.3.2 Resistances . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 220
6.3.3 Step-Size Restriction . . . . . . . . . . . . . . . . . . . . . . . . . 223
6.4 Spreadsheet Implementation . . . . . . . . . . . . . . . . . . . . . . . . . . 224
6.5 Temperature/Time Plots . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 225
6.6 Radial Temperature Profiles . . . . . . . . . . . . . . . . . . . . . . . . . . 225
6.7 Heat Transfer Coefficients . . . . . . . . . . . . . . . . . . . . . . . . . . . . 230
6.8 Numerical Stability and Accuracy . . . . . . . . . . . . . . . . . . . . . . 231
6.8.1 Implicit Method . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 233
Workshop 6.1. Transient Heat Transfer Fins . . . . . . . . . . . . . . . . . . . . 234
xvi Contents

Part III Steady-State Conduction

7 Heat Transfer Fins (and Handles) . . . . . . . . . . . . . . . . . . . . . . . . . . 241

7.1 Motivation for Fins . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 241
7.1.1 Perfect Fin . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 243
7.1.2 Corrected Length . . . . . . . . . . . . . . . . . . . . . . . . . . . . 243
7.2 1-Node Fin Analysis . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 243
7.2.1 Thermally Short Fin Model . . . . . . . . . . . . . . . . . . . . 244
7.2.2 Thermally Long Fin Model for Handles . . . . . . . . . . . 247
7.2.3 Thick Fins . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 249
7.2.4 Orientation of Fins . . . . . . . . . . . . . . . . . . . . . . . . . . . 250
7.3 Half a Mug of Coffee . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 251
7.3.1 Pure Lumped Capacity Result . . . . . . . . . . . . . . . . . . . 252
7.3.2 Modified Lumped Capacity Model . . . . . . . . . . . . . . . 253
7.3.3 Rim Details . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 256
7.3.4 Results of Extended 1-Node Model . . . . . . . . . . . . . . . 260
7.4 3-Node Fin Analysis . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 265
Appendix 1. Constant-Area Fins: Differential Solution . . . . . . . . . . . . 270
Appendix 2. Variable Area Fins: Finite Difference
Numerical Solution . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 282
Appendix 3. Iterative Solutions of Nonlinear
Algebraic Equations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 284
Workshop 7.1: Triangular Fins . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 292
Workshop 7.2: Circumferential Rectangular Fin . . . . . . . . . . . . . . . . . 293
8 Steady-State Conduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 295
8.1 Graphical Method . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 296
8.2 Numerical Method for 2D Steady-State Conduction . . . . . . . . . 302
8.2.1 Finite Element Numerical Analysis . . . . . . . . . . . . . . . 303
8.2.2 Boundary Resistances . . . . . . . . . . . . . . . . . . . . . . . . 307
8.2.3 Case Studies . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 319
8.3 Classic 2D Problem Statement . . . . . . . . . . . . . . . . . . . . . . . . . 325
8.3.1 Numerical Analysis . . . . . . . . . . . . . . . . . . . . . . . . . . 325
8.3.2 1  1 Model . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 326
8.3.3 Few-Node Model . . . . . . . . . . . . . . . . . . . . . . . . . . . . 329
8.3.4 n  m Model . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 331
8.3.5 Comment on Exact Solution . . . . . . . . . . . . . . . . . . . . 336
8.3.6 Results with Internal Heat Generation . . . . . . . . . . . . . 337
Workshop 8.1. Graphical Method on Classic 2D Problem . . . . . . . . . . 339
Workshop 8.2. Graphical and Thermal Resistance Practice . . . . . . . . . 340
Workshop 8.3. Graphical Analysis Practice . . . . . . . . . . . . . . . . . . . . 341
Workshop 8.4. Classroom Fun! . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 341
Workshop 8.5. Storage of Energetic Materials—The Chemically
Reacting Pile . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 342
Workshop 8.6. Cold Plate . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 345
Appendix. MATLAB Code for Classic 2D Problem . . . . . . . . . . . . . . 347
Contents xvii

Part IV Open Systems: Convection and Heat Exchangers

9 Nusselt Number Correlations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 351

9.1 Flow Classification and Dimensionless Parameters . . . . . . . . . . 351
9.2 Nusselt Number Example Calculation . . . . . . . . . . . . . . . . . . . 355
9.2.1 Property Values and Their Temperature
Dependency . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 356
9.2.2 Rayleigh Number . . . . . . . . . . . . . . . . . . . . . . . . . . . . 357
9.2.3 Nusselt Number . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 359
9.2.4 Convection Coefficients . . . . . . . . . . . . . . . . . . . . . . . 361
9.2.5 Natural Convection Film Property . . . . . . . . . . . . . . . . 363
9.3 Application to the Mug of Coffee . . . . . . . . . . . . . . . . . . . . . . 365
9.3.1 Value Selection for Constant Convection
Coefficient Models . . . . . . . . . . . . . . . . . . . . . . . . . . . 365
Workshop 9.1. Wind Chill Effect . . . . . . . . . . . . . . . . . . . . . . . . . . . . 369
Workshop 9.2. Forced Convection for Internal Flows . . . . . . . . . . . . . 372
Workshop 9.3. Two-Node Mug of Coffee
with Variable Convection Coefficients . . . . . . . . . . . . . . . . . . . . . . . . 374
10 Convection Fundamentals . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 377
10.1 Boundary Layer Concepts . . . . . . . . . . . . . . . . . . . . . . . . . . . . 378
10.1.1 Newton’s Law of Cooling . . . . . . . . . . . . . . . . . . . . . 382
10.2 Forced Convection Across an Isothermal Flat Plate . . . . . . . . . 385
10.2.1 Momentum Boundary Layer . . . . . . . . . . . . . . . . . . . . 385
10.2.2 Thermal Boundary Layer . . . . . . . . . . . . . . . . . . . . . . 389
10.3 Natural Convection Across an Isothermal Vertical Plate . . . . . . 396
10.3.1 Mass . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 398
10.3.2 x-Momentum . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 398
10.3.3 Energy . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 400
Workshop 10.1. Boundary Layer Thickness
for the Mug of Coffee . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 404
11 Internal Flows Models . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 405
11.1 Thermal Length . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 405
11.1.1 Basic Thermodynamic and Heat Transfer Relations . . . 408
11.2 Example: Water-Tube Boiler . . . . . . . . . . . . . . . . . . . . . . . . . . 412
11.2.1 1-Node Models with Advective Flow
as a Current Source . . . . . . . . . . . . . . . . . . . . . . . . . . 412
11.2.2 Energy Balance . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 413
11.2.3 Thermal Resistance . . . . . . . . . . . . . . . . . . . . . . . . . . 414
11.2.4 Nusselt Number Correlations . . . . . . . . . . . . . . . . . . . 415
11.3 Momentum Considerations: Pressure Loss . . . . . . . . . . . . . . . . 420
11.3.1 Preheater and Superheater Lengths . . . . . . . . . . . . . . . 422
xviii Contents

11.4 Internal Heat Transfer Flow Models . . . . . . . . . . . . . . . . . . . . . 425

11.4.1 Differential 1D Flow Model . . . . . . . . . . . . . . . . . . . . 426
11.4.2 1-Node Flow Models . . . . . . . . . . . . . . . . . . . . . . . . . 428
11.4.3 Pipe Flows with Constant Heat Flux . . . . . . . . . . . . . . 433
Workshop 11.1. Finned Pipe . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 435
Workshop 11.2. Solar Pipe in Continuous Mode: 2-Node Model . . . . . 436
Workshop 11.3. Solar Pipe in Continuous Mode:
Multi-node Model . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 440
12 Heat Exchanger Fundamentals . . . . . . . . . . . . . . . . . . . . . . . . . . . . 445
12.1 Functional Requirements of a Heat Exchanger . . . . . . . . . . . . . 445
12.2 Double-Pipe Heat Exchangers . . . . . . . . . . . . . . . . . . . . . . . . . 447
12.2.1 Averaged Temperature Differences . . . . . . . . . . . . . . . 448
12.2.2 2-Node Analysis: Well-Mixed Model . . . . . . . . . . . . . 450
12.2.3 2-Node Analysis: Average Temperature Model . . . . . . 456
12.2.4 2n-Node Model . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 459
Workshop 12.1. Double-Pipe Heat Exchangers . . . . . . . . . . . . . . . . . . 460
Workshop 12.2. Serving Coffee at the Right Temperature . . . . . . . . . . 474
13 Evaporation and Mass Transfer Fundamentals . . . . . . . . . . . . . . . 475
13.1 Background . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 475
13.2 Model Development . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 481
13.2.1 Lumped Model Revisited . . . . . . . . . . . . . . . . . . . . . . 481
13.2.2 Model 1: Pure Diffusion Model and Fick’s Law . . . . . 483
13.2.3 Model 2: Advection/Diffusion Model . . . . . . . . . . . . . 487
13.2.4 Model 3: Diffusion from a Point Source . . . . . . . . . . . 492
13.2.5 Model 4: Matching the Air Column
with the Point Source . . . . . . . . . . . . . . . . . . . . . . . . . 495
13.2.6 Comparison of Heat Transfer Coefficients . . . . . . . . . . 497
13.3 Evaporation of a Forgotten Cup of Coffee . . . . . . . . . . . . . . . . 499
13.3.1 Experiment . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 500
13.3.2 Theory . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 501
13.3.3 System Mass Balance . . . . . . . . . . . . . . . . . . . . . . . . . 502
13.3.4 Results and Discussion . . . . . . . . . . . . . . . . . . . . . . . . 509
13.3.5 Surface Temperature . . . . . . . . . . . . . . . . . . . . . . . . . 511
Workshop 13.1. Evaporation Experiments . . . . . . . . . . . . . . . . . . . . . 513
Workshop 13.2. Surface Temperatures During Evaporation . . . . . . . . . 514
Workshop 13.3. A Cooling Mug of Coffee
with Evaporation Effects . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 514
Appendix. Evaluating Dew Points and Mass Fractions . . . . . . . . . . . . 514
 Part I
Modes of Heat Transfer
Thermal Circuits
1

This introductory chapter develops the concept of a thermal circuit and analytical
and numerical methods used to analyze simple circuits. Subsequent chapters will
apply and build on them using heat transfer examples from common experience.
Table 1.1 introduces key elements, namely, resistors, capacitors, batteries, and
current sources. Resistive/capacitive (RC) network diagrams are used as modeling
tools to visualize the flow of heat, an abstract massless quantity. Hence, the term
“thermal circuit.” The focus of this book is to develop heat transfer modeling as a
design tool, rather than as a pure science. The principles apply to other areas in
engineering (and life) practice. Specific tools are developed in this textbook that
pertain to heat transfer phenomena, most of which can be captured in thermal
circuits.

1.1 RC Circuit Driven by a Temperature Source

The discharge of a capacitor through a resistor is modeled by the circuit of Fig. 1.1.
This circuit is a model for the heating or cooling of an object placed suddenly into
an environment at a different temperature, like a hot mug of coffee placed on a
table, or a warm beverage placed in a refrigerator.
An electrical capacitor stores electrical charge (Q) on two plates. The capaci-
tance (C) relates stored charge to voltage between the plates (Q ¼ C(V  Vref),
where Vref is a reference voltage, typically ground). A thermal capacitor stores
thermal energy. For systems that do not experience phase change or chemical
(or nuclear) reactions, the thermal capacitance relates the stored energy to the
temperature (E ¼ C(T  Tref)), where the stored energy is set to zero at some
reference temperature, Tref. An electrical resistor provides a pathway for charge
to flow, driven by a voltage difference across it. The larger the resistance value, the
larger the voltage difference required to move a given quantity of charge. Thermal
resistors provide a pathway for heat to flow, heat being a specific form of energy
that crosses a system boundary, driven by a temperature difference.

# Springer International Publishing Switzerland 2015 3

G. Sidebotham, Heat Transfer Modeling: An Inductive Approach,
DOI 10.1007/978-3-319-14514-3_1
4 1 Thermal Circuits

Table 1.1 Circuit components and associated units in electrical and thermal applications
Component Symbol Electrical SI unit Thermal SI unit

Resistor Ohm C
Watt
Capacitor Farad Joule
C


Battery Volt C

Current source Ampere Watts

Fig. 1.1 RC circuit initiated

by a temperature source
(battery)

A major focus of this book is to develop estimation methods for thermal

capacitance and resistance values based on the three modes of heat transfer (con-
duction, convection, and radiation). Heat is driven from one temperature to a lower
temperature through a medium which impedes the flow of heat, modeled as a
thermal resistance (R) to the flow. The ability of a material to store internal energy
is determined by its thermal capacitance, C. The higher the capacitance of a
material, the lower the change in its temperature when a quantity of heat is added
or removed from it.
Temperature sources (batteries) are locations where the temperature is fixed,
regardless of the rate at which energy is extracted in the form of heat. The
components of Fig. 1.1 are two temperature sources (T0 and T1), a switch (S) a
resistor (R) and a capacitor (C). The normally open switch is momentarily closed
(dashed lines), long enough to charge the capacitor (C) to the battery temperature
(T0), and then released. In response, heat (analogous to electrical current) flows
from the capacitor through the resistor (R) to a second temperature source (T1),
thereby reducing the thermal energy stored in the capacitor (analogous to stored
electrical charge). Eventually, a final equilibrium state is reached at which point the
capacitor is fully discharged. Theory is developed next that describes the response
of the circuit given the fundamental laws of nature, key assumptions, and input
parameter values.
1.1 RC Circuit Driven by a Temperature Source 5

Fig. 1.2 Isolation of the

main node of Fig. 1.1, after
activation by the switch. The
arrows show the direction
assumed for current flow used
in Kirchhoff’s current law

In this text, a large filled “node” is generally used to indicate a temperature value
that is fixed by a battery of known temperature (which may change with time in a
prescribed way). The battery symbol will generally not be shown. For transient
problems, an open “node” is one to which thermal capacitance is assigned, while
small filled symbols (none in this case) indicate nodal temperatures that can be
determined based on neighboring capacitive-carrying and fixed nodal temperatures.
The fundamental law of nature governing this case is the conservation of charge
for the electrical case (the net flux of current into a node is zero: Kirchhoff’s current
law) and the First Law of Thermodynamics for the thermal case (developed
formally in the next chapter). Figure 1.2 shows the circuit after the starting point,
with current (q, with units of energy per unit time in the thermal case) assumed to
flow into the node from both the capacitor and the resistor.
Mathematically, this charge balance, or Kirchhoff’s current law, is:

qR þ qC ¼ 0

The current through the resistor is related to the voltage (T) by Ohm’s law, namely:

ðT 1  T Þ
qR 
R
The current is driven by the voltage difference across the resistor. The larger the
resistance, the lower the current. The (perhaps) familiar equation V ¼ IR is Ohm’s
Law, written in the form here of I ¼ V/R, which is really an unfortunate misrepre-
sentation of the relationship between voltage and current across a resistor. It should
be ΔV ¼ IR, where ΔV is the voltage difference across the resistor, not the absolute
value of the voltage at one specific point.
To establish the relationship between capacitor voltage and charge, consider an
event of time duration Δt, during which time a quantity of charge, Qexit, flows out.
Conservation of charge stated in words is:

Initial Charge ¼ Final Charge þ Charge that Exited

Qinitial ¼ Qfinal þ Qexit

6 1 Thermal Circuits

Rearranging and dividing by the time elapsed:

Qfinal  Qinitial ΔQ Qexit

¼ ¼
Δt Δt Δt
In the limit of short duration events (Δt ! 0), the left hand side becomes a first
derivative, and the right hand side represents the rate at which charge flows out, the
electrical current, qc:

dQ
 qc
dt
The total charge stored in the capacitor is related to the temperature by:


Q  C T  T ref

Taking the derivative and assuming the capacitance does not change with time:

dQ dT
 C
dt dt
The conservation of charge applied to a thermal capacitor is therefore:

dT
C ¼ qc
dt
Care has been taken in defining the signs. For the capacitor, a positive qc represents
current leaving the capacitor (and entering the node) which results in a decrease in
temperature, accounted for by the negative sign. For heat to flow into the resistor,
the temperature difference is expressed as that of the neighbor (T1) subtracted from
the capacity-carrying nodal value (T). If that nodal temperature is higher than its
neighbor, this term will have a negative value.
Inserting these component laws into the charge balance yields the governing
first-order ordinary differential equation (ODE) for the circuit:

dT ðT 1  T Þ
C þ ¼0
dt R
with an initial condition T ð0Þ ¼ T 0 .
This governing equation (and its initial condition) is the result of the model, yet
it is not the solution to the problem. Some calculus (and the assumptions that R, C,
and T1 are constant) is needed to solve it.
It is not necessary, but can be helpful, to define a dimensionless variable for
temperature that starts at a value of 1 and approaches a value of 0 at equilibrium
(when T ¼ Teq).
1.1 RC Circuit Driven by a Temperature Source 7

T  T eq
θ
T 0  T eq

Formally, equilibrium, or steady-state, is achieved when nothing changes with time,

accomplished by setting the time derivative to zero:


dT  T 1  T eq
¼ ¼0
dt eq RC

Since R and C are finite, the equilibrium temperature is therefore equal to the source
T1. For more complicated circuits, it may be more difficult to solve for an
equilibrium value.
A dimensionless time can be defined by normalizing the time by a characteristic
time constant (τ*), whose value will be determined shortly:

t
^t 
t*
Inserting these variables into the governing ODE and rearranging:

 

 
C d T eq þ T 0  T eq θ T 1  T eq þ T 0  T eq θ
 þ ¼0
t* d^t R
Since Teq is constant, its time derivative is zero, and T eq ¼ T 0 so the governing
differential equation becomes:
 
dθ t*
þ θ¼0
^
dt RC

It is now natural to define the time constant as

t*  RC

The values of R and C have been assumed to be constant (technically, their product
is constant). There are situations where R and/or C may vary with time, and
therefore this model would not apply.
The governing equation becomes devoid of all parameters:

dθ
þθ ¼0
d^t
with initial condition θð0Þ ¼ 1.
In a first-order system (i.e., governed by a linear first-order differential equa-
tion), the rate of change of some quantity is proportional to the negative of itself.
The solution to this linear first-order differential equation, plotted in Fig. 1.3, is

^
θ ¼ e t
8 1 Thermal Circuits

Fig. 1.3 Solution of the RC circuit in dimensionless variables

Or, in terms of dimensional variables:

T  T1 t
¼ eRC
T0  T1
Alternatively, this equation can be expressed in terms of the time by taking the
natural logarithm of both sides and rearranging:
 
T  T1
t ¼ RC ln
T0  T1

Example
Q: How much time is needed for the temperature to drop from an initial value of
100  C to a final value of 70  C, given that the ambient (T1) is at 25  C and the
time constant RC has a value of 1.613 h?
h i  7025 
A: time ¼ RC ln TTT 1
0 T 1
¼ 1:613 ln 10025 ¼ 0:819 h

In this case, the dimensionless temperature has a value of 0.6, and, referring to
Fig. 1.3, the corresponding dimensionless time is approximately ½.
To generalize, in one time constant, the dimensionless temperature drops to
within 1/e ¼ 0.368 of its initial value, or it is about 2/3 of the way to equilibrium.
After two time constants, it is close to 10 % of its equilibrium value. A value of
5 time constants is a generally accepted value to indicate that equilibrium has been
reached, for most engineering purposes.
1.2 RC Circuit Driven by a Current Source 9

1.2 RC Circuit Driven by a Current Source

Figure 1.4 shows an RC circuit driven by a current (heat) source. This circuit is a
model for an object, initially at equilibrium with its environment, suddenly exposed
to a radiant heat source, like parking a car in the sun.
In contrast to a voltage source (battery) which fixes the voltage regardless of the
current flow in (or out), a current source fixes the current (heat transfer rate)
regardless of the voltage (temperature). Initially, with the switch open, the voltage
(T) is at the fixed value T1 (fixed by a battery, not shown, but implied by the filled
symbol). At some initial time, the switch is held closed, and current from the source,
Q_ ¼ qs as indicated, flows steadily into the node. Current sources in thermal
systems can include chemical reactions, certain forms of radiation (i.e., solar
radiation) or dissipation of electrical energy.
Kirchhoff’s current law for this circuit is:

q S þ qR þ qC ¼ 0

Introducing the component expressions:

T1  T dT
Q_ þ C ¼0
R dt
with an initial condition T(0) ¼ T1.
As before, a dimensionless temperature can be defined that starts at an initial
value of unity, and is driven toward an equilibrium value:

T  T eq
θ
T 0  T eq

The equilibrium value is obtained by setting the time derivative equal to zero:


dT  T 1  T eq
C  ¼ þ Q_ ¼ 0
dt eq R

Fig. 1.4 RC circuit driven be

a current source
10 1 Thermal Circuits

Solving for the equilibrium temperature:

T eq ¼ T 1 þ RQ_

R and Q_ are inherently positive quantities, and therefore the equilibrium tempera-
ture is above ambient. A dimensionless time can be defined by normalizing the time
by a characteristic time (t*), whose value will be determined shortly:

t
^t 
t*
Inserting these variables into the governing equation and rearranging:




C T 0  T eq dθ T  T T  T θ
¼ Q_ þ
1 eq 0 eq
 ¼0
t* d^t R R

The term in brackets is identically zero from the definition of the equilibrium
temperature, and therefore, upon further rearrangement:
 
dθ t*
þ θ¼0
d^t RC

The same time constant as before naturally emerges, namely:

t*  RC

The governing equation becomes:

dθ
þθ ¼0
d^t
with initial condition θð0Þ ¼ 1.
This governing equation is exactly the same as for the voltage source. The only
difference is in the value of the equilibrium temperature. The underlying physics of
a voltage source and current source are very different, however.

1.3 RC Circuit Driven by Multiple Sources

Figure 1.5 shows an RC circuit in which a node (at T) is driven by a current source
and two separate voltage sources (at T1 and T2, with resistance R1 and R2).
Kirchhoff’s current law for this circuit is:

qS þ qR1 þ qR2 þ qC ¼ 0
1.3 RC Circuit Driven by Multiple Sources 11

Fig. 1.5 RC circuit driven

by multiple sources

Introducing the component expressions:

T1  T T2  T dT
Q_ þ þ C ¼0
R1 R2 dt
with an initial condition T(0) ¼ T0, where T0 can be taken to be any initial
temperature. If, as suggested by Fig. 1.5, the switch is initially open for a suffi-
ciently long time prior to closing, the initial temperature will be between T1 and T2.
Precisely where depends on the relative values of R1 and R2, as will be clear in the
next section.
A general form of Kirchhoff’s current law for a node “i” in thermal contact with
multiple neighbors “j” and a current source is:

dT i X Tj  Ti
Ci ¼ Q_ i þ
dt j
Rij

where Rij is the thermal resistance between node “i” and “j”. As before, a dimen-
sionless temperature can be defined that starts at an initial value of unity, and is
driven toward an equilibrium value:

T  T eq
θ
T 0  T eq

The equilibrium value is obtained by setting the time derivative equal to zero:



dT  _ T 1  T eq T 2  T eq
C  ¼Qþ þ ¼0
dt eq R1 R2
12 1 Thermal Circuits

Solving for the equilibrium temperature:

T1 T2
Q_ þ þ
R1 R2
T eq ¼
1 1
þ
R1 R 2
A general form of the equilibrium temperature for a node “i” in thermal contact
with multiple neighbors “j” and a current source is:

X Tj
Q_ i þ
j
Rij
T i, eq ¼ X 1

j
Rij

A dimensionless time can be defined by normalizing the time by a characteristic

time (τ), whose value will be determined shortly:

t
^t 
t*
Inserting these variables into the governing equation and rearranging:


 
C T 0  T eq dθ _ T 1  T eq T 2  T eq 1 1

¼ Qþ þ  þ T 0  T eq θ ¼ 0
t* d^t R1 R2 R 1 R2

The term in brackets is identically zero from the definition of the equilibrium
temperature, and therefore, upon further rearrangement:
 
dθ t* 1 1
þ þ θ¼0
d^t C R 1 R2

It now makes sense to define the time constant as:

C
t*   
1 1
þ
R1 R2

Or defining an equivalent resistance as:

1 1 1
¼ þ
R R1 R2
1.4 Resistors in Series 13

Then

t*  RC

Notice that the time constant does not involve the current source. The governing
equation becomes:

dθ
þθ ¼0
d^t

with initial condition θð0Þ ¼ 1.

This governing equation is exactly the same as for the temperature source. The
only difference is in the value of the equilibrium temperature, and an equivalent
resistance rather than a single resistance.

1.4 Resistors in Series

In heat transfer, adding layers of insulation adds thermal resistance in series. For
example, putting on a sweater introduces an additional barrier to heat flow from the
torso and arms to the environment. Figure 1.6 shows a simple circuit, called a
voltage divider, in which two resistors are wired in series, separated by two
temperature sources (batteries).
An equivalent circuit with a single resistor (R) between the two batteries is also
shown. The relationship between R and the individual resistors R1 and R2 is sought.
There is only one channel for current to flow from T1 to T2 (if T1 is greater than
T2, otherwise the direction is reversed). Conservation of charge dictates that the
current flowing through both resistors must be the same.
For the equivalent circuit, the current through the equivalent resistor (R) must
equal the current entering R1, consistent with the directions of the arrows. That is,
for resistors added in series:
q12 ¼ qR1 ¼ qR2

where q12 represents the current from T1 to T2. Inserting the component laws for R1
and R2 separately, making the signs consistent with Fig. 1.6:

T1  T
qR1 ¼ q12 ¼
R1

Fig. 1.6 A channel with two resistors in series, and the equivalent circuit
14 1 Thermal Circuits

T  T2
qR2 ¼ q12 ¼
R2
Expressing these equations in terms of the intermediate temperature (T) between
the two resistors:

T ¼ T 1  R1 q12
T ¼ T 2 þ R2 q12

Subtracting these two equations:

0 ¼ T 1  T 2  ðR1 þ R2 Þq12

Rearranging for the heat transfer rate:

T1  T2
q12 ¼
R1 þ R2
For the equivalent circuit:

T1  T2
q12 ¼
R
The equivalent resistance is therefore:

R ¼ R1 þ R 2

The equivalent resistance of resistors in series is the sum of the resistances. The
currents are equal through the resistors. The intermediate temperature can be
expressed in a number of ways. Rearranging for the intermediate temperature
three different (but equivalent) ways:

T1 T2
þ
R R2
T¼ 1
1 1
þ
R1 R2
 
R1
T ¼ T1  ðT 1  T 2 Þ
R1 þ R2
 
R2
T ¼ T2 þ ðT 1  T 2 Þ
R1 þ R2

Adding resistors in series impedes current flow. The voltage in between the two
resistors is intermediate. If the resistors are equal in value, the intermediate temper-
ature will be halfway between T1 and T2. If R1 is greater than R2, a larger voltage
drop will be required across R1 than across R2, and the intermediate temperature
will be closer to T2 than to T1.
1.5 Resistors in Parallel 15

Example Let T1 ¼ 100  C and T2 ¼ 25  C and let R1 ¼ 10  C/W and R2 ¼ 30  C/W.

Calculate (a) the equivalent resistance, (b) the current, and (c) the intermediate
temperature for the series resistance circuit of Fig. 1.6.

(a) The equivalent resistance is R ¼ R1 + R2 ¼ (10 + 30)  C/W ¼ 40  C/W. Notice

that the equivalent resistance is larger than either of the two resistors.
(b) The current is

T 1  T 2 100  25
q12 ¼ ¼ ¼ 1:875 W
R1 þ R 2 10 þ 30
(c) The intermediate temperature is
   
R1 10
T ¼ T1  ðT 1  T 2 Þ ¼ 100  ð100  25Þ ¼ 81:25 C
R1 þ R2 10 þ 30

1.5 Resistors in Parallel

In winter, heat loss through an exterior wall takes places across its windows and the
remainder of the wall simultaneously. The rate of heat loss through the windows is
different than that through the solid wall, even though the air temperatures are the
same inside and outside. The total heat loss across the wall is the sum of these two
“thermal channels,” and these channels act in parallel.
Figure 1.7 shows two resistors wired in parallel between two batteries, and an
equivalent circuit with a single resistor, whose relationship to R1 and R2 is sought.
Unlike the series circuit, there are two distinct channels through which current can
flow. In a heat transfer application, heat lost through a wall of a building takes place
through two parallel channels, through the windows and through the remainder of
the walls.
A nodal balance at either side of the circuit states that the total current entering
(or leaving) the circuit is the sum of the currents flowing through each resistor, and
the voltage drop is the same across both resistors:

q12 ¼ qR1 þ qR2

Inserting component laws:

T1  T2 T1  T2 T1  T2
¼ þ
R R1 R2
16 1 Thermal Circuits

Fig. 1.7 RC channel with two resistors in parallel, and the equivalent circuit

Eliminating the common voltage drop:

1 1 1
¼ þ
R R1 R2

The equivalent resistance can be expressed as:

1
R¼ 1
R1 þ R12
or
R1 R2
R¼
R1 þ R2

The equivalent resistance is less than the smaller of the two resistances. In heat
transfer, heat flows through a wall in a house through parallel channels, i.e., across
solid walls and across windows. The inside and outside temperatures are the same
for windows and solid walls, and the total heat transfer rate is the sum of the heat
through these parallel channels.

Example Let T1 ¼ 100  C and T2 ¼ 25  C and let R1 ¼ 10  C/W and R2 ¼ 30  C/W.

Calculate (a) the equivalent resistance and (b) the current for the parallel resistance
circuit of Fig. 1.7. Compare the results to the series circuit that had the same
batteries and resistors.

(a) The equivalent resistance is

R1 R2 ð10Þð30Þ
R¼ ¼ ¼ 7:5  C=W
R 1 þ R2 10 þ 30
1.6 Iterative Solution Methods 17

The resistance is less than the smaller of the two, while the resistance of the
series circuit was 40 Ω, greater than the larger of the two.
(b) The current is

T 1  T 2 ð100  25Þ C
q12 ¼ ¼ ¼ 10 W
R 7:5 C=W

This current is much larger than the 1.875 W for the series circuit.

1.6 Iterative Solution Methods

“I’m thinking of a number from 1 to 100”

50
Higher
75
Lower
65
Lower
62
Correct. You read my mind!

This dialogue is an example of an iterative method. There is a correct solution, but

not an obvious way to get to it directly. So a feedback loop is used in which a guess
is made, followed by feedback, a subsequent guess using an algorithm, and eventu-
ally, a fairly rapid convergence to the correct solution is found. In this case, the
algorithm is “guess a value that is midway between the latest high and low values”
(This is the so-called regula-falsi method in which the correct result is initially
bracketed, and then methodically arrived at by closing in on it).
There are other algorithms that may or may not work. For example. . .
“I’m thinking of a number from 1 to 100”
1. Higher. 2. Higher. 3. Higher. 4. Higher. . . (etc etc. . .) . . .61. Higher. 62.
Correct! You read my mind!
This algorithm, start at 1 and count up, found the same answer, not as quickly, but
so what, if the only objective is the right answer. A brute force algorithm like this is
perfectly legitimate in many instances.
Here is an algorithm that is not likely to work.
“I’m thinking of a number from 1 to 100”
40. Higher. 40 Higher. 40. Higher. Etc. . .
The algorithm here is “always guess the same number.” It would only work if the
first guess happened to equal the correct answer.
Here is an algorithm in which the feedback is useless.
18 1 Thermal Circuits

“I’m thinking of a number from 1 to 100”

50
Wrong!
25
Wrong!
. . ..
The guesser will eventually find 62, but not because of the feedback.

Iterative procedures are quite common in heat transfer applications. There are
formal methods that will be introduced for specific types of problems, and iterative
procedures can by “invented” as needed to solve problems for which a direct
solution is not possible. You’ll see.
There is a fundamental difference between an iterative and predictive method. In
an iterative method, guesses are made, based on feedback, and hopefully the
algorithm works in the sense that the desired final solution is obtained. The
“trajectory” of the guessed solutions is really not important. It is the final result
that matters. In a predictive method, each step in time is a prediction of what the
actual temperature is. The prediction of the temperature after a finite time step is not
a “guess,” it is a simulation based on the model. Some important predictive
numerical methods are introduced next.

1.7 Numerical Solution 1: Forward Explicit Formulation

(Euler Method)

A few techniques for solving transient problems numerically are introduced in this
introductory chapter, using the basic first-order system to demonstrate the pros and
cons of each method, namely,

dT ðT 1  T Þ
¼
dt RC
Initial Condition : T ð 0Þ ¼ T 0

Note that the values of R and C could be variables.

The derivative in the governing equation can be expressed as a forward finite
difference approximation by taking a finite step in time (Δt), using a superscript to
represent time as p (present) or p + 1 (future). There is no past in this method.

T ðpþ1Þ  T ðpÞ T ð1pÞ  T ðpÞ

¼ ðpÞ ðpÞ
ΔtðpÞ R C
The right hand side represents the rate of heat transfer (divided by C) at the present
time. Rearranging and solving for the future temperature in terms of the present:
ΔtðpÞ  
T ðpþ1Þ ¼ T ðpÞ þ T ð1pÞ  T ðpÞ
RðpÞ CðpÞ
1.7 Numerical Solution 1: Forward Explicit Formulation (Euler Method) 19

with initial condition T ð0Þ ¼ T 0 . In effect, this method extrapolates the slope
forward into time by a finite step (the numerical parameter Δt), without any
adjustments during that step. Notice that in the limit of Δt going to zero, the
Euler method becomes the definition of the first derivative.
The unit of the RC combination is time, and is a characteristic time constant of
the problem. Engineering decisions can often be made knowing no more than that.
This Euler method is called an “explicit” method because the future predicted
value is expressed explicitly in terms of present values, which are known, starting
with the initial condition. Such will also be true for more complicated models (with
more than one unknown nodal temperature). It is easy, and somewhat intuitive, to
implement, and is very powerful. It can run into problems as systems become more
complex, but a legitimate strategy is to implement the method, and if it fails to
produce acceptable results, it can dictate an appropriate “Plan B.”
“If you know the present, you can tell the future.” Of course, you only ever know
so much of the present.

1.7.1 Variable Step Size Explicit Methods

The step size (Δt) is a numerical parameter. It is relatively easy to incorporate logic
that calculates a new value for the step size at every step. Three variations of this
method are:

1.7.1.1 Fixed Step

A fixed step size method is one where the step size is chosen at the start of the
simulation and used throughout:

ΔtðpÞ ¼ Δt

This method seems to be the most obvious and is recommended as the methods are
being learned, yet it can have its drawbacks in practice. The change in the indepen-
dent variable is held constant in this method.

1.7.1.2 Fixed Change

A less obvious, but occasionally useful, method is to allow the dependent variable
to change by a fixed amount:

T ðpþ1Þ  T ðpÞ ¼ ΔT

This method works well if it is known that the temperature change will monotoni-
cally change from beginning to end, which is often the case in heat transfer. The
step size at each step is given by:
20 1 Thermal Circuits

RðpÞ CðpÞ ΔT
ΔtðpÞ ¼
T ð1pÞ  T ðpÞ

1.7.1.3 Fixed Fractional Change

In this method, the step size is calculated that will result in a change in the
dependent variable by a fixed fraction, f, of its deviation from its equilibrium
value Teq, in this case T1:
 
T ðpþ1Þ  T ðpÞ ¼ f T ðpÞ  T ðeqpÞ

The step size for this method is then:

 
T ðpÞ  T ðpÞ 
 
ΔtðpÞ ¼ f RðpÞ CðpÞ  ðpÞ
eq

T 1  T ðpÞ 

Absolute values are needed to guarantee a positive step size. For the RC circuit
studied for now, R and C are constants, and the equilibrium value for T is T1,
making the temperature ratio equal to unity. Therefore, this method will result in a
fixed step size for this case, but not in general.
Also, this method works best, conceptually, for dependent variables that are
inherently positive, like absolute temperature, as opposed to temperature expressed
in Celsius. For example, a 1  C change in a temperature of 100  C is a 1 % change,
whereas the same 1  C change in a temperature of 10  C is a 10 % change. What if
the temperature were 0  C? or 10  C?

1.8 Numerical Solution 2: Backward Implicit formulation

Express the derivative in the governing equation as a backward finite difference

approximation by taking a finite step in time (Δt), and introduce a superscript to
represent time as p (present) or p  1 (past). There is no future, yet. . .

T ðpÞ  T ðp1Þ T 1  T ðpÞ

¼
Δt RC
Rearranging and solving for the present temperature in terms of the past:
 
RC ðp1Þ
T1 þ T
Δt
T ðpÞ ¼  
RC
1þ
Δt

with initial condition T ð0Þ ¼ T 0 .

1.9 Numerical Solution 3: Central Implicit formulation (Crank–Nicolson Method) 21

“The present is a result of the past.” But we want to predict the future! Let the
past become the present, and the present the future by updating “p” to “p + 1”;
 
RC ðpÞ
T1 þ T
Δt
T ðpþ1Þ ¼  
RC
1þ
Δt

It is not clear why this method is called an implicit method in this “1-node” example
because the future predicted value (p + 1) is expressed explicitly in terms of present
values (p). That will not be true for more complicated models (with more than one
unknown nodal temperature).

1.9 Numerical Solution 3: Central Implicit formulation

(Crank–Nicolson Method)

This method is introduced here, but it is not recommended that it be used in heat
transfer applications, because it violates the Second Law of Thermodynamics for
large step sizes, whereas the backward method yields correct equilibrium values.
There are other applications in numerical analysis for which the improved accuracy
of this method are advantageous.
Express the derivative in the governing equation as a central finite difference
approximation by taking a half finite step in time (Δt/2), and introduce a superscript
to represent time as p  1/2 (past) or p + 1/2 (future). There is no present, yet. . .

T ðpþ 2 Þ  T ðp 2 Þ T 1  T ðpÞ

1= 1=

¼
Δt RC
There are three times depicted, past, present and future. To reduce that to two,
approximate the present as the average of the past and future:

T ðpþ 2 Þ þ T ðp 2 Þ
1= 1=

Tð pþ =2
1
Þ  Tð p =2
1
Þ T1 
¼ 2
Δt RC
Let the past become the present, and the future further into the future by updating “p” to
“p + 1/2”. Rearranging and solving for the future temperature in terms of the present:

T ðpþ1Þ þ T ðpÞ
T ðpþ1Þ
T ðpÞ T1 
¼ 2
Δt RC
22 1 Thermal Circuits

 
RC 1 ðpÞ
T1 þ  T
Δt 2
T ðpþ1Þ ¼  
1 RC
þ
2 Δt

with initial condition T ð0Þ ¼ T 0 .

It is not clear why this method is called an implicit method in this “1-node”
example because the future predicted value is expressed explicitly in terms of future
values. That will not be true for more complicated models (with more than one
unknown nodal temperature).

Workshop 1.1: Numerical Solution of First-Order RC Circuit

Consider the model problem of an RC circuit shown, after activation of an initial

state by the switch. Write an Excel workbook or program code that compares the
exact solution (as lines, no markers) and the numerical predictions (as markers with
straight lines between them) from the forward, backward, and central difference
methods for different values of time step, Δt. Allow user input, but use default
values from the following table:
Add a worksheet or code to implement the Euler method with a variable step size
(fixed change).

Input parameter Symbol Value Unit


Initial temperature To 100 C

Ambient temperature T1 0 C

Resistance value R 10 C/W
Capacitance C 0.1 J/ C
Numerical time step Δt 0.5 s

Note: A detailed solution follows. It is advised that this solution not be consulted
prior to doing this workshop yourself. The intent is for as much “self-discovery” to
occur as possible.
Numerical Stability 23

Numerical Accuracy

The three numerical methods are compared with the exact solution in Figs. 1.8, 1.9,
and 1.10 for fixed numerical time steps of 1.0 s, 0.5 s, and 0.1 s, respectively. The
numerical methods are applied to the RC model problem, with a time constant
(RC) of 1 s. The exact solution is shown as a solid line. For the case where the step
size of the numerical approximation equals the physical time constant of the system
(Δt ¼ 1 s here), the explicit method achieves equilibrium in a single step, and then
stays there. The backward implicit method decays too slowly, and the central
method is the most accurate. The accuracy of all methods increases as the step
size decreases. For practical purposes, a step size of 10 % of the underlying time
constant can be considered sufficiently small to simulate the model accurately for
all methods.

Numerical Stability

Figure 1.11 shows the effect of time step on numerical stability for the Euler
method (explicit forward difference method) for step sizes between one and two
time constants. The method involves calculating the initial slope, and extrapolating
forward at that slope for a full step. As shown previously, a step size of 1.0 will
bring the temperature to ambient (0  C here) in a single step. For a larger step, the

Fig. 1.8 Comparison of three numerical techniques on the transient model problem. The time
constant (RC) has a value of 1 s for all three numerical techniques. The exact solution is shown as a
solid line, and straight dashed lines between points are drawn for the explicit and backward
implicit method
24 1 Thermal Circuits

Fig. 1.9 Same as previous with a step size of 0.5 s

Fig. 1.10 Same as previous with a step size of 0.1 s

solution after one step will overshoot its equilibrium value. The next step will
evaluate a positive slope, and will overshoot its equilibrium. The temperature will
resemble a decaying periodic function, and eventually settle to the correct equilib-
rium value. This behavior is numerically stable (the function remains finite), but in
Numerical Stability 25

Fig. 1.11 Explicit forward difference method for step sizes between one and two time constants

a heat transfer context, each step in time violates the Second Law of Thermody-
namics: a warm object cooling to a cold environment cannot become colder than
that environment without some other agent (such as evaporation). Therefore, a step
size of 1 time constant is considered to be the maximum step size allowable for the
explicit Euler method.
Figure 1.12 shows the behavior of the Euler method for a step size of 2 time
constants and a slightly larger value (2.01). For a step size of 2, the solution bounces
back and forth between +100 and 100  C, and will do so indefinitely. For a step
size of 2.01, the solution grows in absolute value with each step, and will eventually
grow to infinity (resulting in a computer error once the size limit of the program
being used is reached). A step size of 2 time constants is therefore shown to be the
maximum step size for numerical stability for the Euler method. This instability can
be seen in the mathematics of the recursion relation for the Euler method, namely,

Δt  
T ðpþ1Þ ¼ T ðpÞ þ T 1  T ðpÞ
RC

in which the absolute value of the future temperature increases with step size (Δt).
The limit for the Euler method is:

lim T ðpþ1Þ ¼ 1
Δt¼1

In general, a numerical method is said to be unstable if an inherently stable physical

problem grows without bounds as successive steps are taken. It is possible, in
general, that a problem is inherently unstable, but that is not the case here.
26 1 Thermal Circuits

Fig. 1.12 Explicit forward difference method for step sizes of 2.00 and 2.01

Figure 1.13 shows the behavior of the backward and central implicit methods for
a step size of 10 (which would be wildly unstable for the forward method). These
methods are shown to be inherently stable, numerically, meaning that their solution
will not increase indefinitely, but rather will approach equilibrium no matter how
large a step size is taken. For the backward method, the recursion relation is
 
RC ðpÞ
T1 þ T
Δt
T ðpþ1Þ ¼  
RC
1þ
Δt

Its limit is

lim T ðpþ1Þ ¼ T 1
Δt¼1

This limit shows that for large step sizes, the solution approaches the equilibrium
value in a single step. On the other hand, for the central method, the recursion
relation is
Numerical Stability 27

Fig. 1.13 Implicit methods (backward and central difference) with a step size of 10 time
constants

 
RC 1 ðpÞ
T1 þ  T
Δt 2
T ðpþ1Þ ¼  
1 RC
þ
2 Δt

Its limit is

lim T ðpþ1Þ ¼ 2T 1  T ðpÞ

Δt¼1

There is no real physical meaning that can be attributed to this limit. For very large
step sizes, the central method will eventually settle to equilibrium, but will do so as
a decaying periodic function. Its behavior violates the Second Law, and in no way
resembles the true behavior of a first-order system.
In future chapters, when multi-node systems are considered, it is possible that
the inherent time constant of one or more nodes is much smaller than all the others,
the so-called stiff numerical problem. In these cases, an implicit method, which is
harder to implement than the explicit method, might be advantageous. While the
central difference method might appear to have better accuracy for step sizes less
than the inherent time constant of the system, the backward difference method is
arguably preferred when the numerical step size is much larger than the inherent
time constant of the most rapidly changing nodal temperature.
28 1 Thermal Circuits

Workshop 1.2: Numerical Solution of RC Circuit with Time

Varying Ambient Temperature

Consider the RC circuit shown (with R ¼ 1  C/W and C ¼ 1 J/ C as defaults), after
activation of an initial state by the switch (T0 ¼ 0  C). In this circuit the ambient
temperature varies with time in a prescribed way. Specifically,

T 1 ¼ T average þ A sin ð2πNtÞ

where Taverage(¼50  C) is the average temperature, A(¼50  C) is the amplitude and

N(¼1 Hz) is the frequency.

Write an Excel workbook or program code that implements the Euler method and
plots T and T1 vs. time (on the same axes) over enough cycles to reach a “steady-state”
(in which there is no cycle to cycle temperature variation). Complete the Results table
in order to investigate the system amplitude and phase response to source frequency.
An analytical solution to this problem can be developed for those so inclined.

Model Development
An energy balance applied to the temperature node is:

dT T 1  T
C ¼
dt R
The Euler method uses a forward finite difference approximation to the first
derivative:

dT T ðpþ1Þ  T ðpÞ

dt Δt
where Δt is the numerical step size, the superscript (p) is an index referring to the
present time, and (p + 1) is one step into the future. Inserting and rearranging for the
future temperature in terms of the present:
 
ðpþ1Þ ðpÞ Δt T ð1pÞ  T ðpÞ
T ¼T þ
C R

For numerical accuracy, the step size should be much smaller (say 10 %) than the
smaller of the inherent response time (RC), or the cycle time (1/N).
Workshop 1.2: Numerical Solution of RC Circuit with Time. . . 29

RESULTS (run cases with different step sizes to observe the effect, but don’t
report them)

N Δt Amplitude Phase lag (deg)

0.01 0.1
0.1 0.1
1 0.1
10 0.01
100 0.001
Lumped Capacity Systems and Overall Heat
Transfer Coefficients 2

This chapter introduces two practical heat transfer concepts; lumped capacity and
overall heat transfer coefficients. The term “lumped capacity system” means that a
finite region of space is considered to be at some average temperature, even though
there could be hot or cold spots in it. The example of a cooling mug of coffee will be
used extensively in subsequent chapters, in which case an analysis of the average
temperature will be conducted, even though the coffee is somewhat hotter than the
mug, and there are cold spots of coffee near the mug wall. In numerical analysis,
subdivisions of a physical object can be made, treating each subdivision as a
lumped capacity element.
The “overall heat transfer coefficient, U” is a parameter that captures the basic
idea of heat transfer, namely, energy transfer that is driven by a temperature
difference. For example, “R value” is the metric used to rate the thermal perfor-
mance of insulating materials, and it is the inverse of an overall heat transfer
coefficient, U. Applied to a wall on a cold day, heat must flow from the inside air
to the inner surface of an exterior wall, then through the wall, which has paint, sheet
rock, studs, insulation, sheathing, vapor barrier, shingles, paint, and then from the
outer surface of the wall to the cold air outside. This textbook builds theoretical
tools for estimating all these effects. The R value is a measure of the overall heat
transfer coefficient (actually, its inverse) and includes all the complicated details
into a single metric.
The focus of this chapter is on the thermal capacity of systems, which depends
on the material properties of density and specific heat. The next chapter will focus
on the thermal resistances of systems, where thermal conductivity and emissivity
are key material properties. The remaining chapters will build analysis tools for a
widening range of applications.
In dealing with complicated situations that involve heat transfer, the approach
will be to start with as global a picture as is reasonable, generally called a 1-node
model, in which all the material in question is lumped into a single “node,” and an
overall heat transfer coefficient between this system and its environment estimated
(which can be rather involved). This first step is the ultimate in the process of

# Springer International Publishing Switzerland 2015 31

G. Sidebotham, Heat Transfer Modeling: An Inductive Approach,
DOI 10.1007/978-3-319-14514-3_2
32 2 Lumped Capacity Systems and Overall Heat Transfer Coefficients

numerical analysis in which space is “discretized.” Time is also discretized using

one of the methods from Chap. 1 (or others that have been developed). If more
detail is needed, the system in question can be broken into two nodes, and then an
interaction between the two nodes occurs, and both of them may interact with the
environment. Then it is logical to define three nodes, or then to “n” nodes, or into
“n
m” nodes for a two-dimensional space, or into “n
m
p” nodes for 3D. Each
node is treated as a lumped capacity system.
That’s a road map of the next several chapters. Welcome aboard!

2.1 A Cooling Mug of Coffee

I, like many, NEED a cup of coffee as soon as possible when I wake up. It takes a
few minutes to prepare the coffee maker and run it through its cycle. Then I pour hot
coffee into a mug. It’s TOO HOT! Grrrrrr. . . So maybe I drop a few ice cubes in it so
I can get a good gulp to start my day. Diluting the coffee is not ideal, but it’s worth it
for that first cup. I wonder if I can predict how long it would take to just wait for
it. A goal of this book is to satisfy that curiosity.

2.1.1 Problem Statement

A volume V ¼ 0.41 L of coffee at 1 atm, 92  C is quickly poured into a room

temperature (21  C) ceramic mug (shown schematically in Fig. 2.1) with a mass of
mmug ¼ 0.365 kg, inner diameter D ¼ 0.083 m, inner height H ¼ 0.098 m, outer
height Ho ¼ 0.109 m, and nominal wall thickness w ¼ 0.0040 m. The mug of coffee
is placed on a table in a still room. Estimate the time it takes for the water to cool to
60  C (above which it is too hot) and to 45  C (defined as “tepid”).

2.1.2 Experimental Solution

An experiment was conducted using a ceramic mug which closely approximates the
mug defined in the problem statement. The actual mug tested has rounded edges, a
handle, and other deviations from the schematic. The idealizations implied in the
schematic diagram represent an early modeling step. Water was brought to a rolling
boil on a stovetop and quickly poured into the mug. A handheld infrared thermom-
eter was used to measure the free surface and the outside wall temperatures as
functions of time. The raw data are plotted in Fig. 2.2 for the entire cooling event,
the end of which, after some 4 h, is defined by the coffee reaching ambient. After an
initial adjustment period, the top and side temperatures fall more or less together
and asymptotically approach ambient. The event will be broken down into distinct
stages in the following paragraphs. The raw data is reported in Appendix 1.
It is emphasized that this experiment was conducted with a handheld device that
measures surface temperature only and was not in a controlled laboratory
2.1 A Cooling Mug of Coffee 33

Fig. 2.1 Schematic of coffee

mug with geometry defined

environment, nor with a detailed error analysis. There is much to be said, but not
here, about the experimental uncertainty in this experiment that would raise into
question many fine details of these temperature curves, especially during the initial
period as the mug is heated. Readers of this text are encouraged to conduct their
own experiments, and apply the theoretical methods developed in this text to those
situations.
Figure 2.3 shows the first 10 min of the cooling event. At the start (immediately
after pouring the coffee), the mug is at ambient temperature and the top surface is
near the boiling point of water at atmospheric pressure. Note that the IR tempera-
ture sensor measures the surface temperature of the water, and strong evaporative
effects of water near its boiling point likely produce temperatures noticeably below
that of the interior. There is an initial phase, called the “mug heating phase” which
34 2 Lumped Capacity Systems and Overall Heat Transfer Coefficients

Fig. 2.2 Temperature of the free surface (top) and middle of the outer side wall (side) as a
function of time for the entire cooling event

Fig. 2.3 First 10 min of the cooling event

2.1 A Cooling Mug of Coffee 35

Fig. 2.4 Temperatures for the first hour of the cooling event

will be investigated in some detail in future chapters. This phase lasts about 3 min,
during which time the side temperature rises to a peak, and the top (surface)
temperature drops quickly to a slightly lower temperature than the side. At the
end of the mug heating phase, the two temperatures are close to each other at
approximately 72  C.
Figure 2.4 shows the first hour of the event, during which time the coffee passes
between practical drinking stages. For the first 16 min, the coffee is too hot to drink.
The subjective definition of “too hot” at 60  C was determined by testing with small
sips (in a separate experiment). The temperature falls after the mug heating phase as
intuition would suggest. During the subsequent cooling phase, the indicated tem-
perature of the free surface is slightly lower than that of the sides of the mug most of
the time. Between 16 and 41 min, the coffee is drinkable, but not tepid. The
sampling rate was changed during this time, and by the end, the temperature
indicated by the IR detector of the top surface is slightly higher than that of the
side wall.
So we’re done, right? After pouring the coffee, I need to wait about 15 min
before I can start sipping, and then 25 min later (40 min after pouring), it’ll be too
late. But what if I put a lid on it, or was using a travel mug, or had a paper cup, or a
bigger mug, or was really interested in a much larger (or smaller) scale type of
practical problem? The main point of this example is to use a simple experiment to
compare to theoretical predictions. The skill of making such a prediction from “first
principles” is the main objective of this text, to which we now turn.
36 2 Lumped Capacity Systems and Overall Heat Transfer Coefficients

2.2 Lumped Capacity Model Development

The first step of a thermodynamic analysis (of which heat transfer is a subset) is
generally to split the universe into two parts, system and surroundings. Doing so
produces a control volume (the region of space that encompasses the system) and a
control surface (the boundary between the system and its surroundings). A lumped
capacity heat transfer model captures the global transient response of the average
temperature of a system to its surroundings at a different temperature. Many
practical problems fall into this category. Furthermore, the basic concepts devel-
oped here form a basis for more complicated analyses when more detail is required.
It is easy, but wrong, to conclude that the lumped capacity approximation means
that the temperature of an object is uniform. Proper application of heat transfer
methods (next chapter) allow for significant spatial variation in temperature, while
still maintaining a lumped capacity model.

2.2.1 Conservation of Energy

The First Law of Thermodynamics (conservation of energy) is the applicable

fundamental law for this event. Choosing an imaginary boundary surrounding the
mug of coffee, shown as dashed lines in the problem sketch, and assuming a closed
system (one in which mass does not cross the boundary, meaning evaporation is
neglected here), the First Law states that the energy (E) stored within the system
can be changed (Δ) by either heat entering (Qin) or work leaving (Wout) the system
across the system boundary:

ΔE ¼ Qin  W out

Expressing the change in energy in terms of an initial and final state:

Efinal  Einitial ¼ Qin  W out

Heat is energy that crosses a defined system boundary due to a temperature

difference, the focus of this text. Work is energy that crosses the boundary due to
a force (mechanical or electrical) acting through a distance. The First Law applies
to a specific process that links an initial with a final state. The stored energy (E) is a
state property, or a snapshot. Its value at any time is independent of what went on
before or after. On the other hand, heat and work are forms of energy that are
associated with energy transfer across the system boundary. If a snapshot is taken,
heat and work are not apparent. On the other hand, if a movie is taken, heat and
work will be “seen” to cross the boundary. Heat and work are “process-dependent”
variables.
In this form, the elapsed time between the initial and final states does not come
into play. In traditional thermodynamic analysis, the prime objective is to relate
overall changes in energy content to the total heat and/or work exchanged. In
2.2 Lumped Capacity Model Development 37

traditional heat transfer analysis, where the details of the process as it occurs are
often the objective, a rate basis is generally more advantageous. Dividing the
energy balance by the elapsed time between states (Δt):

ΔE Qin W out
¼ 
Δt Δt Δt
Consideration of the process that takes place over a “short” time interval produces
the illusion of a snapshot (rather than a process, with a before and after), in which
the rate of change of energy is related to the rate of heat and work exchange. Taking
the limit Δt ! 0 and defining a nomenclature q ¼ Qin/Δt and w ¼ Wout/Δt which
have units of energy per unit time (or J/s ¼ Watt in SI units, or Btu/h in English
units), the First Law for closed systems on a rate basis becomes the following first-
order ordinary differential equation:

dE
¼ qin  wout
dt
Note that thermodynamics texts tend to use Q_ for q and W_ for w.

2.2.2 Lumping the Stored Energy into a Single Node

In general, stored energy can be in the form of kinetic energy, potential energy,
and/or internal energy (u). Each element of mass (dm ¼ ρdV) within the defined
control volume has, in general, a different specific energy, the word “specific”
meaning per unit mass. The total stored energy is the sum of the energy of all the
elemental masses. Expressed as an integral:
ððð  2 

V
E ρ þ gH þ u dV
2
CV

In this expression, the triple integral is evaluated over the control volume
(CV) defined within the system boundary by differential (i.e., vanishingly small)
volume elements (dV). In general, the density (ρ), velocity (V  with a line through it
to distinguish from volume), height (H ) in a gravitational field (g) and internal
energy (u) can all vary with position.
For the problem at hand, the kinetic energy associated with motion of the liquid
and any changes in its potential energy as it rises and falls are assumed from the
outset to be negligible in comparison to the changes in internal energy that arise
during the event. There is motion of the liquid, even if it is not being stirred due to
mild density variations, but it seems plausible that the energy associated with that
motion is not important. Without a much more detailed analysis, the neglect of
kinetic and potential energy cannot be further evaluated, but kept in the back of the
38 2 Lumped Capacity Systems and Overall Heat Transfer Coefficients

mind. The heat transfer principles developed in this text can be applied to those
situations, where kinetic and/or potential energy are important.
In general, the internal energy of a system consists of its nuclear, chemical, latent
(i.e., phase change), and sensible (thermal, or that energy associated with its
temperature) components. However, if nuclear or chemical reactions are absent
and there is no phase change within the system boundary, the only energy form
which changes is the sensible component which relates internal energy to the
temperature through the specific heat (cv, in J/kg/K), a thermodynamic function
defined as:

∂u 
cv 
∂T v

where the partial derivative is taken at constant specific volume and indicates that
the internal energy can be expressed as a function of two other thermodynamic
variables (i.e., specific volume and temperature), in general. Note that lower case
“c” is used to denote specific heat (with units of J/kg/K), while upper case “C” is
used to denote total capacitance (J/K). However, for solids and liquids that are not
undergoing phase change, it is generally acceptable for engineering purposes to
assume that the internal energy depends solely on temperature (not pressure, or
volume, or any other thermodynamic variable), and therefore the partial derivative
becomes a total derivative, and the subscript “v” is no longer required:

du  cdT for liquids and solids

Defining a reference temperature (Tref, analogous to a reference height for potential

energy in a gravitational field) at which point internal energy is defined to have a
value of uref (which is often set to zero), the internal energy is obtained by
integration from the reference temperature to the actual temperature (T):

ðu ðT
 
du ¼ u  uref ¼ cdT  c T  T ref
uref T ref

The last expression defines formally this temperature averaged specific heat (which,
in general, still varies with temperature).
In the present case, with the control volume consisting of two very different
substances (mug and coffee), the spatial triple integral that expresses the total
energy stored can be split into two:
ððð ððð
   
E ρc T  T ref dV þ ρc T  T ref dV
coffee mug

There are two material properties (density and specific heat) that vary with position
because the temperature varies with position, and these integrals cannot be further
2.2 Lumped Capacity Model Development 39

evaluated at this point. However, it is generally justifiable in engineering practice to

also assume that density and specific heat are suitably chosen constants. Defining an
average temperature for each subsystem (coffee and mug), the total energy content
can be expressed as:
   
E ¼ ðρcV Þcoffee T coffee  T ref þ ðρcV Þmug T mug  T ref

The combination ρcV for a given subsystem has units J/K and is given the shorthand
nomenclature of a capital C, with an appropriate subscript. That is:

ðρcV Þmug ¼ Cmug

and similarly for the coffee. It is understood that the density and specific heat are
assumed constant with respect to space, yet could still be functions of temperature
(an effect which could be modeled if required).
Numerical values of specific heat can be estimated in a number of ways, but
generally can be searched for on the Internet or in tables from textbooks to obtain
reasonable approximations. Whenever possible, a careful measurement of an actual
sample (Appendix 2) and/or a value obtained from a manufacturer’s specification
sheet for an actual material in an application is the best practice.

2.2.3 Thermodynamic Analysis of Mug Heating Phase

At the start of the event, the mug and coffee have very different temperatures.
Empirically (Fig. 2.3), after a few minutes, they are at nearly the same temperature.
A demonstration of why the mug temperature and coffee temperatures are so close
to each other is delayed until more heat transfer tools have been developed. The
thermodynamic tools are in place to predict the value of the temperature at the end
of the mug heating phase. Recalling the basic First Law expression:

Efinal  Einitial ¼ Qin  W out

Having defined the system to be the coffee plus mug, and the process to start at an
initial temperature where the mug is at ambient temperature and the coffee is at the
temperature it is poured into the mug (near the boiling point of water):
       
Cmug þ Ccoffee T final  T ref  Cmug T mug, initial  T ref þ Ccoffee T coffee, initial  T ref
¼ Qin  W out

The coffee and mug are assumed to be at the same final temperature (which is the
initial temperature of the main cooling of the coffee and mug together). All the
terms with the reference temperature cancel, and the First Law can be rearranged to
solve for the final temperature:
40 2 Lumped Capacity Systems and Overall Heat Transfer Coefficients

Table 2.1 Parameters Property Coffee Mug

related to thermal capacity
Density (kg/m3) 1,000 1,700
(derived quantities in bold)
Specific heat (J/kg/K) 4,200 800
Volume (cm3) 410 197
Mass (kg) ¼ Density
volume 0.410 0.365
Capacitance (J/K) 1,722 292

Cmug T mug, initial þ Ccoffee T coffee, initial Qin  W out

T final ¼ þ
Cmug þ Ccoffee Cmug þ Ccoffee

The combined system exchanges heat with the environment (Qin) but the primary
heat exchange takes place internally between the coffee and mug, which does not
appear directly here because the system boundary does not include it. It is plausible
that the heat loss to the environment is negligible in the relatively short time (3 min)
of the mug heating phase (compared to the 4 h cooling event). The work exchange is
identically zero in this case. Assuming, therefore, an overall adiabatic process
(Qin ¼ 0), the First Law reveals the final temperature of the mug heating phase to
be a capacitance-weighted average temperature:

Cmug T mug, initial þ Ccoffee T coffee, initial

T final ¼
Cmug þ Ccoffee

Table 2.1 shows property values assumed for the heat capacitance. Coffee is
assumed to have thermal properties of liquid water at nominal room conditions.
The mug values are typical of a ceramic material. In Appendix 2, a rough measure-
ment of the specific heat of the actual mug used is described. A majority, 85.5 % of
the total capacitance is associated with the coffee. The heat capacity of the mug
itself is therefore considered to be small but not negligible, and it is easy to account
for this “thermal mass.” Note that the mass of the mug itself is comparable to that of
the coffee, yet its capacitance is small because its specific heat is less than a quarter
of that of the coffee. Liquid water (and therefore coffee) has an exceptionally high
specific heat value, one of the many unique features of water.
Inserting values for this case:

ð292 J= CÞð21  CÞ þ ð1, 722 J= CÞð92  CÞ

T final ¼ ¼ 81:7  C
ð292 þ 1, 722ÞJ= C

Consider the experimental value (from Fig. 2.3) of approximately 75  C for the
peak side wall temperature (the top surface temperature is a few degrees lower). Is
this good agreement or bad? It could be argued both ways. There is uncertainty in
the experimental temperature values, and these values are taken at two places on the
defined control surface. The theoretical temperature calculated is an average tem-
perature of the system. The temperature of the liquid in the interior is expected to be
at a somewhat higher temperature (how much higher will be investigated in
2.2 Lumped Capacity Model Development 41

subsequent chapters from a theoretical, not experimental, point of view). This trend
is consistent with the discrepancy of the measured and predicted temperatures.
What theory and experiment agree on is the qualitative comparison, namely, that
the final temperature of the mug/coffee combination is much closer to the initial
coffee temperature than the initial mug temperature. The theory explains the
observed result because the capacitance of the coffee is roughly six times that
of mug.
By the way, my Mom likes really hot coffee, so when I make her a cup, I pour
boiling water into a mug first to preheat it, then dump that water out (after about
3 min) before filling it with coffee. The final temperature for the mug heating phase
(initial temperature for the mug/coffee cooling phase) is then much closer to the
boiling point of water, the way she likes it.

2.2.4 1-Node Lumped Capacity Model

In conversation, it is no problem to imagine and speak of a cup of coffee as being at

a certain temperature. When pressed, there would be little problem of
acknowledging that perhaps the temperature of the outside of the mug may be
somewhat different from the temperature of the inside liquid, for example. This
trend might be very obvious if a styrofoam mug were used, or if a cardboard holder
placed outside a paper cup to protect the hand were used. Furthermore, the
temperature of the coffee near the walls of the mug or near the surface may be
different than that in the center. So what is really meant by THE temperature of the
coffee? A simplified mental image (or model) in which the entire object in question
has been lumped into a single representative entity has already been intuitively
made. That intuition is now turned into a formality.
The simplest model lumps the entire system (coffee plus mug) into a single
“node” by defining a single average temperature to both triple integrals of the
energy content, and assuming constant properties (ρ and c):
  
E ¼ Cmug þ Ccoffee T  T ref

The rate of change of the stored energy is then simply:

dE dT
¼C
dt dt
where C ¼ Cmug + Ccoffee is the total thermal capacitance of the combined coffee/
mug system. The temperature is best thought of as an average temperature of the
system, not necessarily that the mug and coffee are at the same temperature. The bar
over the temperature will be dropped from here on, keeping in mind that T means
“average temperature.”
42 2 Lumped Capacity Systems and Overall Heat Transfer Coefficients

Modeling, in this case, means expressing the almost infinite complexity implied
by the imposing triple integral in a much simpler (but not simplistic) mathematical
expression. As a result, clarity is gained at the expense of detail.
ððð  2 
d 
V Modeling dT
ρ þ gH þ u dV ! C
dt 2 dt
CV

And we’re halfway to an RC circuit.

2.2.5 Overall Heat Transfer Coefficient: Empirical Value

The other two terms of the First Law, namely, heat and work, are now addressed.
Work represents mechanical energy transfer across the boundary, such as from
expanding pistons, rotating shafts, electrical input and the like, which do not exist
here, so the work is identically zero. Heat represents the flow of thermal energy
across the system boundary, driven by a temperature difference. The rate of heat
transfer between the coffee (at its average temperature value) and the ambient can
be expressed in the form of Ohm’s Law:

T1  T
qin ¼
R
This model states that the rate of heat transfer is proportional to the temperature
difference driving it and inversely with a property of the system called the thermal
resistance (R). The thermal resistance represents the barrier to heat flow between
the system and surroundings that is NOT accounted for in the temperature differ-
ence driving the heat. The thermal resistance can vary with temperature, in general,
although a suitably chosen average value is frequently justified.
Alternatively, the heat transfer rate from a system to its surroundings can be
modeled as being proportional to the driving temperature difference and to the total
area of the control surface, A. The proportionality factor (not necessarily a constant)
is thus defined as an effective overall heat transfer coefficient, U. Note that there is a
nomenclature problem in that U is used to represent both internal energy and overall
heat transfer coefficient, which mean very different things. In heat transfer:
 
qin ¼ UA T 1  T

The “R value” of building materials (given the symbol ℜ to distinguish it from the
thermal resistance R) is the inverse of U, which is the thermal resistance times the
system surface area:

R ¼ 1=U ¼ RA
2.2 Lumped Capacity Model Development 43

Putting it all together, the energy balance becomes:

dT T 1  T
C ¼
dt R
which is the same differential equation obtained for the electrical circuit, with the
same solution if, and it can be a big “if,” R, C, and T1 are constant:

T  T1 t
¼ eτ
T0  T1
where T0 is the initial temperature and τ is a time constant, defined by:

C
τ¼
UA
The response time of a lumped capacity system increases with its total capacitance
(or thermal mass), and decreases with its surface area, and with the overall heat
transfer coefficient, U. In the present example, estimates of the capacitance and
surface area are possible. Therefore, by measuring the time constant from an
analysis of the cooling curve, an empirically determined value of the overall heat
transfer coefficient (U ) can be obtained by comparing the prediction of this lumped
capacity model for different values of U. The capacitance has already been calcu-
lated, and the total surface area of the control surface is calculated from the
geometry of Fig. 2.1 as shown in Fig. 2.5. More than half of the surface area is
found on the side wall, and the remainder is distributed between the bottom, the
inner rim (i.e., the area above the liquid on the inside surface, which obviously
depends strongly on how much liquid is poured in the mug), the top free surface,
and the top edge (assumed to be squared off). The heat transfer process is quite
different on all these surfaces, as will be explored in subsequent chapters, but the
global view taken at this point does not distinguish that detail.
Figure 2.6 shows the measured free surface temperature (taken to be the best
representation of the average coffee temperature available experimentally) and
predictions of the cooling curve for three input values of the overall heat transfer
coefficient (U ). The initial temperature is taken to be that predicted for the end
of the mug heating phase (81.7  C). After the first 10 min, the data is reasonably
banded between U values of 10 and 20 W/m2/ C. A U value of 15 W/m2/ C
yields a reasonable agreement over the entire event, giving a time constant of
τ ¼ 45.1 min from:
 
J min
ð1, 792 þ 272Þ 
C C 60 s 
τ¼ ¼    ¼ 45:1 min
UA J  m2 
15 2 ð495:6 cm2 Þ 
sm C 10, 000 cm 
2

Since the data points do not fall exactly on any line, the response is not truly first
order which could be due to a variation in the parameters (R, C, or T1) during the
44 2 Lumped Capacity Systems and Overall Heat Transfer Coefficients

Fig. 2.5 Surface area of system boundary broken into regions. For the bottom surface, the small
extra area inside the outer rim is obtained assuming the trapped air gap height is the same as the
mug thickness

Fig. 2.6 Empirical determination of the overall heat transfer coefficient, U for the entire event.
Experimental points for the free surface are shown as markers
2.2 Lumped Capacity Model Development 45

Fig. 2.7 Dimensionless temperature vs. time on semi-logy axes. Initial time in this plot is 3 min
into the cooling event, beyond which time the coffee and mug temperatures are observed to fall
together

event (R being the most likely candidate) or phenomena neglected (like evapora-
tion) may be noticeable. Also, a constant reminder that the experimental data is of
the free surface of the coffee, and the lumped model is of an average temperature of
the system.
There are more sophisticated ways to analyze or test for first-order behavior than
this admittedly brute force approach. However, those methods do not necessarily
improve the basic understanding. Brute force is often acceptable, if not preferable,
to getting side-tracked by unwarranted analysis. That is an engineering
judgment call.
A more sophisticated way to present the same data is to plot the dimensionless
temperature vs. time on semi-logy axes, as in Fig. 2.7. Here the initial time is reset
to zero at 3 min into the cooling event because prior to that mug heating makes a
lumped capacity model unreasonable, during which time the coffee and mug do not
fall together; rather, the coffee is heating the mug. Investigation of those effects
require separating the mug and coffee into separate subsystems, as is considered in
a subsequent chapter. The line on the plot shows an exponential curve fit of the
entire event, with the initial value forced to unity. For a pure first-order response, all
data points would fall on this line. The experimental curve initially falls faster than
the overall response, and slower near the end of the event, as thermal equilibrium is
approached. The R2 value of 0.98 suggests that despite these deviations, this event
is reasonably characterized by a first-order event, with a time constant of the inverse
of the number in the exponential of the curve fit:
46 2 Lumped Capacity Systems and Overall Heat Transfer Coefficients

Fig. 2.8 Dimensionless temperature vs. time on semi-logy axes considering only the first hour of
the event. Initial time in this plot is 3 min into the cooling event at which time the measured surface
temperature is 74.0  C, beyond which time the coffee and mug temperatures are observed to fall
together

C 1
τFullEvent ¼ ¼ ¼ 67:3 min
UA 0:014851

The time constant obtained from Fig. 2.7 is targeted toward a simulation of the
entire cooling event, from hot coffee poured into the mug, until the mug and coffee
reach thermal equilibrium with their environment. In practice, the objective is to
understand when the coffee is within an acceptable range of temperature for
drinking, not once it has become tepid. Figure 2.8 shows a semi-logy plot for the
first hour of the event. The time constant during the first hour of the cooling event is
shorter than the time constant that characterizes the entire event:

C 1
τ1stHour ¼ ¼ ¼ 52:3 min
UA 0:0191
The time constant for the first hour is 22 % shorter than that of the full event. That’s
a big difference, despite the fact that the “R2” value was 98 % for the full event.
That suggests the high R2 values can give a rather misleading sense of confidence.
The final result sought from the engineering problem statement, namely, the
time it takes for the coffee, once poured, to cool first to 60  C, and then to 45  C,
thereby defining a window of opportunity. Expressing the first-order cooling curve
in terms of the time:
2.3 A Melting Cup of Ice/Water 47

 
T  T1
t ¼ τ ln
T0  T1

The initial temperature (T0) is taken to be the equilibrium temperature the mug and
hot coffee come to during the mug heating phase (81.7  C), and the time constant is
the experimentally based value of 45.1 min. Therefore, the time to reach 60  C is:
 
60:0  21:0
t60  C ¼ ð45:1 minÞln ¼ 13:8 min þ 3 min offset ¼ 16:8 min
74:0  21:0

The time to reach 45  C is:

 
45:0  21:0
t45  C ¼ ð45:1 minÞln ¼ 35:7 min þ 3 min offset ¼ 38:7 min
74:0  21:0

Of course, this window of opportunity agrees with the direct data plots (Fig. 2.4)
because the time constant was empirically determined, not developed from first
principles. The next chapter introduces the main theoretical heat transfer tools that
can be used to predict this behavior from first principles.
A word about units and absolute temperature is in order. Several quantities have
temperature as one of the units, and those units can be specified as either  C or K
because the quantity itself is based on a temperature difference. For example, the
units of cv are J/kg/K or J/kg/ C because cv is the change in energy divided by the
change in temperature. This author prefers writing these units in the “code” version
J/kg/K rather than the J/kg·K version (which looks like K is in the numerator) or the
J/(kg·K) version. As another example, the overall heat transfer coefficient U is
based on the driving temperature difference, and therefore it is W/m2/K or equiva-
lently W/m2/ C. Some quantities require absolute units, such as the universal gas
constant whose units must be J/kmol/K and not J/kmol/ C.

2.3 A Melting Cup of Ice/Water

At times, I like a nice tall glass of iced coffee. Everyone knows that if you place a
cup of iced coffee on a table top, the ice will eventually melt, diluting the coffee in
the process. The question is, how long does that take? This problem involves heat
transfer, where phase change occurs in the system. In addition, it is another scenario
where an overall heat transfer coefficient can be measured.

2.3.1 Problem Statement

An ice/water slurry is prepared by filling a plastic cup with 0.295 kg of ice and
adding liquid water to completely fill the cup. Solid (ice) and liquid are initially at
the melting point (0  C). The cup is placed on a table in a still room at 18  C.
48 2 Lumped Capacity Systems and Overall Heat Transfer Coefficients

Determine the time it takes for half of the ice to melt, and the overall heat transfer
coefficient, U, based on the exposed surface area of the cup plus slurry to the
environment. The cup has an inner diameter D ¼ 0.074 m, inner height H ¼ 0.144 m,
outer height Ho ¼ 0.166 m, and nominal wall thickness w ¼ 0.0038 m.

2.3.2 Experimental Solution

To prepare this experiment, ice from a conventional household freezer is placed in

the cup, weighed, and allowed to sit at rest in a refrigerator for sufficient time that
the cubes warm from the freezer temperature to the melting point, but not long
enough to begin to melt. A lot could be said about how to determine that, either
experimentally or theoretically. To make a long story short, the change in internal
energy of solid ice from freezer temperature to melting point is much less than the
change in internal energy of solid melting to liquid. As a result, when removed from
a freezer space and placed in the cup, the ice will first warm up to near the melting
point before it begins to melt. That can be proved, just not yet. The liquid water is
prepared by pouring water from a larger container packed with ice and liquid water
filling the void spaces and filled to the brim. In a rough experiment, the cup
dimensions and initial masses of ice and water were those given in the problem
statement. The cup was placed on an inverted plastic container (approximately 5 cm
high) with trapped air inside to minimize (but not eliminate) the heat transfer
through the bottom surface. The test was terminated when, visually, the ice was
approximately half melted. Then the remaining liquid and solid were separated and
the masses measured.
It is more difficult to monitor the progress of this event by measuring tempera-
ture than the cooling of hot coffee because the phase change takes place at a fixed
temperature, namely, the melting point. A measurement of the temperature cannot
distinguish a slurry that is mostly solid from one that is mostly liquid. It is possible
to roughly measure the height of the bottom of the ice/liquid slurry as a preliminary
measure. However, the shape of the ice likely changes with time, and therefore the
packing ratio of solid to liquid volume changes as well.
On the other hand, it would seem that the masses of the liquid and solid could be
measured as the event unfolds. However, that would require separating the phases,
making the measurement of liquid and solid, and then putting them back into the
cup. This procedure would introduce a significant loading error since the experi-
ment called for not disturbing the event, and keeping it stationary. A loading error
in measurements occurs when the act of taking a measurement changes the event
that is to be measured. It is a closely related concept to the Heisenberg uncertainty
principle in quantum mechanics. An alternative would be to repeat the experiment
many times, and change only the length of time of the event. At the end of each
experiment, the liquid would be separated from the solid by pouring it off.
2.3 A Melting Cup of Ice/Water 49

In the experiment conducted, after 3.5 h (210 min), the remaining ice and liquid
were separated and the mass of ice remaining was measured to be 0.125 kg (42.3 %
of original).

2.3.3 Theoretical Model Prediction: Energy Analysis

While the percentage of liquid and solid changes continually, the temperature of the
slurry does not change with time, as a first approximation. In practice, since ice has
a lower density than liquid water, as ice melts, the slurry will separate into two
parts; an ice/liquid slurry in good contact floating above a single phase liquid below.
The liquid on the bottom portion of an ice/water slurry is not constrained to the
melting point. In fact, for an undisturbed scenario like this, the liquid water will
tend to stabilize at 4  C, the temperature where the liquid water density is maxi-
mum. Assuming a constant system temperature makes theoretical modeling simpler
than the cooling mug of coffee. Since the rate of heat transfer is driven by the
temperature difference, it remains constant, unlike the hot coffee cooling where the
heat transfer rate decreases as the coffee approaches ambient temperature.
Defining the cup and slurry (solid and liquid) as a closed system, an energy
balance is:
 
dE d Ecup þ dEsolid þ dEliquid
¼ ¼ Q_ in  W_ out
dt dt
The work done by the system on the surroundings is identically zero. The heat
transfer rate can be expressed as before with an overall heat transfer coefficient,
Q_ in ¼ UAðT 1  T Þ where U is an overall heat transfer coefficient, A is the surface
area of the system, T1 is ambient temperature, and T is the average system
temperature. In this model, the system will be assumed to remain at the melting
point as long as some ice remains.
The energy (E) of each component in general consists of three components:
internal, kinetic, and potential energy. There is little motion in this example, and the
energy of each component is therefore the internal energy per unit mass, u. For the
cup, which does not change phase, there is only sensible internal energy. That is:
 
Ecup ¼ mcup ccup T  T ref

Since nothing in this equation is changing with time, the first term of the energy
equation will be zero when the time derivative is applied. The thermal capacitance
of the cup is not involved in this constant temperature model.
The internal energy of the solid and liquid water have different specific internal
energies, given by uL for the liquid and uS for the solid. Their difference is the heat
of fusion (or heat of melting): uL  uS ¼ uLS ¼ umelt which for water at 0  C has a
value of 330 kJ/kg. The internal energy of the system is thus:
50 2 Lumped Capacity Systems and Overall Heat Transfer Coefficients

 
Ecup ¼ mcup ccup T cup  T ref þ mliquid uL þ msolid uS

The total mass of water remains constant: mH2 O ¼ mliquid þ msolid

Therefore, the mass of liquid can be eliminated:
 
Ecup ¼ mcup ccup T cup  T ref þ ðmH2 O  msolid Þuliquid þ msolid us

Taking the time derivative:

dE dmsolid dmsolid dmsolid

¼ uL þ us ¼ umelt
dt dt dt dt
The energy balance becomes:

dmsolid
umelt ¼ UAðT 1  T Þ
dt
Every parameter can reasonably be modeled as being constant during this event,
and therefore a simple integration, with the initial mass of ice being mICE,0 yields:

UAðT 1  T MP Þ
msolid ¼ mICE, 0  t
umelt

Note that the system temperature has been set equal to the melting point of the
slurry (TMP ¼ 0  C). The mass of ice decreases linearly with time, and it is an easy
matter to calculate the time it takes to melt the ice to any value. In particular, the
total melting time of the ice (tmelt) is determined by setting mice ¼ 0 and solving for
the time:

mICE, 0 umelt
tmelt ¼
UAðT 1  T MP Þ

The only difficult parameter to evaluate is the overall heat transfer coefficient.

2.3.4 Results

The results are reported in Fig. 2.9 (spreadsheet structure) and Fig. 2.10 (plot of
theory and experiment). For the calculation (Fig. 2.9), the surface area was taken to
be the sum of the side surface and the top surface. The bottom surface was assumed
to be well insulated, as attempts to do so were made. The initial ice fraction (0.522)
was determined using “goal seek” to set the known initial mass by changing the ice
fraction. The worksheet was copied twice and values entered in the overall heat
transfer coefficient to encompass the experimental result (Fig. 2.10).
2.3 A Melting Cup of Ice/Water 51

Fig. 2.9 Spreadsheet calculation of the melting cup of ice

The heat transfer coefficient is determined empirically to be a value between 5.0

and 5.5 W/m2/K. This value is of the same order of magnitude, but lower (by a
factor of roughly 3) than the cooling mug of coffee. The external convection
environment in the two examples are comparable (object sitting in a still room).
There are differences, however, such as the thermal conductivity of the material, the
effect of temperature on heat transfer coefficient, and direction of the plume created
that might account for the observed difference in overall heat transfer coefficient.
These effects constitute much of the subject of ensuing chapters.
52 2 Lumped Capacity Systems and Overall Heat Transfer Coefficients

Fig. 2.10 Model predictions for several values of heat transfer coefficient and experimental
results for the melting cup of ice

Appendix 1. Raw Data for Cooling Mug of Coffee

Time (min) Top ( C) Side ( C)

0 92 21
0.33 80 60
0.67 78 70
1 75 74
1.5 72 75
2 73 75
2.5 74 75
3 74 75
4 73 74
5 72 73
6 70 71
7 69 70
8 68 69
9 67 68
10 68 67
11 65 66
12 65 64
13 64 63
14 62 62
(continued)
Appendix 1. Raw Data for Cooling Mug of Coffee 53

(continued )
Time (min) Top ( C) Side ( C)
15 61 61
16 60 60
17 60 59
18 59 58
19 58 58
20 57 57
21 57 56
22 56 56
23 57 56
24 56 55
25 55 55
30 52 51
35 49 48
40 46 45
45 44 43
50 42 42
55 41 40
60 40 39
70 36 35
80 35 34
90 33 33
100 32 31
110 30 30
125 28 28
140 27 27
165 25 25
180 24 24
210 23 23
250 22 22
54 2 Lumped Capacity Systems and Overall Heat Transfer Coefficients

Appendix 2. Measurement of the Mug-Specific Heat

Problem Statement

A mug of mass mmug ¼ 0.335 kg is placed in a large pot of warm water at

Tmug ¼ 41  C and maintained for several minutes. The mug is removed, placed on
a countertop, and 500 mL of water at Twater ¼ 14.5  C is poured into it. After
approximately 2 min, the temperature of water in the mug settles at Tfinal ¼ 17.5  C.
Determine the specific heat of the mug.

Analysis

The First Law of Thermodynamics is applied to a system that consists of the mug
and water poured into it:

ΔE ¼ Q  W

The heat Q represents heat exchange between the mug/water and the environment
during the 2 min settling period, and while not zero, is assumed to be negligible.
There is no work exchange, and evaporation effects are neglected. Therefore, the
change in energy is zero, or the initial energy equals the final energy:

Einitial ¼ Efinal

Considering the energy of the two components (water and mug) relative to any
reference temperature, Tref, and assuming constant specific heats:
   
mmug cmug T mug  T ref þ mwater cwater T water
  T ref 
¼ mmug cmug T final  T ref þ mwater cwater T final  T ref

The terms involving the reference temperature all cancel, and the energy balance
can be solved for the specific heat of the mug:
  
mwater T final  T water
cmug ¼ cwater
mmug T mug  T final

Inserting numerical values, using the density of water to be 1,000 kg/m3 and the
specific heat of liquid water to be 4,200 J/kg/K:
  
0:5 17:5  14:5
cmug ¼ 4200 J=kg=K
0:335 41:0  17:5

cmug ¼ 800 J=kg=K

Workshop 2.1. Cooling Coffee Data 55

Workshop 2.1. Cooling Coffee Data

In a rough experiment, hot coffee was poured into a room temperature ceramic mug
located on a table in a still room. A handheld infrared temperature sensor was used
to measure the temperature of the free surface (top) of the coffee, and the outer side
wall temperature (side) as a function of time. The data is contained in an excel
file (CoffeeData.xlsx, listed also in Appendix 1) which has columns for time, top
surface temperature, and side wall temperature.

IN CLASS EXERCISE: (follow this general approach in documenting future

workshops)

• Open a WORD file and save it as “WKSP_2_1_****.docx”, where **** is the

first four letters of your last name. Add front matter to the DOC file (i.e., name,
date, title). You will be writing a short technical memo, whose target audience is
your peer group (but someone not doing this exercise. . .). Write a few sentences
of Introduction.
• Download “CoffeeData.xlsx”, or enter raw data from Appendix 1. Rename it
“CoffeeData_****.xlsx”.
• Prepare a well-formatted plot of the top and side temperatures vs. time for the
entire cooling event. Remove grid lines. Add an indication on the plot that
indicates clearly the window of time for which the coffee is considered to be
drinkable (define over 60  C as too hot, and 45  C as too cold). Copy the figure
and paste special (as an “enhanced metafile”) into the DOC file and adjust its
size. Add a numbered figure caption. Write a few sentences to describe the
figure.
• In excel, make a copy of the figure and format the x-axis to focus on the first
5 min of the cooling event. Format, copy, and paste special, and add a few
sentences.
• In excel, format the x-axis to be on a logarithmic scale, showing the entire event.
Format, copy, and paste special, and add a few sentences.
• Write a concluding paragraph, and submit the DOC file.
56 2 Lumped Capacity Systems and Overall Heat Transfer Coefficients

Workshop 2.2. Euler Method Simulation of Cooling Mug

of Coffee

A mug of coffee, initially at T0 ¼ 75  C, is placed on a table in a still room at

T1 ¼ 21  C. After 20 min, the coffee temperature is measured to be 57  C. The
coffee continues to cool for a total of 250 min. The total capacitance of the coffee
plus mug is 2,010 J/K, and the total surface area is 496 cm2. Use a numerical
method to determine the inherent time constant for this problem and overall heat
transfer coefficient, and simulate the entire cooling event.

Model Development

The First Law of Thermodynamics applied to the system defined by coffee plus
mug (neglecting evaporation, making it a closed system with no mass flows across
the boundary) is:

dE
¼ Q_ in  W_ s
dt

There is no shaft work, no significant potential energy or kinetic energy changes

(so E is internal energy), and no phase change (so the internal energy E ¼ C
(TTref), where C is a total heat capacitance (in J/K), treated as a constant, and
Tref is a reference temperature). Assume that the rate of heat transfer is proportional
to the instantaneous difference between ambient and coffee temperature
(proportionality constant UA, where U is an effective overall heat transfer coeffi-
cient, NOT internal energy) and A is the total surface area of the system. The First
Law becomes:
 
dC T  T ref dT
¼C ¼ UAðT 1  T Þ
dt dt
Workshop 2.2. Euler Method Simulation of Cooling Mug of Coffee 57

Upon rearrangement:

dT ðT 1  T Þ ðT 1  T Þ
¼ ¼
dt ðC=UAÞ τ

The combination C/UA must have units of time for dimensional consistency, and
that is defined as the time constant for this problem, τ ¼ C/UA. Note that the rate of
change of temperature is proportional to the temperature difference driving the heat
transfer, which decreases with time, whether it is a heating event (a cold beer) or a
cooling event (hot coffee).

Numerical Solution

Forget that this first-order ODE with an initial condition has a simple closed-form
(i.e., analytical) solution. The goal of the workshop is to develop a powerful
numerical technique that will be applied to problems for which an analytical
solution is impossible.
The Euler method “discretizes” time and expresses first derivatives as forward
difference approximations with a small but finite step in time (Δt), namely,

dT T future  T present T ðpþ1Þ  T ðpÞ

 ¼
dt Δt Δt
The superscripts are not powers, they are indices, with (p) meaning present and
(p + 1) meaning one step into the future. Inserting this approximation into the
governing ODE:
 
T ðpþ1Þ  T ðpÞ T 1  T ðpÞ
¼
Δt τ

Solving for the temperature one step into the future yields the Euler formula for
this case:
" #
ðpþ1Þ ðpÞ T 1  T ðpÞ
T ¼T þ Δt
τ

Notice that on the right hand side, the temperature has been given a superscript
(p) because it is changing with each step. The ambient temperature and time
constant are not changing, so they do not need superscripts. They are fixed input
parameters. The time step is a numerical parameter that can be changed by the user.
A simulation is considered accurate if the predicted results are insensitive to the
choice of time step.
58 2 Lumped Capacity Systems and Overall Heat Transfer Coefficients

Guided Spreadsheet Implementation

• Open “CoffeeData.xlsx” and rename it “WKSP2_******.xlsx”.

• On a blank worksheet renamed “SIMULATION”, enter all fixed physical
quantities in a block labeled “INPUT PARAMS” (i.e., T0, T1, C, A). These
values are known and will not change. Below that enter a block labeled
“RESULTS” and enter a time constant in a cell to any value. It is a constant in
the simulation, but its value is unknown and will be solved for (It is a type of
“eigenvalue”). Add a cell to calculate U ¼ C/τ/A in SI units (W/m2/K).
• Below that, start a labeled block “NUMERICAL SIMULATION”. Enter a
numerical step size, Δt, with a value of 0.5 min (but you can change this later
to explore its effect. . .).
• In a “header row,” label three columns; for index (p), time (min), and tempera-
ture ( C).
• In the next row, enter 0 for “p” and time. Enter reference to the cell, where initial
temperature is (T0). This row represents the initial condition, and these values are
known inputs.
• In the next row, enter the three appropriate formulas, each of which refers to the
initial condition row, and to input or numerical parameters. Use dollar signs ($)
to lock in rows and/or columns as appropriate for copying the formula down. The
index (p) is incremented by 1, the time is incremented by Δt, and the temperature
is incremented by the Euler formula for this problem.
• Copy this row down until the time elapsed equals the total simulation time
(250 min, which requires 500 rows for this time step).
• Create a scatter plot (with straight lines, no markers) of temperature vs. time, and
move it near the top of “SIMULATION” so you can monitor it.
• Add a series that enters the two data curves. Change its format to “markers with
no lines.”
• Manually change the value of the time constant to obtain a better fit (it will never
be perfect, there are assumptions in the model that are never completed
satisfied. . .).
• Now use “goal seek” (in Data, What if) by setting the temperature in the cell at
20 min to 57  C by changing the time constant.
• Play around with the value of the numerical time step (Δt), then set it back to
0.5 min.
• Upload the file.

C’mon, admit it. This was fun!

Workshop 2.3. Floating Thermometer Response Time

All physical (as opposed to optical) temperature measuring devices are subject to
both static and dynamic errors. A static error occurs when the sensor is in thermal
equilibrium with its environment, but its temperature is different than the object or
Workshop 2.3. Floating Thermometer Response Time 59

fluid it is intended to measure. For example, a temperature sensor placed in sunlight

will be hotter than the fluid surrounding it. A dynamic error occurs when the
temperature being measured changes with time, and the temperature sensor is not
in thermal equilibrium. This workshop focuses on dynamic response, specifically,
the determination of the inherent response of a particular thermometer to a sudden
change in temperature. The response depends on the nature of the fluid the ther-
mometer is placed in.
A measurement of the response time of a temperature sensor in a particular fluid
environment can be obtained by the following general steps:

1. Precondition the temperature of the temperature sensor to a different tempera-

ture than the fluid.
2. Suddenly place the thermometer into the fluid.
3. Measure the indicated temperature as a function of time until the thermometer
achieves equilibrium with the fluid.
4. Analyze the data for response time, and overall heat transfer coefficient.

In designing the experiment, the temperature of the fluid should not change
substantially in the time it takes for equilibrium to be achieved, which can require
patience. It is important to know the final equilibrium temperature to determine the
true response time.
A response time experiment was conducted with a floating thermometer, an
inexpensive liquid-in-glass sensor designed to measure the temperature of con-
sumer aquariums. The construction consists of a glass cylinder of length 10.8 cm
and outer diameter 1.1 cm (with rounded ends). Tiny balls made of lead are located
at one end which causes the thermometer to float vertically in water, with approxi-
mately 10.0 cm submerged (0.8 cm above the surface). Above the lead balls is a
reservoir of sensing liquid and a smaller cylinder in which the thermometer fluid
rises and falls with temperature. A calibrated scale placed adjacent to this inner
cylinder allows for measurement. For thermal modeling purposes, the mass of the
whole system can be considered to be dominated by the lead.

Measurement of Response Time

The response was determined for two different fluid environments: ambient air and
nominally room temperature water. In both cases, the thermometer was
preconditioned to a lower temperature by allowing it to equilibrate with a cold
water bath. The raw data is given in Fig. 2.11. Blank columns are left for the
calculation of the dimensionless temperature and the response time, which can be
calculated for each individual data point. Note that the initial temperature is the first
data point. The fluid temperature (Tinf), reported at the top of each table, was
determined after approximately five times longer than the last reported data point.
In addition to calculating the response time associated with each point, prepare
plots of the raw data (temperature vs. time) and a plot of the dimensionless
60 2 Lumped Capacity Systems and Overall Heat Transfer Coefficients

Fig. 2.11 Spreadsheet templates and raw data for dynamic response of a floating thermometer

temperature vs. time on semi-logy axes. An exponential fit (with the x-intercept
forced to a value of unity) can be added to each curve. The degree to which the
system obeys a first-order response is determined by the consistency of the individ-
ual response times, or equivalently from the linearity of the exponential fit.

Measurement of Overall Heat Transfer Coefficient

The overall heat transfer coefficient (U) for the two cases can be determined from
the definition of the response time, namely, τ ¼ RC ¼ C/(UA). An estimate of
the total capacitance (C) can be made by assuming that the thermal properties
(i.e., specific heat) is approximately that of the lead used to weigh it down. The
mass of the thermometer can be determined by measuring the fraction of the length
that is above water when floating, and using Archimedes’ principle (the weight of
an object equals the weight of the displaced fluid).
Heat Transfer Modes: Conduction,
Convection, and Radiation 3

Heat transfer is energy transfer driven by a temperature difference. A first step in

estimating the rate of heat transfer between, say, a hot mug of coffee and its
environment is to identify the modes of heat transfer and the thermal channels of
heat flow between two locations. There are different pathways, or thermal channels,
that heat can take from hot coffee to ambient; directly from the coffee to the air
above the free surface, through the side wall of the mug, or through the bottom of
the mug and through the floor on which the mug rests.
Within each of these pathways, heat flow encounters discontinuities in material
properties, from coffee, to mug, to air. The mathematical expression of the different
underlying physics associated with these discontinuities gives rise to three modes of
heat transfer: conduction, convection, and radiation. In contrast, there are two
mechanisms of heat transfer (conduction and radiation) that refer to the physical
phenomena that give rise to the energy transfer. The mode of heat transfer by
convection is based on a conduction mechanism. A thermal channel is defined as a
“pathway” for heat to flow from a hot fluid or surface to a cold one, expressible in
terms of a series combination of heat transfer modes.
In this chapter, the practical application of a single-paned window is used to
develop the engineering tools (and the thinking process behind them) used in heat
transfer analysis. This example involves a single thermal channel. The basic
method is applied more generally, and to the coffee/mug problem with its multiple
channels, in subsequent chapters.

3.1 Example: Single-Paned Window

On the north side of a building, (so that there is no solar gain), a single-paned glass
window (shown schematically in Fig. 3.1) of thickness Δx ¼ 0.012 m, height
H ¼ 2 m, and width W ¼ 8 m (into the page) is exposed on the inside to room air
maintained at TH ¼ 25  C. The outside of the window is exposed to a crosswind of
10 miles per hour (¼4.47 m/s) and an outside air temperature of TC ¼ 0  C.

# Springer International Publishing Switzerland 2015 61

G. Sidebotham, Heat Transfer Modeling: An Inductive Approach,
DOI 10.1007/978-3-319-14514-3_3
62 3 Heat Transfer Modes

Fig. 3.1 Cross-section of single-paned window with representative temperature distribution.

Both surfaces are exposed to a fluid (air, in this case) and are enclosed by far walls which exchange
thermal radiation with the window

Estimate the rate of heat transfer through the window (in Watts), the temperature
of the inside and outside surfaces of the glass (T1 and T2) and the daily energy
cost associated with heat loss through the window, assuming the price of heating to
be p ¼ $20/GJ.

3.2 Heat Transfer Modes and Thermal Channels

The three modes of heat transfer used in engineering practice are introduced in this
section. A brief definition of the modes is followed by the development of a thermal
channel and its representation as a resistance network, and finally the mathematical
models used for the modes are defined.
Conduction: Heat transfer through a solid is called conduction, and is governed by
Fourier’s Law, defined shortly. In the window problem, conduction heat transfer
takes place between the inner and outer surfaces of the glass of a known thickness
Δx. In the mug of coffee example, conduction occurs across the mug wall.
Convection: Heat transfer between a fluid (gas or liquid) and a solid surface is called
convection and is governed by Newton’s Law of Cooling. There are two convective
3.2 Heat Transfer Modes and Thermal Channels 63

surfaces in the window problem, the inside surface, which is exposed to room air,
and the outside surface, which is exposed to outside air (with a crosswind). In
reality, the physical mechanism of convection is conduction across a thin layer of
fluid, called a thermal boundary layer, shown with thicknesses δ1 and δ2 in Fig. 3.1.
However, the thickness of these layers is unknown, and they are heavily influenced
by the motion of the fluid.
Radiation: Heat transfer between two solid surfaces that “see” each other and
thereby exchange electromagnetic energy, or radiation, is governed by the Stefan–
Boltzmann Law. Like convection, radiation is associated with a specific surface,
and there are two of them in the window problem. However, the physical mecha-
nism of heat transfer is very different from that of convection. Accelerating charged
particles (electrons and protons) in the solid emit photons (or electromagnetic
waves) that carry energy with them. Simultaneously, photons that emanate from
other surfaces that impinge on a surface can be absorbed. Therefore, a net exchange
of energy occurs. Radiation is much more difficult to model than conduction and
convection, in general. Good engineering judgment, from experience, dictates
when it is essential to include it, and how to do so in an approximate way.
Mass Transfer: In addition to the three modes of heat transfer, energy flows
associated with mass crossing the boundary can be important. There are two
mechanisms for mass transfer effects: advection and diffusion. Advection (often
called “convection” but with a very different meaning than the mode of convection
between a fluid and solid surface) is the energy associated with bulk fluid flow across
a boundary. The enthalpy is the natural thermodynamic property that accounts for
both internal energy flowing in and the work required to do so (the “flow work”). For
example, fluid flowing into a heat exchanger carries energy with it, and exits the heat
exchanger at a different energy level. The other way energy crosses the boundary is
by molecular diffusion (driven by concentration gradients). Evaporation effects fall
into this category. These effects are developed in Chapter 13, and results will be
discussed in relation to the cooling mug of coffee problem.
As will be seen, conduction and convection are naturally expressed in an Ohm’s
Law form, where the heat transfer rate (current) is proportional to a temperature
difference (voltage drop), and inversely with a thermal resistance (which may be a
variable). There are important practical cases where radiation can be modeled in
Ohm’s Law form as well and others where it can be modeled as a current source.
These cases are developed in this chapter. Cases where these models do not
capture the controlling physics are left to other textbooks. Mass transfer by advec-
tion can be modeled as a dependent current source (i.e., the current source depends
on the enthalpy values entering and exiting, developed in Chapters 11 and 12).
Mass transfer by diffusion can be modeled either as a current source, or in an Ohm’s
Law fashion, but with effective R values that are extremely temperature dependent.
A lumped capacity (or 1-node) thermal resistance model of the single-paned
window is shown in Fig. 3.2. All the capacitance of the window ðC ¼ ρcV ¼
ρcΔxHwÞ has been lumped into a single node, placed at the midway point of the glass,
and given a temperature with a bar over it to indicate that it is an average window
64 3 Heat Transfer Modes

Fig. 3.2 Lumped capacity (i.e., 1-node) model of a single-paned window

temperature, and it varies with time, in general. This node is represented with an open
circle, indicating that it carries capacitance and is governed by a differential equation,
in the general case. Nodes are placed at the inner and outer surfaces of the window
because these locations are discontinuous in space (air on one side, glass on the other),
and there is a distinct change in heat transfer mode from one side to the other. These
temperatures are unknown (prior to analysis), but can be related to the capacitive node
through algebraic manipulation and are therefore represented by filled circles.
There are four fixed temperatures represented by batteries; fluid and far wall, on
each side of the window. The fluid (air) in the immediate vicinity of the window
exchanges heat by the mode of convection, and a thermal resistance with a subscript
“h” is used (because “h” will be used in defining a convection coefficient). In addition,
each surface exchanges heat with “walls” that can be far removed from the window.
The average temperature of these walls is treated as fixed. The window surfaces
exchange heat with these removed surfaces through the mode of radiation, and a
thermal resistance with a subscript “r” is used for these modes. For example, the outer
window surface “sees” anything that an observer inside the building looking outside
would see; other buildings, the street, the sky. The inner window surface “sees” what
someone would see looking in: ceiling, floor, walls, furniture, people, etc. In general,
the fluid temperature in the vicinity of the window and the far wall temperatures
could be different from each other. A wisely chosen average value is required if the
floors, ceilings walls, and occupants are at widely different temperatures.
The mode of heat transfer between the inner and outer window surfaces and the
midpoint of the window is conduction, and conductive thermal resistances (Rk/2)
are placed between each surface node and the interior node. The total conductive
resistance between the inner and outer surface (Rk) has been split into two equal
resistance values.
The RC network of Fig. 3.2 has four thermal channels; from each of the four
fixed ambient temperatures to the average temperature of the window itself. After
setting up the general transient solution, this problem will be reduced to a single
thermal channel from inside to outside by considering only the steady-state solu-
tion, and setting the fluid and radiation wall temperatures equal to each other.
Then heat will flow from inside to outside through a single thermal channel.
3.2 Heat Transfer Modes and Thermal Channels 65

A nodal energy balance applied to the window node is:

dT T H, fluid  T T H, walls  T T C, fluid  T T C, walls  T

C ¼ þ þ þ
dt R1, h þ Rk =2 R1, r þ Rk =2 R2, h þ Rk =2 R1, r þ Rk =2

Each of the four heat transfer rate terms on the right side model the four thermal
channels. If the five thermal resistances are treated as constants, this equation can be
recast into a general first-order system with a steady-state temperature and an
effective RC time constant. Alternatively, the energy balance can be solved numer-
ically, and that would allow for models for which the thermal resistances depend on
temperature.
The temperatures of the two window surfaces (T1 and T2) are not directly solved
for in the governing equation. However, they are interior temperatures within
thermal channels and, as derived in Chap. 1, they can be expressed in terms of
their neighbors as:

T H, fluid T H, walls T C, fluid T C, walls

R1, h þ R1, r þ Rk =2 R2, h þ R2, r þ Rk =2
T T
T1 ¼ and T2 ¼
R1, h þ R1, r þ Rk =2 R2, h þ R2, r þ Rk =2
1 1 1 1 1 1

However, the window problem at hand is really a quasi-steady state problem. The
problem statement states that the inside and outside temperatures are fixed. Even if
the inside or outside temperatures were changing with time, in practice, it is
expected that they would do so gradually, and the window would maintain the
quasi-steady temperature obtained by setting the time derivative to zero. Formally,
it would have to be shown that the rate of heat flowing into the capacitor is small
compared to the rate of heat that flows from the inside surface (T1) to the outside
surface (T2). Here is a case that could certainly be tested by running a transient
simulation as outlined, and then neglecting the capacitance effects. The problem
statement would be modified to specify the manner in which either the hot or cold
ambient temperatures varied with time.
At steady-state, then, heat does not flow into or out of the capacitor, and in
representing the network, the capacitor can be removed. In this case, heat from the
inside window surface to the outside surface takes place through two series resistors
of Rk/2. The average temperature of the window (by a voltage divider) equals the
geometric average of the inside and outside temperatures. The simplified thermal
resistance network that carries no capacitance is shown in Fig. 3.3. In general, for
problems that are considered to be at steady-state, RC networks need not show the
inherent capacitances associated with material because there is no change in energy
stored, as is assumed in this case.
A further simplification can be made if it is reasonable to assume that the fluid
temperatures and the radiation wall temperatures are equal. That is, if

T H, fluid ¼ T H, walls ¼ T H and T C, fluid ¼ T C, walls ¼ T C

66 3 Heat Transfer Modes

Fig. 3.3 Single-paned window resistance network at steady-state

Fig. 3.4 Single-paned window resistance network for steady-state conditions, with equal fluid
and wall temperatures, and shown in terms of a single equivalent resistance

then only a single battery on either side can be used, and the simplified resistance
network is shown in Fig. 3.4. Notice now that radiation and convection act in
parallel with each other, and in series with conduction (and the convection/radiation
on the other side). Care must be taken in assuming that these two driving
temperatures are the same. There are important practical cases where an object
exchanges heat with a radiation partner at a very different temperature than the fluid
in the immediate vicinity of the object.
This problem has been reduced to a model of heat flow through a single thermal
channel from one temperature source to another. Within this channel, all three
modes of heat transfer are involved, with distinct temperature drops along the way.
3.3 Fourier’s Law of Conduction 67

The heat transfer rate through the window, as sought from the original problem
statement, is expressed in terms of the overall temperature drop and thermal
resistances and, equivalently, in terms of an overall heat transfer coefficient (U):

TH  TC
q¼ ¼ UAðT H  T C Þ
Req

where the equivalent thermal resistance is:

1
1
1 1 1 1
Req ¼ þ þ Rk þ þ
R1, h R1, r R2, h R2, r

The basic laws that govern the modes of heat transfer, and methods for estimating
resistance values, are developed in the following sections.

3.3 Fourier’s Law of Conduction

Focusing first on the conductive heat transfer that takes place across the window
itself between the inner and outer surfaces of the glass, Fig. 3.5 shows a schematic
of a solid wall consisting of two surfaces (1 and 2) with solid material between them
that are maintained at different average temperatures. It is emphasized that these
surface temperatures are not generally equal to their adjacent fluid temperatures.
Conduction through the wall is governing by Fourier’s Law, which states that, at
steady-state, the rate of heat transfer between two surfaces (1 and 2) with solid
material between them is proportional to the area of the surfaces (A12, which can be
an average area if they are different), the temperature difference between them
(T1  T2), and inversely with the distance between them (Δx12, which can be an
average distance). The proportionality factor (not necessarily a constant) is a
material property called the thermal conductivity (k12), which has SI units W/m/K.
Fourier’s Law of Conduction is expressed mathematically as:

k12 A12 ðT 1  T 2 Þ T 1  T 2
qconduction, 1!2 ¼ ¼
Δx12 R12

Fig. 3.5 Rectangular solid

wall. The surface on the left is
considered to be surface
1, and the surface on the right
surface 2. All six other faces
are considered to be “edge”
faces
68 3 Heat Transfer Modes

where a generic definition of a conductive resistance has been introduced as:

Δx12
R12 ¼
k12 A12
In general, since the area of the two surfaces can be different, and the distance
between them might vary somewhat, these expressions are approximate for a more
general geometry than a plane wall.
Figure 3.6 shows the thermal conductivity of a variety of materials (solids,
liquids, and gases) at nominal room temperature (25  C). The thermal conductivity
of most materials is a mild function of temperature, but a suitably chosen average
value can be used for most engineering applications.
For the window problem, the conductive resistance between in inner and outer
surfaces is:
 
Δx12 0:012 m 1, 000 W 
R12 ¼ ¼    ¼ 0:781 C

k12 A12 ð0:96 W=m= CÞð2 m  8 mÞ  kW  kW

Alternatively, Fourier’s Law can be expressed in terms of the heat flux crossing a
given area at a location x (A(x), where the area is normal to and can vary with x) by
postulating that the rate of heat transfer is proportional to the surface area, as before,
and to the temperature gradient perpendicular to the area. The thermal conductivity
is the proportionality factor. That is:

∂T
qconduction, x ¼ kAðxÞ
∂x
The minus sign accounts for the direction of heat flow from high to low tempera-
ture, according to the Second Law of Thermodynamics. That is, if the temperature
increases in the direction of x (dT/dx > 0), then heat will flow in the negative
x direction.
The Appendix contains a derivation of the thermal resistance for a general
geometry in which the area can be expressed as a function of a single dimension.
The result for the general formula is:

xð2
1 dx
R12 ¼
k12 AðxÞ
x1

This formula is applied to three well-defined geometries: rectangular, cylindrical,

and spherical. The result R12 ¼ Δx12 =ðk12 A12 Þ is the exact result for a rectangular
geometry, or a plane wall (Cartesian, where the area and distance are constant).
Across a cylindrical shell (where the distance is constant but the area varies), the
thermal resistance for a pipe of length L into the page, outer radius r2 and inner
radius r1 is:
3.3 Fourier’s Law of Conduction 69

Fig. 3.6 Thermal conductivity of various materials. Adapted from http://www.

engineeringtoolbox.com/thermal-conductivity-d_429.html, accessed 1/30/2011
70 3 Heat Transfer Modes

 
r2
ln r1
R12 ¼ for a cylindrical shell
2πk12 L
For a spherical shell, with outer radius r2 and inner radius r1, the thermal
resistance is:


1 1 1
R12 ¼  for a spherical shell
4πk12 r 2 r 1

For any other geometry, an approximate model can be used, which involves first
defining two surfaces, calculating an average distance between them, and an
average area. A general approach is:

1. Identify two surfaces of the solid in question between which a conductive

resistance is to be estimated.
2. Calculate the area of these two surfaces and the geometric average (A12).
3. Estimate an average distance between the two surfaces (Δx12). This step can be
difficult to conceptualize for complicated shapes.

The conductive resistance is R12 ¼ kΔx 12

12 A12

3.4 Newton’s Law of Cooling

The mode of heat transfer between a solid surface and an adjacent fluid is termed
convection. It is governed by Newton’s Law of Cooling, which states that the rate of
heat transfer between a surface and an adjacent fluid is proportional to the surface
area and to the temperature difference between the surface and the fluid just outside
a thermal boundary layer (a region of influence within which the temperature
changes from the surface temperature to the free stream fluid temperature). The
proportionality factor (not necessarily a constant) is termed the convection coeffi-
cient, with the symbol “h” (not enthalpy, a totally different quantity). Newton’s
Law of Cooling is expressed mathematically as:

  T surface  T fluid
qconvection ¼ hAsurface T surface  T fluid ¼
Rconvection
A thermal resistance associated with a convection boundary has been introduced,
which is inversely proportional to the convection coefficient and the surface area. In
some models, the convection coefficient can be a function of temperature (and thus
variable) and the surface area could change with time (i.e., a moving piston changes
the area of the cylinder wall exposed to the gases it contains).
The mechanism of convection is actually conduction. Heat flows by conduction
across the thermal boundary layer, which is usually thin compared to the size of the
3.4 Newton’s Law of Cooling 71

Table 3.1 Approximate convection coefficient ranges (SI units)

Gases Liquids
(W/m2/ C) (W/m2/ C)
Natural convection
A natural convective flow environment is one in which 1–10 100–1,000
there is no imposed flow velocity on the surface.
Gravity causes an induced buoyant flow, however
Forced convection
A forced convective flow environment is one in which 10–100 1,000–10,000
the surface experiences an identifiable imposed velocity. The
wind chill effect is a common experience of forced convection

physical object (except in microscales), driven by a temperature difference. That is,

heat is conducted from the fluid to the surface through the fluid in the thermal
boundary layer of thickness δ.

1 δ
Rconvection ¼ 
hAsurface kfluid Asurface

That is:

kfluid
h
δ
As might be intuitive, the thermal conductivity of the fluid is directly linked to the
convection coefficient (a measure of the effectiveness of the heat transfer between
the fluid and an adjacent solid). In addition, the thinner the thermal boundary layer,
the higher the convection coefficient. This thickness is influenced by the fluid
motion, as well as fluid properties (thermal conductivity, density, and specific
heat) and will be explored more fully later.
Chapter 9 develops a procedure for estimating convection coefficients using
dimensionless quantities (Nusselt number correlations) that characterize the flow
and temperature fields. Until then, appropriate values (or formulas) for convection
coefficients will be given in problem statements. Table 3.1 shows the order of
magnitude range expected for convection coefficients in SI units. These rules of
thumb can be used as a first guess, or as a reality check for a detailed calculation.
The art of making appropriate choices for numerical values for convection
coefficients and/or whether they are treated as variables (in space and/or tempera-
ture) is developed with experience.
For the window problem as stated, the inside window surface is exposed to room
air that is nominally still. That is, there is a natural convection with air as the fluid,
and a convection coefficient value between 1 and 10 W/m2/ C is expected. The
outside window is exposed to a crosswind. That is, there is a forced convection with
air as the fluid, and a convection coefficient between 10 and 100 W/m2/ C is
expected.
72 3 Heat Transfer Modes

For the window problem considered here, the ballpark estimates for the window
problem are:
W
h1 ¼ 5:0 for the inside surface (typical of natural convection in air)
m2 K
W
h2 ¼ 15:0 2 for the outside surface (typical of forced convection in air)
m K
The convective thermal resistances in the window problem are therefore:
 
1 1 1, 000 W 
R1, h ¼ ¼   ¼ 12:5 C

h1 A ð5 W=m2 =KÞ  ð16 m2 Þ kW  kW

 
1 1 1, 000 W 
R2, h ¼ ¼   ¼ 4:17 C
h1 A ð15 W=m2 =KÞ  ð16 m2 Þ kW  kW

These resistances are significantly larger than the conductive resistance (0.78  C/
W), indicating that there is a larger barrier to heat flow at the two convective
surfaces compared to the conductive resistance across the window itself. However,
radiation acts in parallel, and it must be estimated before a final conclusion can be
made on that score. . .

3.5 Stefan–Boltzmann and Kirchhoff’s Radiation Laws

The mechanism of radiation heat transfer is fundamentally different than that of

conduction. Conduction involves the direct interaction of adjacent molecules that
vibrate at different energy levels (indicative of temperature), and energy is trans-
ferred from regions of higher to lower temperature. On the other hand, radiation
heat transfer is the result of energy transport associated with electromagnetic
waves, or photons, which requires no medium. Each photon carries energy with it
in the amount ђν (where ђ is Planck’s constant and ν is the frequency) at the speed
of light.
A black body is one that absorbs all photons incident upon it. The ability of an
object to absorb photons is closely related to its ability to generate (emit) photons.
The Stefan–Boltzmann Law states that the total power emitted from a black body
with a surface area A and absolute temperature T is given by:

qBlackBody ¼ σAT 4 ðin WattsÞ

where σ ¼ 5.669(10)8 W/m2/K4 is the Stefan–Boltzmann constant. This equation

must be in absolute (K or R) temperature units. This energy flux represents an
3.5 Stefan–Boltzmann and Kirchhoff’s Radiation Laws 73

integration over all wavelengths of light. This law does not depend on the material
in any way, other than its blackness. However, no real bodies are perfectly black,
and they emit less total radiation than a black body. For engineering purposes, the
introduction of the so-called grey body defines a material property, the emissivity
(ε), that represents the ratio of the actual emitted power to what a black body
would emit:

qGreyBody
ε¼
qBlackBody

The emissivity is a crude, but very successful, modeling tool for estimating radia-
tion heat transfer. There is an extremely complex relationship between wavelength
and emissivity, influenced strongly by quantum mechanical effects. The molecules
of a real body tend to emit radiation strongly in localized bands of frequency.
Emissivities of representative material surfaces are shown in Fig. 3.7. Highly
reflective materials tend to have low emissivities (bottom plot). Most other
materials tend to have somewhat higher emissivities.
Not only do surfaces emit radiation, but they can also absorb incident radiation.
The absorptivity (α) is defined as the fraction of incident light that is absorbed by
the material. At a given wavelength of light, the absorptivity equals the emissivity
(Kirchhoff’s Law of Thermal Radiation).
However, the emissivity can be very different at different wavelengths. For
example, glass has a relatively high emissivity (and absorptivity) for infrared
electromagnetic radiation, but a low emissivity (and absorptivity) to visible light.
Therefore, windows are transparent to visible light, but are opaque (with a high
emissivity) to infrared, which is where terrestrial thermal radiation is centered.
To add further to the complexity, while a given surface is emitting radiation
(from the vibration of its molecules), photons that strike it from other sources can
either be transmitted (pass through, like visible light through a window), absorbed
near the surface, or reflected. Keeping track of all these phenomena involves
complicated geometric calculations, and a general treatment of radiation is beyond
the intended scope of this first course in heat transfer.
Fortunately, useful engineering approximations to radiation heat transfer can
often be made, so that radiation can often be included in a thermal resistance model.
In such limiting cases, radiation heat transfer involves heat transfer from (or to) a
surface. Radiation is like convection in that way. However, because the physical
mechanism is entirely different, the net heat transfer rate from (or to) the surface is
the algebraic sum of radiation and convection. That is, radiation and convection act
in parallel, not series. Convection involves thermal contact with the fluid immedi-
ately adjacent to the surface. Radiation involves thermal contact with distant
surfaces as electromagnetic waves travel between them. Two situations allow for
simplified models: an object completely surrounded by a much larger object, and a
radiation source imposed on an object.
74 3 Heat Transfer Modes

Fig. 3.7 Emissivity of various materials. The uncertainty (not shown) of all these values is quite
high. Data adapted from http://www.engineeringtoolbox.com/emissivity-coefficients-d_447.html,
accessed 2/5/2012
3.5 Stefan–Boltzmann and Kirchhoff’s Radiation Laws 75

3.5.1 Case 1: Surrounded Object

If two objects, separated by a gas (or vacuum) that does not significantly absorb
radiation, can “see” each other, they exchange heat via radiation. When a convex
object (one that cannot see itself) with a surface temperature of T1 is completely
surrounded by a much larger object with a surface temperature T2 (shown
schematically in Fig. 3.8), the rate of heat transfer due to radiation between them
can be approximated as:
 
qrad, 1!2 ¼ ε1 σA1 T 41  T 42

where

ε1 is the emissivity of object 1 (with a value between 0 and 1)

σ ¼ 5.669(10)8 W/m2/K4 is the Stefan–Boltzmann constant
A1 is the surface area of object 1

These temperatures must be in absolute units (K or R).

Since this relationship is not linear with respect to temperature, it cannot be used
in this form for Ohm’s Law analogy (i.e., Ohm’s Law is NOT V4 ¼ IR). However,
by expanding the temperature term, it can be rewritten in an alternate form that
isolates on the temperature difference (T1  T2), namely:
 
qrad, 1!2 ¼ ε1 σA1 T 21 þ T 22 ðT 1 þ T 2 ÞðT 1  T 2 Þ ¼ hr A1 ðT 1  T 2 Þ

where a radiation coefficient, hr, has been “defined” as:

 
hr ¼ ε1 σ T 21 þ T 22 ðT 1 þ T 2 Þ

The heat transfer rate can now be expressed as a thermal resistance between the two
objects:

T1  T2 T1  T2
q12 ¼ ¼
ð1=hr A1 Þ Rrad

Since the rate of heat transfer by radiation is now expressed in terms of a tempera-
ture difference, the temperatures of the nodes in a resistance network need not be in
absolute units, as long as the radiation coefficient (hr) is calculated using absolute
temperature.

Fig. 3.8 Radiation exchange

between a surrounded convex
object and a surrounding
cavity
76 3 Heat Transfer Modes

50
45 ε1 = 0.8
Radiation coefficient (W/m2/K)

40 T1 = 298K

35 "h2"
30
25
h
20
15
10
5 "h1"

0
300 350 400 450 500 550 600 650 700 750 800
T2 (K)

Fig. 3.9 Radiation coefficient for object 1 completely surrounded by object 2

The radiation coefficient depends strongly on the two surface temperatures, but
only on the emissivity of the smaller, surrounded, object. If these surface
temperatures are not known beforehand, the radiation coefficient cannot be
evaluated initially, and an iterative solution technique is needed. However, an
excellent first (possibly last) approximation can be made by “guessing” the surface
temperatures. To help make an intelligent guess, notice that, for T2 > T1, the value
of hr can be bracketed by:

}hr1 } ¼ ε1 σ4T 31 < hr < ε1 σ4T 32 ¼ }hr2 }

Figure 3.9 shows these radiation coefficients for a surface 1 maintained at room
temperature for temperatures up to 800 K. For moderate temperature differences
between object 1 and 2, the radiation coefficient can be reasonably assumed to be
constant, on the order of 5–10 W/m2/K.
Since radiation and convection act in parallel, radiation is important when the
radiation coefficient is of the same order or much larger than the convection
coefficient. This situation occurs almost always when the convection environment
is natural convection of a gas (in which case hconvection is of the order of 1–10
W/m2  C). Radiation can be important in other convection environments, especially
at high temperatures, such as in furnaces.

3.5.2 Case 2: Radiation Source

Solar radiation incident on the Earth is approximated by a different model; the sun
does not completely surround the Earth. However, solar radiation can be treated as a
current source (Fig. 3.10).
3.5 Stefan–Boltzmann and Kirchhoff’s Radiation Laws 77

Fig. 3.10 Radiation modeled as a current source to a surface at temperature T1

Fig. 3.11 Thermal

resistance model of the Earth

A flat surface held perpendicular to the sun’s rays outside the influence of the
atmosphere receives a solar flux of 1,400 W/m2 (the “solar constant”). This radia-
tion is partially attenuated as it travels through the Earth’s atmosphere. On a bright
sunny day around noon, a flat plate held normal to the sun will receive at most
approximately 1,000 W/m2. Some of this radiation is reflected, some is absorbed,
and the rest is transmitted. How much radiation is absorbed depends on the
material. Black objects absorb a lot; reflective objects absorb a little.
Other radiation sources can be treated as a current source. For example, in a
toaster oven, electrical energy is converted to thermal energy in a resistive element,
creating a glowing hot wire coil. Thermal radiation emitted by the coils is absorbed
by the surface of the toast which warms, dries, chars, and eventually burns.

3.5.2.1 Example: Estimate the Steady-State Temperature of the Earth

The Earth is bathed in a constant bombardment of photons from the sun, a portion of
which is absorbed by the atmosphere, ground and oceans, the rest being reflected.
This absorption of energy is modeled as a current source, labeled “Solar” in the
resistance network shown in Fig. 3.11. Simultaneously, the Earth is modeled as an
object completely surrounded by a larger “object,” namely outer space. The spectral
distributions of the absorbed solar and emitted terrestrial electromagnetic waves are
very different (the solar flux peaks in the visible light spectrum, while the emitted
flux peaks in the infrared).
At steady-state, the rate that energy is absorbed by the Earth (from the sun)
equals that emitted by the Earth (to space):

qsolar absorbed ¼ qemitted

 
αI 0 Ashadow ¼ εσAsurface T 4E  T 4space

where α is the net integrated absorptivity of the Earth (equal to approximately 0.9),
Io is the solar constant (1,400 W/m2), Ashadow ¼ πR2E is the shadow area the Earth
(RE ¼ 6.378 km is the Earth’s radius), Asurface ¼ 4πR2E is the surface area of the
Earth, ε is the net integrated emissivity of the Earth (approximately equal to the
absorptivity), TE is the temperature of the Earth in K, and Tspace is the temperature
of outer space (0 K). In this example, the use of the resistance coefficient hr is not
78 3 Heat Transfer Modes

300

295

α = 0.9
TEarth (K)

290

285
α = 0.8
280

275

270
0.7 0.75 0.8 0.85 0.9 0.95 1
Emissivity, ε

Fig. 3.12 Predicted Earth’s temperature as a function of radiation parameters

really needed because outer space is near absolute zero, and the equation can be
solved directly for the average temperature of the Earth:
 

0:25  
α I0 α 0:25
TE ¼ ¼ 280:1 K
ε 4σ ε

Note that this temperature is independent of the size of the spherical planet, and a
first approximation to other planets can easily be made using the appropriate solar
constant based on the planets distance from the sun.
The temperature of the Earth from this model is plotted in Fig. 3.12 as a function
of emissivity for two values of absorptivity (0.8 and 0.9). As emissivity increases,
the temperature decreases. As absorptivity increases, the temperature increases.
Notice that for α ¼ ε, the predicted temperature of the Earth is 280 K ¼ 7  C ¼ 44  F,
which is about right. This simplified model captures the controlling physics of a
rather complex problem!

Comment on Global Warming

This model yields insight to the much discussed and controversial greenhouse
effect, in which radiation emitted from the Earth’s surface is absorbed by gases in
the atmosphere. Light emitted from the Earth’s surface and absorbed by the
atmosphere would result in a decrease in the Earth’s effective emissivity, and result
in a higher temperature. The extent to which an apparent warming trend can be
attributed to the combustion of fossil fuels will likely remain controversial forever,
as climate is a fundamentally nonlinear chaotic process. Nevertheless, anything that
can be done to responsibly reduce our reliance on fossil fuels is a good idea.
3.5 Stefan–Boltzmann and Kirchhoff’s Radiation Laws 79

3.5.3 Case 3: Radiation Between Infinite Plates

Consider two plates that are placed close together so that the gap between them is
much less than the width and height, i.e., a double-paned window. There is thermal
radiation exchange between the two inner surfaces that enclose the gap. However,
these two surfaces that “see” each other do not fit the thermal radiation criterion of
an object completely surrounded by another, case 1. However, the following
derivation yields a simple formula for radiation between these surfaces that can
be cast into an Ohm’s Law form.
Photons are continually being emitted in random directions by both surfaces
(1 and 2) at a rate that is governed by the Stefan–Boltzmann Law. In a short period
of time (Δt), the photon energy that is emitted by surface 1 (the hotter surface, say)
per unit area is:

n1 ¼ ε1 σT 41 Δt

The photon energy simultaneously emitted by surface 2 is:

n2 ¼ ε2 σT 42 Δt

For the case of infinitely large plates, all of the photons emitted by 1 will strike
surface 2 and vice versa. That is, the fraction of photons emitted by surface 1 that
strike surface 2 is unity. One of the three things will happen to emitted photons.
A fraction (α2) will be absorbed, a fraction will be transmitted (τ2, i.e., the photon
passes through) and a fraction will be reflected (ρ2). Mathematically:

α2 þ τ2 þ ρ2 ¼ 1

Similarly, for photons that strike surface 1:

α1 þ τ1 þ ρ1 ¼ 1

There is no energy exchange between surface 1 and 2 associated with transmitted

photons. However, photons emitted by 1 that are reflected by 2 will return and strike
surface 1, their original source. In turn, some of these photons will be reflected
back to surface 2. Eventually, all the photons emitted by surface 1 will be either
absorbed by surface 2, absorbed by surface 1, or transmitted (in either direction).
Any photons that are emitted by 1 and ultimately absorbed by 2 results in an energy
exchange between surface (and vice versa), expressed as:

q1 to 2 ¼ n1 ðα2 þ α2 ρ1 ρ2 þ α2 ρ1 ρ2 ρ1 ρ2 þ α2 ρ1 ρ2 ρ1 ρ2 ρ1 ρ2 þ   Þ

The first term represents the photons emitted by 1 that are absorbed by 2. The
second term represents photons that reflect from 2, then reflect from 1 and are
80 3 Heat Transfer Modes

absorbed by 2 on the second pass. The third term represents those photons that
reflect again. In summation form:

X
1
E1 to 2 ¼ n1 α2 ðρ1 ρ2 Þi ¼ energy absorbed by surface 2 of photons emitted by 1:
i¼0

Simultaneously, the number of photons emitted by 2 and ultimately absorbed by 1 is

given by:

X
1
E2 to 1 ¼ n2 α1 ðρ1 ρ2 Þi ¼ energy absorbed by surface 1 of photons emitted by 2:
i¼0

Recognizing the binomial expansion (a Taylor series expansion), namely:

X
1
1
ðρ1 ρ2 Þi ¼
i¼0
1  ρ1 ρ2

The net energy exchanged is:

ð n1 α 2  n2 α 1 Þ
E12 ¼ E1 to 2  E2 to 1 ¼
1  ρ1 ρ2

The rate of energy exchange (per unit area) is therefore:

 
E12 ε1 α2 σT 41  ε2 α1 σT 42
q12 ¼ ¼
Δt 1  ρ1 ρ2

In formal radiation theory, Kirchhoff’s Radiation Law dictates that at a given

wavelength, the emissivity and absorptivity are equal (α1 ¼ ε1 and α2 ¼ ε2).
Therefore:

ε1 ε2  
q12 ¼ σ T 41  T 42
1  ρ1 ρ2

Note that if either surface has a low emissivity, the thermal radiation exchange
between surfaces is small. Therefore, in practice, only one surface need be coated
with a “low-e” coating to substantially reduce the radiation exchange across the gap
of a double-paned window.
Opaque surfaces are those which do not allow electromagnetic energy to pass
through them. That is:

τ¼0 and ρ ¼ 1  α ¼ 1  ε for opaque surfaces

For window glass, the transmittance of visible light is very high, by design.
However, the longer wavelength thermal radiation for nominal Earth’s ambient
3.5 Stefan–Boltzmann and Kirchhoff’s Radiation Laws 81

temperatures is strongly absorbed or reflected, and very little thermal radiation at

those wavelengths passes through. That is, window glass is generally opaque to
thermal radiation.

ε1 ε2  
q12 ¼ σ T 41  T 42
1  ð 1  ε1 Þ ð 1  ε2 Þ

ε1 ε2  
¼ σ T 41  T 42 for opaque surfaces
ε1 þ ε2  ε1 ε2

Expressed in an Ohm’s Law form (by expanding the temperature term twice):

q12 ¼ hrad ðT 1  T 2 Þ

where

ε1 ε2  
hrad ¼ σ T 21 þ T 22 ðT 1 þ T 2 Þ
1  ð 1  ε1 Þ ð 1  ε2 Þ

Expanding the denominator:

ε1 ε2  
hrad ¼ σ T 21 þ T 22 ðT 1 þ T 2 Þ
ε1 þ ε2  ε1 ε2

As before, the temperatures in the radiation coefficient are in absolute units

(Kelvin), while in Ohm’s Law form they can be expressed in Celsius. Note that
the radiation coefficient is zero if either surface has a zero emissivity (smooth
surfaces). A “low e coating” would only need to be applied to one inner surface to
minimize the radiation effect. Also, the coefficient in brackets evaluates to unity if
both emissivities are unity.

3.5.4 Radiation in the Single-Paned Window Problem

Both surfaces of the single-paned window can be treated as a completely

surrounded object. The exposed surface on the inside “sees” three walls and the
floor and ceiling of the room it is in.
Kirchhoff’s Law can therefore be expressed as:
 
qH, walls!1 ¼ ε1 σA1 T 2H, walls þ T 21 ðT H, walls þ T 1 ÞðT H, walls  T 1 Þ

Kirchhoff’s Law can be put into an Ohm’s Law form:

T H, walls  T 1
qH, walls!1 ¼ hr, 1 A1 ðT H, walls  T 1 Þ ¼
Rr , 1
82 3 Heat Transfer Modes

where the radiative coefficient for surface 1 is:

 
hr, 1 ¼ ε1 σ T 2H, walls þ T 21 ðT H, walls þ T 1 Þ

Similarly, for the outside window surface:

T 2  T C, walls
q2!C, walls ¼ hr, 2 A2 ðT 2  T C, walls Þ ¼
Rr , 2

where the radiative coefficient on the outside depends on the outside window
surface temperature (T2) and the outside walls (TC, walls) that the outer window
surface “sees”:
 
hr, 2 ¼ ε2 σ T 2C, walls þ T 22 ðT C, walls þ T 2 Þ

The radiative coefficient is a strong function of absolute temperature (and is

inherently positive). In the window example, the surface temperatures (T1 and
T2) are not known a priori. However, an iterative procedure can be set up to
determine it. That is, make a first approximation by “guessing” T1 and T2, then
evaluating for the heat transfer rates and then the surface temperatures. Then use
this updated “guess” for T1 in a second iteration and continue to iterate as needed. A
good first approximation that will demonstrate the sensitivity of the radiative
coefficient to temperature is to set T1 equal to the inside wall temperatures, and
T2 to the outside wall temperatures. This guess will overestimate hr,1 and underes-
timate hr,2. A value for the emissivity of 0.9 will be used on both surfaces.
 
hr, 1ðFIRST APPROXÞ ¼ ε1 σ T 2H, walls þ T 2H, walls ðT H, walls þ T H, walls Þ ¼ 4ε1 σT 3H, walls
 
hr, 1ðFIRST APPROXÞ ¼ 4ð0:9Þ 5:669ð10Þ8 ð25 þ 273Þ3

hr, 1ðFIRST APPROXÞ ¼ 5:40 W=m2 = C

Notice that the temperature unit for radiative coefficient can be either  C or K, since
it was defined from Newton’s Law of Cooling from a temperature difference,
whereas K must be used to evaluate hr. For the outside window surface:
 
hr, 2ðFIRST APPROXÞ ¼ ε2 σ T 2C, walls þ T 2C, walls ðT C, walls þ T C, walls Þ ¼ 4ε2 σT 3C, walls
 
hr, 2ðFIRST APPROXÞ ¼ 4ð0:9Þ 5:669ð10Þ8 ð0 þ 273Þ3

hr, 2ðFIRST APPROXÞ ¼ 4:15 W=m2 = C

Notice that both of these radiative coefficients are of the same order of magnitude as
the convection coefficients. That means that radiation is important and should not
be neglected in this problem. The corresponding radiative resistance values are:
3.6 Single-Paned Window: Results 83

 
1 1 1, 000 W 
R1, r ¼ ¼   ¼ 11:5 C
hr, 1 A ð5:4 W=m2  CÞð16 m2 Þ kW  kW
 
1 1 1, 000 W 
R2, r ¼ ¼   ¼ 15:1 C
hr, 2 A ð4:15 W=m2  CÞð16 m2 Þ kW  kW

This treatment of thermal radiation is a preliminary treatment and should be

considered a good means to establish if radiation is important. If radiation is
shown at this level of analysis to be dominant, a more thorough development of
radiation theory might be warranted.

3.6 Single-Paned Window: Results

With all the pieces in place, the resistance network, and its reduction to a single
equivalent resistance, is shown with resistance values in ( C/kW) in Fig. 3.13.

Fig. 3.13 Single-paned window resistance values (in  C/kW) and its reduction to a single
equivalent resistance. This calculation is the first iteration, in which the surface temperatures
were assumed in order to calculate the radiative resistances
84 3 Heat Transfer Modes

The window surface temperatures can be solved for using voltage divider
formulas. For example, the general formula derived in Chap. 1 for a node “i”
with thermal neighbors “j”:
X Tj
R
j, neighbors ij
Ti ¼ X
1
j, neighbors
Rij

Referring to the middle network of Fig. 3.13 and using a series resistance “jump”
across the immediate neighbor to a known temperature:

TH
=R1, eff þ TC=ðRk þR2, eff Þ =6:0 þ 0=ð0:78þ3:3Þ
25

T1 ¼ ¼ ¼ 10:1  C
=R1, eff þ 1=ðRk þR2, eff Þ
1
=6:0 þ 1=ð0:78þ3:3Þ
1

=ðR1, eff þRk Þ þ TC=ðR2, eff Þ 25=

TH
=
ð6:0þ0:78Þ þ ð3:3Þ
0

T2 ¼ 1 ¼ ¼ 8:18  C
=ðR1, eff þRk Þ þ 1=ðR2, eff Þ 1
=ð6:0þ0:78Þ þ 1=ð3:3Þ

Recall that the radiative resistances used a first approximation that the surface
temperatures were equal to the adjacent fluid temperatures. A second approxima-
tion (iteration) uses the temperatures just calculated to give an improved guess at
the surface temperatures, and therefore improved resistance estimates. This itera-
tive procedure is demonstrated in Fig. 3.14, which is a screenshot of a spreadsheet
designed for that purpose. The top block contains input parameters (which can be
simply changed in the spreadsheet to run different scenarios), followed by calcula-
tion of the fixed thermal resistances in the model developed here. Finally, the
iteration table shows the effect of improved “guesses” for this case study. Each
row represents a different iteration (index “k”). The two temperatures are the
“guesses” for the temperatures. The first iteration (k ¼ 0) is an input set equal to
the fluid temperatures (but any values could be manually placed). Then the radiative
coefficients are calculated based on these guesses, followed by the radiative
resistances, effective resistances (convection and radiation in parallel), and finally
the net equivalent resistance. For the next iteration, the two temperatures are
calculated based on the resistance values calculated from the previous iteration,
and therefore are improved “guesses.” It is apparent that only three iterations are
needed to converge to within three significant figures. However, in light of the
uncertainty in the conductive resistances (k value), and both convective resistances,
it can be argued that the first approximation is sufficient for this case.
This general procedure for performing iterative calculations arises frequently in
heat transfer applications, and will be returned too often, in different variations.
Finally, to answer the original question of the problem statement, the rate of heat
transfer through this window (of area 16 m2) is 2.436 kW. The total energy
3.6 Single-Paned Window: Results 85

INPUT PARAMETERS
w= 8 m, window width
H= 2 m, window height
Dx= 0.012 m, window thickness
TH = 25 oC, inside air and wall temperatures
TC= 0 oC, outside air and wall temperatures
k= 0.96 W/m/K, thermal conductivity of glass
h1= 5 W/m2/K, convection coefficient on inside
h2= 15 W/m2/K, convection coefficient on outside
e= 0.9 emissivity of glass
FIXED THERMAL RESISTANCES
Rk= 0.78125 oC/kW
Rh1= 12.5 oC/kW
Rh2= 4.166667 oC/kW

ITERATION TABLE
k T1 T2 h1,r h2,r R1,r R2,r R1,eff R2,eff Req q(kW)
0 25 0 5.40 4.15 11.57 15.05 6.01 3.26 10.05 2.487
1 10.05736 8.114674 5.01 4.34 12.48 14.40 6.25 3.23 10.26 2.437
2 9.779702 7.875645 5.00 4.34 12.50 14.42 6.25 3.23 10.26 2.436
3 9.776913 7.873849 5.00 4.34 12.50 14.42 6.25 3.23 10.26 2.436
4 9.776881 7.873826 5.00 4.34 12.50 14.42 6.25 3.23 10.26 2.436
5 9.776881 7.873826 5.00 4.34 12.50 14.42 6.25 3.23 10.26 2.436
6 9.776881 7.873826 5.00 4.34 12.50 14.42 6.25 3.23 10.26 2.436

Fig. 3.14 Worksheet designed for single-paned window with iteration procedure to calculate
radiative coefficients. The values for the first iteration (k) differ from the calculations in the text
because of round-off error in those hand calculations

transferred is the rate multiplied by the time, and the energy cost is the energy price
multiplied by the total energy transferred:


        
$20 2:436 kJ  GJ  103 J 3, 600 s 24 h
Energy Cost ¼    9      
GJ s 10 J kJ   h   day 
$4:21
¼
day

This energy cost estimate scales with the total surface area of window (a parallel
heat transfer concept). It may not seem like a lot, but it can add up over the course of
a year, and over a building that has many windows.
Engineering Significance: Notice that since the conductive resistance is much
smaller than the effective resistances on both sides of the window, increasing the
thickness of the window is not an effective means to reduce heating costs associated
with windows. On the other hand, introducing a second pane of glass with an air gap
86 3 Heat Transfer Modes

(or other gas) between them has the effect of introducing more convective/radiative
thermal resistances to the thermal channel. Indeed, double-paned windows very
effectively reduce the heat transfer rate. Workshop 3.1 develops simplified models
for analyzing double-paned windows.

Appendix. Conductive Thermal Resistances in Various

Geometries

Consider the cylindrical shell of inner radius r1, outer radius r2, and length L (into
the page) shown schematically in cross-section in Fig. 3.15. The inner surface is
maintained at a temperature T1 and the outer at T2. Conceptually, the shell can be
represented in rectangular coordinates by “cutting” the projection at a random
place, and straightening it out to reveal the projection shown of a wall of constant
thickness (Δr ¼ r2  r1), but with a cross-sectional area that varies from 2πr1L on
one side to 2πr2L on the other. The coordinate “r” on the original view will be
replaced by the coordinate “x” in the projection.
At steady-state, the heat transfer rate crossing the inner surface must equal the
rate crossing the outer surface. Furthermore, the heat transfer rate at any surface in
between must be equal. Invoking Fourier’s Law, in which the conductive heat
transfer rate in the x direction is proportional to the area normal (i.e., perpendicular)
to the heat flow and to the temperature gradient (with the thermal conductivity, k,
being the proportionality factor):

dT
qx ¼ kAðxÞ ¼ constant @SS
dx
The goal of the analysis is to express the conduction between the two surfaces in
terms of an Ohm’s Law form:

Fig. 3.15 Schematic

of a cylindrical shell
and its projection as a
one-dimensional
rectangular section
Appendix. Conductive Thermal Resistances in Various Geometries 87

T1  T2
q12 ¼
R12
In general, the thermal conductivity is a mild function of temperature. Separating
the variables (T and x) in the Fourier’s Law expression:

dx
kdT ¼ qx
AðxÞ

Integrating from inner to outer surface (where the position goes from x1 to x2 and
the temperature from T1 to T2, and noting that qx is a constant):

Tð2 xð2
dx
kdT ¼ qx
AðxÞ
T1 x1

Defining an average thermal conductivity on the left hand side:

Tð2

kdT  kðT 2  T 1 Þ
T1

The heat transfer rate can therefore be expressed as:

kðT 2  T 1 Þ T 1  T 2
qx ¼ q12 ¼ xð2 
R12
dx
AðxÞ
x1

A general form for the conductive thermal resistance is therefore:

xð2
1 dx
R12 ¼
k A ð xÞ
x1

For well-defined geometries (i.e., Cartesian, cylindrical, spherical coordinates), this

integration can be carried out directly. Other geometries in which the appropriate
area can be expressed in terms of a single coordinate are possible as well. For
geometries that are not well defined, an approximation for conductive resistance
between two surfaces follows as:

ðaverage distance between surfacesÞ

R12   
k ðaverage area of surfacesÞ
88 3 Heat Transfer Modes

Applied to Cylindrical Coordinates

For the cylindrical shell:

xð2 ðr2

1 dx 1 dr 1 r2
R12 ¼ ¼ ¼ ln
k A ð xÞ k 2πrL 2πkL r1
x1 r1

If the approximate resistance value is expressed in this geometry:

ðaverage distance between surfacesÞ r2  r1 r2  r1

R12    ¼ 2πr1 Lþ2πr2 L ¼
k ðaverage area of surfacesÞ k 2 πkLðr 1 þ r 2 Þ

The ratio of the exact to approximate resistance values is:

R12, EXACT ð1 þ r2 =r1 Þlnðr2 =r1 Þ

¼
R12, APPROX 2ð r 2 = r 1  1Þ

The latter form (in terms of the ratio of radii) is obtained by dividing numerator and
denominator by r1. A plot of this ratio is shown in Fig. 3.16. It is clear that the
approximate value underestimates but approaches the exact value when the ratio of
radii approaches unity, the thin shell approximation. However, the error is less than
10 % up to r2/r1 ratios of 3.

1.5
Ratio of Exact to Approximate Thermal

1 r2 ln r2
Resistances: Cylindrical Geometry

R12,EXACT r1 r1
1.4
R12,APPROX r
2 2 1
r1
1.3

1.2

1.1

1
1 2 3 4 5 6 7 8 9 10
r2/r1

Fig. 3.16 Comparison of exact thermal resistance to approximate thermal resistance for a
cylindrical shell
Workshop 3.1. Double-Paned Window Models 89

Exercises: Derive the exact conductive thermal resistance for

• A spherical shell with inner radius r1 and outer radius r2 (see how the floor
conductive resistance is treated in Appendix 1 of Chapter 4. . .)
• A rectangular duct of aspect ratio (width to height) AR, length L, with inner
width w1, and thickness Δx
• A truncated cone with smaller area A1, larger area A2, and a distance L between
them (with insulated sides). . .

Workshop 3.1. Double-Paned Window Models

In an attempt to reduce the heat loss across a window, the single-paned window
from the opening example of this chapter is replaced with a double-paned window
in which a second pane of glass is added, with a gap between the plates (g). Estimate
the rate of heat loss and the energy cost as a function of gap width, and compare to
the single-paned window result. Compare the effects of typical window glass (with
an emissivity of 0.93 on both surfaces) and the performance with one of the inner
surfaces coated with a “low-e coating” (emissivity of, say 0.3). Also, compare the
replacement of air with argon.
It is a good idea to stop reading and think about this problem first and draw your
own thermal resistance network before going on.

Model Development
Heat transfer across the window is still a single thermal channel, but inclusion of a
second pane with a gap introduces additional thermal resistances. There is radiation
exchange between the two gaps (modeled as case 3: radiation between infinite
plates), and there is a parallel heat exchange mechanism by conduction and/or
convection. Two models for the latter are considered at this point: narrow gap and
wide gap.

Narrow Gap Model

In this model, the air (or other gas, like Argon) that is trapped between the two
panes of glass is assumed to be stationary. Therefore, it behaves like a solid and heat
exchange between the inner surfaces takes place through a straight conduction
mode. However, radiation between the two inner plate surfaces acts in parallel.
A thermal resistance network (at steady-state) for this model is shown in Fig. 3.17.
There are two parallel channels across the gap, a conductive resistance and a
radiative resistance.
The gap resistance is given by:

g
Rk, gap ¼
kgas A
90 3 Heat Transfer Modes

Fig. 3.17 Thermal resistance network for a double-paned window with the narrow gap model

The total heat transfer rate is:

TH  TC
q12 ¼ ¼ UAðT H  T C Þ
Req

where the equivalent resistance is:

1 1 1
Req ¼ þ Rk þ 1 þ Rk þ 1
1
Rr , 1 þ R h, 1
1
Rrad, gap þ Rk, gap
1
R r , 2 þ R h, 2
1

In terms of equivalent resistances where parallel channels exist:

Req ¼ RH, equiv þ Rk þ Rgap, equiv þ Rk þ RC, equiv

The equivalent gap resistance can be expressed as:

!
Rk, gap g 1
Rgap, eq ¼ Rk, gap
¼ hrad, gap g
Rrad, gap þ1 kair A
kair þ1

The equivalent gap resistance increases with the gap width. Conceptually, the gap
acts like an insulating layer of material added, and since the thermal conductivity of
air (0.023 W/m/K) is much lower than that of glass (0.96 W/m/K), this added layer
is more effective than simply making the glass thicker. However, the radiation
channel across the gap reduces the effective insulating property (notice the dimen-
sionless parameter in the denominator). Given an effective radiation coefficient of
hrad,gap ¼ 6 W/m2/K, from experience, and kair ¼ 0.023 W/m/K, the value of the gap
width where the conduction and radiative resistances are equal (setting the dimen-
sionless parameter to unity) is g ¼ kair/hrad,gap ¼ 0.0038 m ¼ 3.8 mm. That is, for
gaps below this, radiation is dominant, and conduction dominates for wider gaps.
Since manufactured double-paned windows set the gap spacing to be typically
10–20 mm, this comparison suggests, tentatively, that radiation effects are minor
in comparison to straight conduction for actual windows. However, the assumption
that air in the gap does not move must be questioned. The air in the gap does not
remain stationary, and an additional convective-like mode of heat transfer occurs,
and described in the following wide gap model.
Workshop 3.1. Double-Paned Window Models 91

Fig. 3.18 Thermal resistance network for a double-paned window using the wide gap model

Wide Gap Model

Since the gap contains a fluid and there is a temperature difference between the two
panes of glass, the fluid near the colder pane is denser than the fluid near the hotter
plane. A buoyancy force will cause the hotter fluid to rise and the colder to sink, and
a natural circulation pattern will develop in the gap. Fluid friction opposes the
motion. In this introductory chapter, a conceptual model is used in which there is a
boundary layer adjacent to each inner surface that is small compared to the gap
width. Therefore, the bulk of the fluid in the middle of the gap is considered to be
stagnant, and a classic natural convection mode of heat transfer governed by
Newton’s Law of Cooling exists between each glass surface and the bulk fluid in
the gap, which is at an intermediate temperature to that of the two glass surfaces. A
thermal resistance network for this model is shown in Fig. 3.18.
In the wide gap model, heat flows from the hotter inner surface (Ta) into the bulk
of the gap (at Tgap) by a natural convection mode, and in series, from the bulk fluid
to the colder inner surface (Tb). The equivalent resistance of this channel is:

1 1 1
Req ¼ þ Rk þ 1 þ Rk þ 1
1
Rr , 1 þ Rh1, 1 Rrad, gap þ 1
RH, gap þRC, gap Rr, 2 þ 1
Rh, 2

As with the narrow model, the equivalent resistance is:

Req ¼ RH, equiv þ Rk þ Rgap, equiv þ Rk þ RC, equiv

The equivalent gap resistance can be expressed as:

 
RH, gap þ RC, gap
Rgap, eq ¼ 
RH, gap þRC, gap
Rrad, gap þ1

The gap resistance consists of two convective resistances in series, and a reducing
factor due to radiation. Notice that the size of the gap does not affect the prediction
of this model.
Conceptually, and in practice, the conductive resistance across the glass itself
was shown to be negligible for the single-paned window, so that the heat transfer
92 3 Heat Transfer Modes

rate is controlled by two convective/radiative equivalent resistances in series that

are of comparable magnitude. The second pane of glass introduces two additional
convective resistances that are of the same order, and therefore, the heat transfer
rate through the double-paned window should be approximately half that of the
single-paned window. Adding a third pane of glass (a triple-paned window)
introduces two more convective resistances, so the rate of heat transfer is approxi-
mately a third of that of a single-paned window. Notice the diminishing returns. A
second pane cuts the original heat transfer rate by 50 %, a third pane cuts it by only
an additional 17 %, a fourth pane would cut an additional 8 %, and so on.

Additional Workshop Ideas

Workshop 3.2. Heat Loss Through Walls

Compare the rate of heat loss per unit area (report results as “R-Value” in  C m2/W)
across an exterior wall constructed with 2 4 studs, spaced by 16 inches with that
of 2 6 studs, spaced by 24 inches. Consider the inner surface to be finished with
sheetrock, and the outer surface to be finished with ¼00 plywood, a vapor barrier,
and siding. Consider the cases with and without insulation in the gaps. Note that the
actual dimensions of a 2 4 are not 2 4 in.

Workshop 3.3. “Critical Radius of Insulation”

The outer surface of a 100 m long electrical wire of radius rwire ¼ 0.005 m is not to
exceed T1 ¼ 100  C (it doesn’t matter what the wire material is, just what its outer
temperature is). The wire is exposed to a fluid at Tinf ¼ 0  C with an effective
convective/radiative coefficient of 15 W/m2/K. Calculate the steady-state rate of
heat loss (from which maximum amperage could be determined) for a bare wire.
Next, a layer of insulation (with thermal conductivity kins ¼ 0.15 W/m/K) is to be
added. What insulation thickness (Δx) will maximize the heat transfer rate between
the pipe and ambient (hence amperage)? What is the corresponding heat transfer
rate and outer surface temperature of the insulation? Wait a minute. . . Isn’t insula-
tion supposed to REDUCE the heat transfer rate? What’s up with that? Use
resistance networks (and exact formula for conductive resistance across insulation
layer).
Workshop 3.1. Double-Paned Window Models 93

Workshop 3.4. “Wearing Short Pants”

Compare the steady-state rate of heat loss from your legs for a hot summer day with
long pants versus short pants. For both scenarios, draw a thermal resistance network
(omit capacitances since steady-state values are specified.). The skin can be set at a
fixed temperature of, say 32  C (a value which can vary due to thermoregulatory
effects). For the case of short pants, the skin is exposed to a fluid at one ambient
temperature and as an enclosed surface in radiation thermal contact with walls that
could be at a different temperature than the fluid. For example, in summer, the
exterior wall of the room could be at a higher temperature than the room, and in
winter, the wall could be cold. For the case of long pants, there is an air gap between
the skin and the pants, which can be modeled as two natural convection resistances
(skin to air, air to pants) and a radiation between infinite planes (skin to pants).
Conduction occurs across the pants and radiation/convection on the outside. What if
it were a sunny day? Compare black and white pants in that case (note that due to
very different wavelengths, the absorptivity of solar radiation is higher for black
pants than white pants, but the emissivity at the Earth’s temperatures can be about
the same and a value near unity).

Workshop 3.5. “Temperature of the Planets”

Estimate the average temperature of all the planets, assuming first that the effective
absorptivity of solar insolation equals the effective thermal emissivity of thermal
radiation. Obtain the average distance from the sun from a published or online
source. Compare the published temperature values. Then use that data to determine
the effective ratio of solar absorptivity to emissivity for each planet. For this
problem, the solar flux (intensity times area) leaving the sun is the same exiting
any spherical shell enclosing the sun. That is, I ð4πr 2 Þ ¼ Constant, where “I” is the
solar intensity (¼1,400 W/m2 for the Earth) and “r” is the distance from the sun
center to the distance of the spherical shell (¼1.49(10)11 m for the Earth).
 Part II
Transient Conduction
1-Node Transient Models
4

A basic philosophy of this book is to develop a modeling approach that considers

“the big picture” first, and then more detail only if needed. Since heat transfer rates
are driven by temperature differences, the first step in analyzing heat transfer
problems is usually to solve for the temperature field, which depends on spatial
position and time. Numerical analyses involve the discretization of space (breaking
the physical region into discrete subelements) and time (taking finite steps forward
in time).
This chapter considers “1-node” models, where the region of space in question is
lumped into a single node. Time dependence often results in easily solved ordinary
differential equations, or ones that can be easily solved numerically. By treating a
defined system as a single object with average properties, that is, as a single node, a
first approximation can often be made that is either sufficient to make design
decisions, or directs attention to intelligently develop a more detailed approach.
Furthermore, key functional relationships and associated insights are obtained with
relatively little mathematical complexity.
The so-called exact solutions involve breaking the spatial region of influence
into an infinite number of pieces (by making them infinitesimally small) and by
stepping forward in time by an infinitesimal amount. There is certainly mathemati-
cal elegance and historical relevance in the exact solutions of these partial differ-
ential equations, but the philosophy implied in solving problems that way is the
opposite of that adopted here. Furthermore, what is often lost in translation is that
the “exact solution” applies to a problem that has gone through an extensive set of
approximations to the physical problem to make it “fit” a model that has a closed-
form mathematical solution. There is a certain false sense of security in how
“exact” those models are to real physical problems.
The first section considers the cooling mug of coffee as a lumped element. Next,
the heating or cooling of a general homogeneous object is investigated in a general
way. Firstly, the straightforward case is considered in which the size of the node is
constant, but its temperature changes. Secondly, several cases are considered in
which the size of the node changes with time. This class of problems serves to

# Springer International Publishing Switzerland 2015 97

G. Sidebotham, Heat Transfer Modeling: An Inductive Approach,
DOI 10.1007/978-3-319-14514-3_4
98 4 1-Node Transient Models

introduce important concepts of thermal boundary layers and thermal lengths that
will be useful in subsequent chapters, particularly in gaining a fundamental
understanding of boundary layer phenomenon in fluids that form the basis for
estimation of convection heat transfer coefficients.
The applications developed in this chapter correspond to cases that are devel-
oped in traditional textbooks, with exact solutions. The methods developed here can
be extended to a much broader set of problems.

4.1 Lumped Capacity Model for Coffee/Mug Problem

In this section, the lumped capacity model is applied to the coffee/mug problem. All
of the capacitance of the system (mug plus coffee) is lumped into a single capacitor.
A thermal resistance network of multiple channels (top, side, and bottom) is
developed to model parallel paths for heat to flow from the coffee interior to the
ambient. A first model includes all convection and conduction modes (Model 1).
Then effects of radiation on the outer surfaces are included (Model 2). Finally, the
effect of evaporation at the coffee-free surface is included (Model 3). In all these
models, the resistance values are treated as constants, suitably chosen to represent
an average value during the cooling event. Use of constant resistances results in a
first-order system, and the entire event is characterized by a single overall heat
transfer coefficient (U) and a time constant (RC).

Engineering Problem Statement (Repeated)

A volume V ¼ 0.41 L of coffee at 1 atm, 92  C is quickly poured into a room
temperature (21  C) ceramic mug (shown schematically in Fig. 4.1) with a mass of
mmug ¼ 0.365 kg, inner diameter D ¼ 0.083 m, inner height H ¼ 0.098 m, outer
height Ho ¼ 0.109 m, and nominal wall thickness w ¼ 0.0040 m. The mug of coffee
is placed on a table in a still room. Estimate the time it takes for the water to cool to
60  C (above which it is too hot) and to 45  C (defined as “tepid”).

4.1.1 Model 1: Neglect Radiation and Evaporation

Figure 4.2 shows a resistance network that breaks down the total heat flux from the
hot coffee (at T, representing an average temperature of the interior of the coffee) to
ambient (at T1) into three parallel channels (top, side and bottom), each of which
consists of several resistances in series. Each individual resistance represents a
mode of heat transfer between one identifiable surface or fluid and another, with
nodes to identify each surface. Appendix 1 details the calculation of the individual
resistance values.
In this model, all the capacitance of the system (coffee plus mug) has been
lumped into a single node, and therefore the smaller filled nodal temperatures can
be related to the main capacity-carrying node, T, and the known ambient tempera-
ture through the methods developed in Chap. 1. To reiterate, open symbols are
4.1 Lumped Capacity Model for Coffee/Mug Problem 99

Fig. 4.1 Schematic diagram

of coffee mug with geometry
defined

Top View

D W
L

V H
coffee H0

mug

Side View

reserved for nodes for which capacitance has been attributed, and therefore their
temperature is governed by a differential equation, not an algebraic relationship
between neighbors. At times, big filled symbols (rather than a battery symbol) will
be used to represent known temperatures. Small filled symbols represent unknown
intermediate temperatures of nodes that have no capacitance assigned to them.
The results of this model will be compared to the experimental data from
Chap. 2. It is noteworthy that the experimental method (an IR detector) gave direct
measurements of the outer surface of the side wall (Tos) and of the free surface (Ts),
not of the bulk average.
For the side channel, heat must first be transferred from the interior of the liquid to
the inner surface of the mug side wall. The resistor Rcs (for coffee/side) represents
that convection heat transfer mode. Next, the heat must cross the side wall through
the mug, from the inner surface to the outer surface, and that conduction mode is
represented by the resistor Rws (wall/side). Finally, heat must flow from the outer wall
to the ambient fluid, and that is a convective mode represented by Ras (ambient/side).
These three thermal resistances act in series, and therefore the total resistance of the
side channel is their algebraic sum. Once numerical values are estimated, judgments
about their relative importance can be made.
For the bottom channel, heat must flow from the interior of the coffee to the inner
surface of the bottom, or floor through a convective resistance (Rcb for coffee/
bottom). Heat must then cross through the mug wall through a conductive resis-
tance (Rwb for wall/bottom). Heat flows from the bottom of the mug across an air
gap then into the floor through a conductive path (Rfb for floor/bottom).
100 4 1-Node Transient Models

Fig. 4.2 Lumped capacity T∞

thermal resistance model
superimposed on a schematic
of a mug of coffee (top) and at
then a stepwise breakdown to
Ts
a single equivalent thermal
ct
resistance cs ws as T∞
T
C Tis Tos
cb
Tib
wb
Tob
fb

T∞

ct at
T(t) top
cs ws as T∞
side
cb wb fb
C bottom

top
T(t)
side T∞

bottom
C

T(t)
R
T∞
C

For the top channel, the free surface of the coffee does not rigidly fit into the
modes defined in Chap. 3. Convection was defined as heat transfer between a fluid
and a solid surface. However, the coffee-free surface is a boundary between two
fluids (air and coffee). Heat flows from the interior to the surface and then from the
surface to the ambient above it, and so two resistances in series that are like
convection resistances are introduced for the top channel (Rct for coffee/top and
Rat for air/top).
4.1 Lumped Capacity Model for Coffee/Mug Problem 101

Fig. 4.3 Spreadsheet implementation of extended lumped capacity Model 1: neglect radiation
and evaporation

Figures 4.3 and 4.4 show results from a spreadsheet implementation of the
model. The block of input quantities represents input values. Several of these inputs
have been calculated in other spreadsheets (and future chapters) and manually
input into this spreadsheet. The values chosen for the convection coefficients
(6.8 W/m2/ C for air/surface and 470 for coffee/surface locations) were determined
in a separate spreadsheet (detailed in a future chapter) as a suitable average value
for the entire cooling event.
All quantities in this block are fixed constants. More sophisticated models might
allow for variable coefficients, but then a numerical simulation of the event would
be required. On the other hand, the assumption of constant heat transfer coefficients
(suitably chosen averages) allows for a closed-form solution to the lumped capacity
model, with a net determined time constant and an overall effective heat transfer
coefficient (RC and U, shown in the results block). Figure 4.4 applies to
Model 3 discussed shortly (which includes radiation and evaporation effects).
The effects of radiation and evaporation are “turned off” in Fig. 4.3 by setting the
corresponding “h” values to a small number (a “zero” cannot be used, as the model
divides by this value), resulting in effectively infinite resistance values.
102 4 1-Node Transient Models

Fig. 4.4 Derived parameters used in Models 1, 2, and 3, with the following exceptions: for Model
1, the thermal resistances Rat,evap and Rat,rad evaluate to large (effectively infinite) values by setting
the corresponding heat transfer coefficients to low values (they can’t be zero, since the model
divides through by them). In Model 2, the thermal resistance Rat,evap is large

The plot of temperature vs. time in Fig. 4.3 shows that the predicted cooling is
noticeably slower than the experimental result. Recall that an experimental value of
U was banded between approximately 10 and 20 W/m2/K (the experimental result
being not a pure first-order response). The predicted overall U value of 6.0 W/m2/K
is below that range.
The table of channel resistances reveals more about the controlling routes for
heat to flow from the coffee to ambient. The net channel resistance is much lower
through the side wall than through the bottom channel, which is lower than the top
channel. This effect is largely due to the surface area (recall the side surface area is
more than half the total exposed area). Within the side channel, the resistance is
dominated by the outer equivalent resistance. One result of this effect is that the
outer surface temperature is much closer to the coffee temperature than ambient, as
confirmed experimentally.
4.1 Lumped Capacity Model for Coffee/Mug Problem 103

4.1.2 Model 2: Add Thermal Radiation

Radiation effects have not been included in the model of Fig. 4.2. The alternative
model of Fig. 4.5 adds radiation effects. Here, the mug of coffee sits in a room
whose walls and ceiling (the big outer box) may, in general, be at a different
temperature (T1w) than the air in the room. The exposed surfaces on the sides
and top of the coffee/mug system are in direct contact with the air surrounding them
(and modeled as convection), but in addition, these exposed surfaces “see” the walls
of the room it is in. The physical mechanism of this radiation heat transfer is
fundamentally different than the convection. Notice that now there are two
“ambients,” fluid and radiation wall.
Figure 4.6 shows this resistance network recast. At the top and side surfaces
exposed to air, the channels split, or branch. Consider, for example, the side
channel. In order for heat to flow from the coffee ultimately to ambient, it must
first flow from the coffee to the submerged portion of the inner surface of the mug.
From there, heat flows across the mug wall to its outer surface. The outer surface
branches into two channels, one to the fluid immediately adjacent to it (at T1) and
the other to the walls of the room (at T1w). Notice that the temperature at the mug
outer surface is represented by an open symbol. The temperature there cannot be
related to two other known temperatures by a simple voltage divider calculation
(recall Chap. 1). This network cannot be modeled as a simple RC network because
of the branching.

Fig. 4.5 Resistance network superimposed on a mug of coffee schematic that includes radiation
effects
104 4 1-Node Transient Models

Fig. 4.6 Another thermal T∞w

resistance network for a mug
of coffee

top
T(t)
side T∞
bottom
C
R

Fig. 4.7 Resistance network top

when the walls of the room T(t)
side T∞ = T∞w
are at the same temperature as
the air within it bottom
C
R

Figure 4.7 shows the resistance network for the case when the walls of the room
and the air within it are assumed equal. The convective and radiative resistances act
in parallel on the top and bottom surfaces, and the entire network can be reduced to
a single R value, so that a simple RC circuit model can be used. When radiation and
convection act in parallel (to the same ambient temperature), the net heat transfer
coefficient is the sum of the two modes: hequiv ¼ hconvection þ hradiation .
Figure 4.8 shows the spreadsheet results when these radiation effects are
included. The value chosen for the radiation coefficient (6.6 W/m2/ C) was deter-
mined in a separate spreadsheet (detailed in a future chapter) as a suitable average
value for the entire cooling event. The model now shows good agreement between
theory and experiment, with perhaps a slightly slower initial cooling rate. The outer
equivalent resistance value is approximately half of what it was for the model that
neglected radiation.

4.1.3 Model 3: Add Evaporation Effects

Figure 4.9 shows the resistance network with evaporation effects included as a third
parallel channel between the free surface of the coffee and ambient. Figure 4.10
shows the spreadsheet result when this effect is included. The energy exchange
associated with the evaporation is modeled with an equivalent convection coeffi-
cient (hevap). While this energy exchange is by a mass heat transfer mechanism, not
a heat transfer mechanism, it can be expressed as being proportional to the temper-
ature difference (with the coefficient having a strong temperature dependence). The
selection of a specific constant value requires judgment, and a formal way of doing
that is the subject of Chapter 13. In the present case, introducing an approximate
4.1 Lumped Capacity Model for Coffee/Mug Problem 105

Fig. 4.8 Model 2: add radiation effects

at,rad
at,evap
at,conv top
T(t)
ct as,rad T∞ = T∞w
side
cs ws as,conv
C bottom
cb wb fb

Fig. 4.9 Resistance network with evaporation included. A valuable exercise is to write in pencil
the values from Fig. 4.4 directly onto this sketch, visualizing the heat transfer being represented

evaporative channel increases the overall heat transfer coefficient from 10.9 to
11.9 W/m2  C, a 9.1 % increase. That is considered to be significant, but not
dominant. The effect of radiation is much more important, especially since radia-
tion acts in parallel with convection on the side wall, with its much larger surface
area than the free surface.
Figure 4.11 shows results (from an analysis in Chapter 13) of the effect of
temperature on the three heat transfer coefficients between the coffee-free surface
106 4 1-Node Transient Models

Fig. 4.10 Model 3: add evaporation effects

30
Heat Transfer Coefficient (W/m2/K)

25

20

15 Sum

10 Radiative

5
Convective Evaporative
0
25 35 45 55 65 75 85 95

Coffee Temperature (oC)

Fig. 4.11 Results of a more detailed estimate of the heat transfer coefficients on the coffee-free
surface. The ambient temperature is 25  C
4.2 Heisler and Gurney–Lurie Charts from a 1-Node Lumped Capacity Analysis 107

and ambient. For natural convection, the convective heat transfer coefficient is a
function of the temperature difference between the solid and air surface. It is
actually zero when the coffee and ambient are at the same temperature, increases
sharply with temperature difference, then gradually increases. The radiative coeffi-
cient is a function of both coffee and ambient temperatures, and does not show a
sharp drop as the coffee temperature approaches ambient. Radiation becomes the
dominant mode of heat transfer as the coffee approaches thermal equilibrium with
ambient.
The evaporative term increases sharply as the coffee temperature approaches the
air temperature, and is an artifact of the use of the resistance model. In fact,
modeling evaporation as a dependent current source makes more sense in this
case. However, for coffee temperatures above 27  C in this case, the evaporation
coefficient is generally lower than both convection and radiation until approxi-
mately 90  C, above which the evaporation effect increases dramatically. This real
effect is due to the rapid increase in vapor pressure as the boiling point is
approached from below. The main point here is to argue that evaporative effects
are considered secondary to radiation effects, unless the coffee is within a few
degrees of its boiling point. As with radiation, for applications in which evaporative
effects dominate, other resources can be consulted.

4.2 Heisler and Gurney–Lurie Charts from a 1-Node

Lumped Capacity Analysis

In traditional heat transfer textbooks, exact solutions to transient problems of

homogeneous objects (i.e., uniform thermal properties of density, specific heat,
and thermal conductivity) that have a spatially one dimensional geometry (rectan-
gular, cylindrical, or spherical coordinates) are presented in generalized plots of
center temperature vs. time, temperature vs. position, and total heat transferred
vs. time. These problems (referred to as either Heisler or Gurney–Lurie charts) are
classified as “conduction with convection boundary conditions” and they are all
basic heating (or cooling) events of finite-sized objects at an initial temperature
suddenly placed in a fluid environment at a different temperature.
The lumped capacity model approach developed in Chaps. 1, 2, and 3 provides a
good first approximation to this class of problems and is formally developed here.
Furthermore, the two key dimensionless parameters, Biot number and Fourier
number, naturally emerge. The Biot number gives information about the spatial
distribution of temperature, that is, whether the object heats or cools uniformly (low
Biot number), or whether the surface temperature approaches the environment
temperature while the center lags behind (high Biot number). The Fourier number
is a dimensionless time, where real time is normalized by a characteristic time it
would take the interior of the object to adjust if its surface temperature were to
suddenly be changed (i.e., a high Biot number event). These variables are devel-
oped naturally from a generalized lumped capacity analysis and developed here for
a specific manufacturing-type application.
108 4 1-Node Transient Models

4.2.1 Example: Quenching a Solid Rod

In a manufacturing process, a stainless steel rod (k ¼ 16 W/m/ C, ρ ¼ 7,820 kg/m3,

c ¼ 490 J/kg/ C) of length L (into the page) and diameter D at an initial temperature
of Tinitial ¼ 300  C is placed in a quenching oil bath at Tfluid ¼ 30  C in order to
harden and cool it (Fig. 4.12). Using a lumped capacity model and assuming a
uniform convection coefficient on the surface of h ¼ 300 W/m2/ C and negligible
radiation determine the time required to cool the bulk of the material to 100  C and
the surface temperature at that time for diameters of 1 mm, 1 cm, 10 cm, and 1 m.

Model Development
A general analysis will first be performed, and then applied to the specific case in
question. A thermal RC network superimposed on the sketch lumps the entire heat
capacity into a single node. A conductive resistance between the bulk interior of the
rod and its surface acts in series with a convective resistance between the surface
and the surrounding oil. Note that the capacitive-carrying node (open symbol) is not
placed at the geometric center for conceptual reasons, as the numerator of the
conductive resistance is an average distance between the interior and the surface. A
convective resistance between the surface of the object and the surrounding fluid
takes places at a well-defined surface. If radiation is important, it acts in parallel
with convection, and can be combined into a single convective/radiative resistance
(if the fluid and radiation wall temperatures can be considered to be equal).
For constant resistance and capacitances, the solution for the (dimensionless)
temperature as a function of time is:

T  T fluid t
θ ¼ eðRk þRh ÞC
T initial  T fluid

This relationship can be used for any object of any shape and construction for which
estimates can be made of the total capacitance (the object need not be uniform), a
conductive resistance between the interior and the surface and a convective resis-
tance at the surface. Conceptually, the temperature represents an average tempera-
ture of the object, and in general, it is recognized that the temperature is not uniform
in the object.

T(t)
T∞
+
Rk Rh
Fig. 4.12 Schematic and C
superimposed RC network for
a solid rod undergoing a
cooling or heating event
4.2 Heisler and Gurney–Lurie Charts from a 1-Node Lumped Capacity Analysis 109

At any time, the surface temperature is related to the average temperature by a

voltage divider relationship:

T  T fluid T surface  T fluid

¼
Rk þ R h Rh
Rearranging:

T surface  T fluid Rh 1 1
¼ ¼ ¼
T  T fluid Rk þ Rh Rk =Rh þ 1 Bi þ 1

where the Biot number is introduced as the ratio of the conductive to convective
resistance. This relationship is independent of time. Rearranging for the surface
temperature:

T  T fluid
T surface ¼ T fluid þ
Bi þ 1
The convective resistance is approximated as:

1 1
Rh ¼ ¼
ðConvection Coef:ÞðSurface Area exposed to fluidÞ hAs

The conductive resistance is approximated as:

Ave: Distance bet: Capacitive Node and Surface D

Rk ¼ ¼
ðThermal ConductivityÞðAve: Area bet: Capacitive Node and SurfaceÞ kA

There is more uncertainty in precisely defining the conductive resistance compared

to the convective, but a reasonable first approximation of very complicated cases
can usually be made.
The Biot number can therefore be expressed as:



Rk hD As
Bi ¼ ¼
Rh k A
Note that the Biot number is proportional to the average distance between the
interior node and surface. For a given material in a given convective environment,
the larger the object, the larger the Biot number.
Consider the limiting cases. For large Biot number (Bi  1), in which case it is
more difficult for heat to flow by conduction through the material than it is for heat to
flow from the surface to the adjacent fluid, the surface temperature approaches the
fluid temperature. For the case of small Biot number (Bi  1), the surface tempera-
ture equals the average temperature. That is, the temperature in the object is uniform.
Heat flows easily from the interior to the surface, and is “hung up” at the surface.
110 4 1-Node Transient Models

This analysis demonstrates that the Biot number yields insight into the spatial
temperature distribution in a solid. All else being equal, small objects tend to heat
or cool uniformly, while large objects tend to have larger thermal gradients
within them.
For a homogeneous object, the total capacitance is C ¼ ρcV, where V is the total
volume of the object and the average temperature can be expressed as:

 t   
T  T fluid ρcVRk
R
1þ h = Rk ðBiþ1
Bi
Þ kAt

θ¼ ¼e ¼e ρcVD

T initial  T fluid

θ ¼ eðBiþ1ÞFo
Bi

where the Fourier number (Fo) has been introduced as a dimensionless time:

kAt αt
Fo ¼
ρcVD AD
V

The Fourier number is a measure of how far along a cooling or heating event an
object is. For short time (Fo  1), the dimensionless temperature is close to unity,
or the average temperature is close to the initial temperature. For long time
(Fo  1), the average temperature approaches ambient.
The effect of the shape of the object is manifested in the Biot and Fourier
numbers. There is a certain degree of latitude in precisely defining the average
conduction distance and average conduction area, consistent with lumping the
entire solid into a single node. The Heisler and Gurney–Lurie charts are developed
for the three classic coordinate frames that allow for exact theoretical analysis,
namely, Cartesian (wall), cylindrical and spherical. Interpretations of these cases in
which geometric averages are used are summarized in Fig. 4.13. Notice that the
Cartesian geometry (Fig. 4.14) consists of a wall (surface area of width  height)
subjected to the same fluid environment on both sides. The centerline acts as an
insulated boundary, and only half of the wall needs be considered. In the detailed
solution to Workshop 4.1, several different interpretations of the conductive resis-
tance for the cylindrical case are developed.
This dimensionless temperature (θ) vs. dimensionless time (Fourier number) is
plotted in Fig. 4.15 for fixed values of the Biot number. The Heisler and Gurney–
Lurie charts contain plots of the centerline temperature in this form. The spatial
variation is displayed as the dimensionless surface temperature vs. the inverse Biot
number (Fig. 4.16). The Heisler and Gurney–Lurie charts are of similar form, but
show the temperature at various locations relative to the centerline. In this analysis,
only the surface temperature relative to the average temperature is directly avail-
able. Note that these “cooling curves” become independent of Biot number for large
4.2 Heisler and Gurney–Lurie Charts from a 1-Node Lumped Capacity Analysis 111

Fig. 4.13 Biot and Fourier numbers for classic geometries. Note that α = k/ρ/c

L L
T∞ T∞
Ts T(t) T(t) Ts
Rh Rk Rk Rh
Centerline

C C

Fluid Solid Solid Fluid

L
T(t) Ts T∞
Centerline

Rk Rh
C

Solid Fluid

Fig. 4.14 RC network of a wall subjected to the same fluid on both sides. Top network shows a node
on each side, and could be used as a 2-node model (next chapter) if the fluid temperatures were
different on either side. For the symmetric case, the centerline acts as an insulated boundary due to
symmetry and the equivalent RC network (bottom) is considered in this section. The bottom schematic
applies directly to cylinders and spheres, where the centerline is a point (of zero surface area)

Biot number, in which case the surface temperature is close to the fluid temperature,
and therefore the cooling rate is controlled by the rate of conduction to the interior.
Low Biot number (high conductive, relative to convection) events result in uniform
temperatures within the solid, and the cooling or heating is controlled by the
convection.
112 4 1-Node Transient Models

1
0.9 Bi = 0.01
0.8
Dimensionless Temperature

0.7
0.6 Bi = 0.1
0.5
0.4
0.3
Bi = 1
0.2 Bi = 10
Bi = 100
0.1
0
0 2 4 6 8 10
Fourier Number (dimensionless time)

1
Bi = 0.01

Bi = 0.1
Dimensionless Temperature

0.1

Bi = 1
0.01

Bi = 100 Bi = 10
0.001

0.0001
0 2 4 6 8 10
Fourier Number (dimensionless time)

Fig. 4.15 Dimensionless temperature vs. Fourier number for various Biot numbers. Linear axes
(top) and semi-logy axes (bottom)

4.2.2 Application to Quenching a Solid Rod

For a solid rod of diameter D, length L, the volume is:


πD2 L
V¼ 4
4.2 Heisler and Gurney–Lurie Charts from a 1-Node Lumped Capacity Analysis 113

1
0.9
0.8
(Tsurface-Tfluid)/(Tave-Tfluid)

0.7
0.6
0.5
0.4
0.3
0.2
0.1
0
0.001 0.01 0.1 1 10 100 1000

1/Biot #

Fig. 4.16 Dimensionless surface temperature plotted against inverse Biot number

The surface area is

As ¼ πDL

The average distance (for the conduction distance) is taken to be half the radius
(quarter the diameter) and the average area is taken to be the geometric average
between the surface and the centerline:

D ¼ =4
D

1 πDL
A ¼ ðAs þ 0Þ ¼
2 2
A consideration of other conduction models is developed in the detailed solution to
the workshop presented at the end of this chapter.
The Biot and Fourier numbers, therefore, are:
!

Rk hD=4 2As
Bi ¼ ¼
Rh k As þ 0

hD
Bi ¼
2k
kAt ktπDL=2
Fo ¼ ¼
ρcVD ρc πD4 L D4
2
114 4 1-Node Transient Models

8αt
Fo ¼
D2
For the cooling event defined, the dimensionless temperature is specified as:

T  T fluid 100  30
θ¼ ¼ ¼ 0:259
T initial  T fluid 300  30

Solving the cooling expression for the Fourier number:

ðBi þ 1Þ
Fo ¼  lnθ
Bi
And for the time:



D2 Fo ρcD2 2k
t¼ ¼ ð1 þ BiÞlnθ
8α 8k hD



ρcD hD
t¼ 1þ lnθ
4h 2k

The cooling time increases linearly with diameter for low Biot numbers and
increases with the diameter squared for large Biot number events.
Results for the four diameter cases are summarized in Fig. 4.17. The cooling
time increases strongly with cylinder diameter. The temperature is nearly uniform
for small diameters, and large internal temperature gradients are apparent for large
diameters.

INPUT PARAMETERS
k 16 W/m/K
ρ 7820 kg/m^3
c 490 J/kg/K
h 300 W/m^2/K
Tinitial 300 oC
Tfluid 30 oC
Tfinal 100 oC
DERIVED PARAMETERS
α 4.18E-06 m^2/sec
θ 0.259
RESULTS
D (m) 0.001 0.01 0.1 1
Biot # 0.009375 0.09375 0.9375 9.375
Fourier # 145 15.7 2.79 1.49
cooling time (sec) 4.35 47.1 835 44,722
Surface temperature (oC) 99.3 94.0 66.1 36.7

Fig. 4.17 Spreadsheet solution for the quenching solid cylinder problem
4.3 Thermal Boundary Layers in Thick Walls 115

4.3 Thermal Boundary Layers in Thick Walls

In the 1-node fixed lumped capacity model considered in the previous section, the
capacitance encompassed the entire solid object. However, at the instant the fluid
temperature is changed, there is a region near the surface that responds immedi-
ately, while the interior is unaffected for a period of time. That is, there is a time
delay between when the thermal signal at the surface occurs and when the interior
responds. This section addresses that deficiency by introducing variable-sized
nodes. The heating or cooling of a finite-sized object subjected to a sudden change
at its surface can therefore be analyzed in two distinct phases: a thermally thick
phase (phase I) and a lumped heating (or cooling) phase (phase II).
This phenomenon introduces the notion of a thermal boundary layer or a
penetration layer. A thermal disturbance at the surface propagates, or penetrates,
into the solid. The size of this boundary layer region grows as time progresses.
A thermally thick wall is defined as one in which the physical thickness of the wall
is larger than the penetration distance. Whether a given wall behaves as thermally
thick or not depends on the time duration after a change at the surface occurs.
While this chapter is focused on thermal boundary layers growing with time in a
solid, the concept of a thermal boundary layer is at the heart of modeling thermal
boundary layers growing in a fluid, and gives rise to an understanding of convection
coefficients. Two combination thermal properties combinations will naturally appear;
thermal diffusivity and thermal inertia. The thermal diffusivity (α ¼ k=ρ=c) determines
pffiffiffiffiffiffiffi
how thick a boundary layer is. The thermal inertia (Γ ¼ kρc ) determines the heat
transfer rate from the surface to the interior.
This section is long, and readers anxious to develop numerical methods can
skip directly to Chap. 5 from here, or skim the remainder of this chapter, especially
the plots.

4.3.1 Suddenly Heated Thermally Thick Wall

The first variable-sized node case is one in which the surface temperature of a thick
wall is suddenly changed to a new value by some means and held there. The basic
growth of a penetration layer naturally emerges. An application of this case is the
so-called contact temperature, that is, the temperature that occurs at the interface of
two solid objects at different temperatures when suddenly placed in contact.

4.3.1.1 Example: Cooking a Hamburger

A 1 in. (0.0254 m) thick hamburger is initially at a uniform temperature
Tinitial ¼ 0  C (i.e., just finished thawing). The hamburger is to be cooked by placing
it on a hot skillet in a manner such that the temperature of the hamburger in contact
with the skillet will be at a fixed temperature Tsurface ¼ 200  C. Determine the time
it takes for the center of the hamburger to begin to warm (at which point the
hamburger is flipped).
116 4 1-Node Transient Models

Model Development: 1-Node Solution—Variable Capacitance

A 1-node model where the temperature of the node is known, but its size changes
can be developed to reveal the underlying relationship between space and time in
this type of thermal boundary layer. A characteristic penetration depth is deter-
mined that can be useful in constructing a more detailed grid, among other things.
The RC network is shown for this 1-node model in Fig. 4.18. In this model, the
size of the grid changes with time. Two nodes (surface and interior) are shown at
time t and a future time t + Δt. All the capacitance of the boundary layer is lumped
into the interior node. In this model, these nodal temperatures are known and
constant in time, with the node at the interior of the solid representing a growing
region of space as the thermal wave spreads away from the surface. The tempera-
ture of the interior node is taken to be the geometric
  average temperature
between
the surface and initial solid temperature T ave ¼ T surface þ T initial =2 . The dis-
tance δ is interpreted to be the distance from the surface to the region of influence at
a given time, and is the depth of penetration of the thermal signal. The average
difference between the surface and a location inside the region defined by node
some fraction, “f” of the total size associated with the node. A value of f ¼ ½ is
consistent with the conceptual model. However, the exact solution will be used to
“calibrate” and will be shown to be somewhat less than ½.
The capacity-carrying interior node does not represent a closed thermodynamic
system. Rather, because the size increases, the boundary at the leading edge of
the node moves into the wall at an instantaneous speed of dδ/dt. Mass crosses
this boundary at a rate m_ ¼ ρ dδdt A, and carries enthalpy (internal energy plus flow

work) into the node at a rate of mc _ p T initial  T ref , represented by a (variable)
current source, with Tref being a reference temperature assigned a thermal energy
value of zero.

Fig. 4.18 1 Node model (with variable R and C) at time t and a short time later, t + Δt
4.3 Thermal Boundary Layers in Thick Walls 117

The energy balance for the interior node is:

dE T surface  T ave 
¼ _ p T initial  T ref
þ mc
dt R
which states that the rate of change of energy stored in the node equals the rate at
which heat enters from the surface plus the rate at which energy enters at the leading
edge. The energy stored in the node is changing because of its size (not its
temperature, which is constant). The conductive thermal resistance also changes
with time as the thermal wave propagates into the interior, spreading out. The rate
of change of energy stored is:
0 1
ððð Tðave
dE d B C d
 
¼ ρ@ ¼
cv dT AdV ρcv T ave  T ref Aδ
dt dt dt
CV T ref

For a solid, c ¼ cv ¼ cp. Assuming that the size of the node is proportional to the
penetration distance (δ) and expressing the conductive resistance as the conductive
distance (fδ, where f is constant to be determined) divided by the thermal conduc-
tivity of the solid (ks) and contact surface area (A):

d
  T surface  T ave dδ 
ρc T ave  T ref Aδ ¼ fδ
þ ρcp T initial  T ref A
dt kA
dt

Recognizing that only the depth δ changes with time, and rearranging and grouping
factors (note that the reference temperature cancels):



dδ k T surface  T ave 1
¼
dt ρc T ave  T initial f δ

With the average temperature is taken to be the geometric average

(Tsurface + Tinitial)/2, the temperature ratio (Tsurface  Tave)/(Tave  Tinitial) equals
unity and the governing differential equation becomes

dδ α
¼
dt f δ

where the combined thermal property α ¼ k/ρ/c called the “thermal diffusivity”
with SI units m2/s has been introduced. The left hand side represents the rate of
change of energy stored associated with the growing size of the node. The right
hand side represents the heat transfer rate into the nodal volume and decreases
as the penetration depth increases. As the penetration layer grows, it becomes
self-insulating.
118 4 1-Node Transient Models

Separating variables and applying integration limits:

ðδ ðt
f δdδ ¼ αdt
0 0

Performing the integration yields a simple (but not simplistic) expression for the
thermal penetration depth:
pffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
δ¼ ð2=f Þαt

This expression shows that the penetration depth increases with the square root of
time. Also, the higher the thermal diffusivity of the material, the faster it grows. The
penetration depth increases with the square root of the thermal conductivity and
reflects competing effects. The conduction heat transfer rate is proportional to the
thermal conductivity, but the thicker the boundary layer, the greater the self-
insulating effect. The higher the combination ρc (which represents energy required
to change the temperature per unit volume), the harder it is for the thermal
disturbance to penetrate, with a square root dependency.
Expressing the penetration depth in terms of time:

f δ2
t¼
2α
This form shows that the time it takes for a thermal “disturbance” to propagate a
distance δ increases with the distance squared. Doubling the distance from the
surface increases the time it would take for penetration by a factor of four. This
result is due to the self-insulating property of the penetration layer.
Note that the growth of the thermal boundary layer does not depend on the
magnitude of the step change in surface temperature that initiates it. The leading
edge of the boundary layer can be envisioned as a signal that something has
happened at the surface, and it does not matter what that something is.
The rate of heat transfer from the surface to the interior is given by:

T surface  T ave kA  kA 
qc ¼ ¼ T surface  T ave ¼ pffiffiffiffiffiffiffiffiffi T surface  T ave
R fδ 2f αt

Expressing the thermal diffusivity in terms of its components (α ¼ k/ρ/c), and

recognizing that the temperature difference between the surface and the average
temperature is half of the overall temperature difference between the surface and
infinity (Tsurface  Tave) ¼ (Tsurface – Tinitial)/2:
"sffiffiffiffiffi# rffiffiffiffiffiffiffi!
1 kρc 
qc ¼ A T surface  T initial
8f t
4.3 Thermal Boundary Layers in Thick Walls 119

pffiffiffiffiffiffiffi
The material property formed by the combination Γ ¼ kρc is termed the “thermal
inertia” or the “thermal hardness.” The rate of heat transfer decreases with time
(inverse square root dependency), again demonstrating the self-insulating nature.
At the moment the surface temperature changes (t ¼ 0), the heat transfer rate is
infinite, a result of the temporal discontinuity posed in the engineering problem
statement. As time passes, the wave penetrates deeper into the material, reducing
the temperature gradient that drives it.
The exact solution to this problem is given by:

 rffiffiffiffiffiffiffi!
1 kρc 
qc ¼ pffiffiffi A T surface  T initial
π t

This exact solution differs from the 1-node model only in the value of the constant. The
1-node model reproduces the functional dependency on material properties and time
exactly. In this case, the 1-node solution can be “calibrated” against the exact solution
by a suitable choice of the average conductive distance (which defines “f”) such that:
sffiffiffiffiffi
1 1
¼ pffiffiffi
8f π

Squaring both sides and rearranging:

π
f ¼ ¼ 0:39
8
This value, somewhat less than half the size of the node, reflects that a more exact
solution would not predict a linear temperature distribution, but one skewed toward
the surface.
Table 4.1 shows material properties for several representative materials, sorted
in order of increasing thermal diffusivity. Liquid water and air are included as the
only fluids. A plot of the penetration distance as a function of thermal diffusivity
shown for these representative materials at various elapsed times is shown in
Fig. 4.19. Each boundary layer grows with a square root dependency on time.
The higher the thermal diffusivity (α), the faster it grows.

4.3.2 Application to Cooking a Hamburger

Results of this model to the cooking hamburger are shown in Figs. 4.20 and 4.21. The
penetration depth is shown as a function of time until it reaches the centerline in
Fig. 4.20, at which point the hamburger can be flipped. Not modeled yet is the
response of the cooked portion after flipping (best done with a more detailed
numerical model), during which time the temperature within the hamburger
redistributes. It is apparent that in the first minute, the thermal wave penetrations
120 4 1-Node Transient Models

Table 4.1 Room temperature thermal properties of selected substances (sorted by thermal
diffusivity)
(kρcp)0.5
Substance ρ (kg/m3) Cp (J/kg/ C) k (W/m/ C) α ¼ k/ρcp (m2/s) (J/m2/ C/s0.5)
Wood 420 2,720 0.11 9.63E08 354
Meat 950 3,520 0.4 1.20E07 1,157
H2O(l) 1,000 4,200 0.57 1.36E07 1,547
Glass 2,700 840 0.78 3.44E07 1,330
Brick 1,600 840 0.69 5.13E07 963
Steel 7,800 465 54 1.49E05 13,995
Cast Iron 7,200 447 52 1.62E05 12,937
Air 1.23 1,005 0.023 1.86E05 5.33
Aluminum 2,707 896 204 8.41E05 22,244
Copper 8,954 383 386 1.13E04 36,383

1000
1 year
H2O(L)

Brick

100 1 month
Glass
Meat
Penetration Distance (m)

1 day
10
1 hour
1
1 min
0.1
1 sec
Aluminum
Copper

0.01
Cast Iron
air
Steel

0.001
1.E-07 1.E-06 1.E-05 1.E-04 1.E-03
2
Thermal Diffusivity (m /sec)

Fig. 4.19 Penetration depths as a function of thermal diffusivity for selected materials

to approximately 0.5 cm. It takes two more minutes to propagate to 1.0 cm, and
almost 6 min to reach the centerline (1.27 cm).
Figure 4.21 shows temperature plotted against position at 1 min intervals. The
vertical dashed line shows the hamburger at the instant it is placed on the grill. After
1 min, the penetration layer has reached 0.5 cm, and the straight line between the
fixed surface temperature and the edge of the penetration layer gives a first approxi-
mation to the temperature distribution in the solid at that time. The penetration layer
4.3 Thermal Boundary Layers in Thick Walls 121

1.4
Centerline
Penetration Depth (cm) 1.2

0.8

Flipping Time
0.6

0.4

0.2

0
0 1 2 3 4 5 6
Time (minutes)

Fig. 4.20 Penetration depth vs. time for a cooking hamburger

200
180
160
140 time
Temperature (oC)

120
Centerline

100 6 min
Top Surface

80
60 1 min
40
20
0
0 0.5 1 1.5 2 2.5 3
Distance from Cooking Surface (cm)

Fig. 4.21 Temperature profiles for cooking hamburgers. Straight lines from fixed surface tem-
perature to the initial temperature at the penetration depth is shown at 1 min intervals

continues to grow, but the spacing between lines decreases. The temperature gradi-
ent driving the event (Fourier’s law of conduction) decreases with time.
This response, of course, is consistent with experience. If a cooking hamburger
is cut in half at some point to check, there will be a cooked layer near the cooking
surface, and a raw hamburger above it.
122 4 1-Node Transient Models

4.3.2.1 Example: Contact Temperature

A block of solid material A (with density ρA, specific heat cA, thermal conductivity kA),
initially at a uniform temperature TA, is suddenly brought into good thermal contact
with material B (with density ρB, specific heat cB, thermal conductivity kB), at a different
temperature TB. Determine the temperature at the point of contact, the temperature
distribution in the wall and the heat transfer rate at the wall as functions of time.
Figure 4.22 shows an RC network for this problem. Both materials are assumed
to experience a suddenly heated wall event, with the temperature at the surface
suddenly changed to the contact temperature value, yet to be determined. No
assumption has been made about whether the contact temperature changes with
time, but it will be shown to be invariant.
Since the interface is a surface area, with no capacitance associated with it
(and the ability to store energy), the rate at which heat enters from one side must
equal the rate of heat transfer to the other:

qA ¼ qB

Applying the result from the suddenly heated wall problem, choosing signs such
that if A is hotter than B, heat flows from the interior of A (at TA) to the contact
point (at Tc) and from the contact point to the interior of B (at TB) or vice versa:
"sffiffiffiffiffi# rffiffiffiffiffiffiffiffiffiffiffiffiffiffi! "sffiffiffiffiffi# rffiffiffiffiffiffiffiffiffiffiffiffiffiffi!
1 k A ρA c A 1 kB ρB cB
Að T A  T c Þ ¼ Að T c  T B Þ
8f t 8f t

Cancelling terms and rearranging:

sffiffiffiffiffiffiffiffiffiffiffiffiffiffi
TA  Tc k B ρB c B Γ B
¼ ¼
Tc  TB k A ρA c A Γ A

pffiffiffiffiffiffiffi
where the thermal inertia, Γ  kρc, naturally appears.

Fig. 4.22 1-Node model for two material (A and B) at different initial temperatures brought into
contact, with variable R and C at two instants of time
4.3 Thermal Boundary Layers in Thick Walls 123

This expression can be rearranged to solve for the contact temperature:

ΓA T A þ ΓB T B
Tc ¼
ΓA þ ΓB
which shows that the contact temperature is an average temperature weighted by
the thermal inertia. The higher the thermal inertia, the harder it is to change its
surface temperature, from which the term “thermal inertia” is based. Notice that
there is no time dependency, and therefore the contact temperature is immediately
established and maintained constant as time progresses, as long as both materials
exhibit thermally thick behavior. The rate of heat transfer from A to B (or vice
versa) decreases with time as a thermal boundary layer grows in both materials
(at different rates due to different thermal diffusivities), but the contact temperature
does not change with time.
The temperature difference ratio on the left hand side indicates that the contact
temperature resides between A and B. Whether it is closer to A or to B depends on
the ratio of the thermal inertia values. Consider the limits. If A has a much higher
inertia than B (ΓA  ΓB and the temperature ratio is therefore small), then Tc is
close to TA. If A and B have equal inertias, then Tc is right in the middle.
A table of contact temperatures for a variety of material combinations is shown
in Fig. 4.23. In this case, the hotter material is at 100  C and the colder material is at
0  C. The temperature of the contact point is listed in the corresponding box.
For example, if a hot brick is placed on a cold steel plate, the contact temperature
will be 6.4  C. That temperature is much closer to the steel because the thermal
inertia of steel is more than ten times that of the brick.
A good classroom demonstration is to pass around two room temperature objects
of similar shape but made of materials of very different thermal inertia. The objects
are known to be at the same temperature, say room temperature. Each student (at a
warmer temperature than the objects) will grab the objects, one in each hand, and
will think that one is hotter than the other. The sensation of hot or cold depends
on the contact temperature, which will be different, even though the objects are
known to be at the same temperature. Note that thermal properties of human flesh
are close to that of water, so choosing a brick or other ceramic (a ceramic mug will
do nicely) and a metal will give a good difference.

Fig. 4.23 Contact temperatures for combinations of representative materials

124 4 1-Node Transient Models

4.3.3 Suddenly Heated Thermally Thick Cylinder

A solid cylinder of radius R0 at an initial temperature Tinitial has its surface

temperature suddenly changed to Tsurface. Determine the penetration distance as a
function of time.
In this case, a variable capacitance node is used (Fig. 4.18), except that the
conduction area as well as the conduction distance changes with time as the thermal
boundary layer grows inwardly from the surface toward the centerline. The conduction
from the outer layers is focused somewhat toward a decreasing surface area at the
leading edge of the penetration layer.
In this case, the conductive resistance is taken to be the exact solution for
conduction between cylindrical shell with outer radius of the pipe surface (R0)
and an inner surface at a radius a fraction of the penetration depth from the surface
(R0  fδ). The volume of the nodal element is the exact value (rather than the thin-
wall approximation).
The nodal balance on the variable capacity node is:

dE T surface  T ave 
¼ _ T initial  T ref
þ mc
dt Rk
d h    i
ρc πR20  π ðR0  δÞ2 L T ave  T ref
dt

2πkL T surface  T ave dδ 
¼   þ ρc2π ðR0  δÞ T initial  T ref
ln R0Rf0 δ dt

Expanding and separating variables:


R0
ðR0  δÞln dδ ¼ αdt
R0  f δ

Dimensionless parameters will now be introduced. There is an obvious characteris-

tic dimension for which to normalize the penetration depth, namely, the cylinder
radius. Time can be normalized by a characteristic time (t*) to be determined:

δ
δ^ ¼
R0
t
^t ¼
t
Inserting these variables into the governing equation and rearranging:


  αt
^ ^ ^
1  δ ln 1  f δ dδ ¼  2 d^t
R0
4.3 Thermal Boundary Layers in Thick Walls 125

It is now possible to define the characteristic time for this problem by setting the
coefficient on the right hand side equal to a numerical value. That is, let

R20
t ¼
2α
Note that the same characteristic time as the finite cylinder is used (with the 2 in the
denominator) to make the dimensionless time the same as the Fourier number
derived previously for a cylinder with a fixed size node.
The governing differential equation for the penetration depth as a function
of time is:
 
1  δ^ ln 1  f δ^ dδ^ ¼ d^t =2

A good exercise in integrating by parts, not shown, yields the following solution,
with an initial condition that the penetration depth is zero at the initial time:
 

 1  f δ^     1 3 2
^ ^ ^
2ð1  f Þ ln 1  f δ  1  1  f δ ln 1  f δ   2 þ ¼ ^t
f2 2 2f f

This function is plotted in Fig. 4.24 for two values of f. The value 0.4 is based
on the experience from calibrating the thick wall problem, and the value 0.5
is the geometric average value. This result indicates that the initial thick wall heating
(or cooling) stage of a large Biot number (so that the surface temperature
equals the fluid temperature) solid cylinder will occur until the dimensionless time

0.8 f = 0.4
Penetration depth (d/r0)

0.6 f = 0.5

0.4

0.2

0
0 0.02 0.04 0.06 0.08 0.1 0.12 0.14 0.16
2
Dimensionless time (2at/r0 )

Fig. 4.24 Dimensionless penetration depth as a function of dimensionless time for the suddenly
heated cylinder
126 4 1-Node Transient Models

(Fourier number) equals approximately 0.1. For longer times, the fixed capacity
node is more appropriate as the cylinder gradually warms (or cools) as a lumped
thermal mass.

4.3.4 Thermally Thick Wall Suddenly Exposed

to a Hot (or Cold) Fluid

A variation of a suddenly heated wall is one in which a thick wall (with density ρ,
specific heat c, thermal conductivity k) at an initial temperature Tinitial is suddenly
exposed to a fluid with a constant convective/radiative coefficient h at a
temperature Tfluid. This case is more in line with the cases covered by the Heisler and
Gurnie–Lurie charts than the case where the surface temperature is changed to a known
value. It is more difficult to solve because the average temperature in the penetration
layer changes with time as the layer grows. Another variation (not considered here)
would involve subjecting the surface to a constant heat flux, with or without a simulta-
neous convection to the fluid.

4.3.4.1 Example: Boiling and Baking Boneless Chicken Breasts

While chefs might cringe at the thought, boneless chicken breasts can be cooked by
dropping them into boiling water. Alternatively, chicken breasts can be baked in an
oven. A series of photos of boneless chicken breast (approximately 3.8 cm thick)
after being boiled for different times and cut in half is shown in Fig. 4.25.
The evolution of a cooked layer is revealed. Even though the shape of the chicken
breast is clearly not an infinitely thick wall, for short enough times, the growth of a
nearly uniform thickness of cooked layer from the surface is apparent.
Both of these cooking scenarios can be reasonably modeled as the placement of a
solid object into a fluid (water for boiling, air for baking) at a different temperature.
In modeling this behavior, the only difference between cooking methods is the fluid
temperature (100  C for boiling, 177  C for baking at 350  F) and the convection
coefficient assumed (500 W/m2/K for boiling, 15 W/m2/K for baking). Both cases
are treated as a thick wall at an initial temperature (0  C) suddenly exposed to a fluid
at different temperatures. A general treatment is developed (using dimensionless
parameters) and then applied to the cooking chicken case.
A thermal resistance model for this case is shown in Fig. 4.26 at time t, and a
short time later, t + Δt. In this model, heat flows from the ambient fluid to the wall
surface through a convective/radiative resistance, and from the surface to the
interior (to the capacity-carrying node) through a conductive resistance in series.
The size of the capacity-carrying node increases, and therefore the leading edge
grows by conduction into a region at the initial temperature, as before. Mass crosses
this boundary, and the energy flux is modeled as a current source. The speed of the
leading edge is the rate of change of the boundary layer thickness. The temperature of
the capacity-carrying node is at the geometric average
of the surface temperature and
the initial temperature (T ave ¼ T surface þ T initial =2). The only difference between this
model and the case of a wall whose surface temperature is suddenly changed is that the
surface temperature in this case is not fixed.
4.3 Thermal Boundary Layers in Thick Walls 127

Fig. 4.25 Photographs

of similarly sized
(approximately 3.8 cm thick)
boneless chicken breasts cut
in half after boiling for
different times

An energy balance applied to the variable capacity-carrying node is:

d
  T surface  T ave 
ρcAδ T ave  T ref ¼ _ T initial  T ref
þ mc
dt Rk

The current source represents the enthalpy flux into the node (the flow of internal
energy plus the flow work). The mass flux is related to the rate of change of the
boundary layer thickness:

dδ
m_ ¼ ρVA ¼ ρ A
dt
128 4 1-Node Transient Models

Fig. 4.26 RC network for a thermally thick wall suddenly subjected to a hot (or cold) fluid

The conductive resistance is equal to the conductive distance divided by the thermal
conductivity and area. The conductive resistance is taken to be a fraction (f) of the
thickness of the boundary layer. Conceptually, f is approximately ½ and could be
set to that value or “calibrated” against the published exact solution for this
problem.
The energy stored in the capacitive-carrying node depends on the size of the
node and its average temperature, both of which change with time in this case.
Using the chain rule and expanding all terms:
 
 dδ dT kA T surface  T ave dδ 
ρcA T ave  T ref þδ ¼ þ ρcA T initial  T ref
dt dt fδ dt

The reference temperature term cancels on both sides. Rearranging:


dδ dT ave ðk=ρcÞ T surface  T ave
ðT ave  T initial Þ þ δ ¼
dt dt fδ

The first term represents the change in energy associated with the growing size of
the node, and the second term represents the change in energy due to the gradually
changing surface temperature. Dividing through by the temperature change
(recognizing that the average temperature is midway between the surface and the
initial temperatures), and multiplying through by δ and dt:

d T surface  T initial α
δdδ þ δ 2 ¼ dt
T surface  T initial f

Dimensionless variables reveal the underlying relationships between space and

time in this problem. A dimensionless surface temperature is introduced that is
4.3 Thermal Boundary Layers in Thick Walls 129

defined to have a value of unity initially, and to decay to zero as the surface
temperature approaches the fluid temperature as time goes to infinity:

T surface  T fluid
θs ¼
T initial  T fluid

The boundary layer thickness in normalized by a characteristic dimension (δ*)

whose value will be determined through the course of the dimensional analysis:

δ
δ^ ¼ 
δ
Similarly, time is normalized by a characteristic time (t*) whose value will be
determined:

t
^t ¼
t
The energy balance becomes, after manipulation:

dθs αt
δ^ dδ^ þ δ^ 2 ¼ 2 d^t
1  θs f δ

The dimensionless parameter on the right side can be set equal to unity, revealing
the fundamental relationship between space and time:

αt
¼1
f δ 2

Finally, the parameter-free energy balance is:

dθs
δ^ dδ^ þ δ^ 2 ¼ d^t
1  θs
Again, both the size and the average temperature of the boundary layer are
changing with time. This equation cannot be integrated, yet, as there are two
dependent variables (penetration distance and surface temperature).
Figure 4.26 shows that the surface temperature represents a voltage divider
between the ambient fluid and the interior:

T fluid  T surface T surface  T ave

¼
Rh Rk
130 4 1-Node Transient Models

Rearranging:

Rk T surface  T
¼
Rh T fluid  T surface

f δ=kA T surface  T surface þ T initial =2
¼
1=hA T fluid  T surface

Rearranging:

hδ T surface  T initial
2f ¼
k T fluid  T surface

Introducing the dimensionless variables:


hδ ^ T fluid þ θs T initial  T fluid  T initial 1  θs
2f δ ¼   ¼
k T fluid  T fluid þ θs T initial  T fluid θs

The parameter hδ*/k is a natural Biot number for this problem, and can be used to
define the characteristic dimension by setting:

hδ
2f ¼1
k
The characteristic penetration distance is therefore naturally defined as:

k k
δ ¼
2f h h

Returning to the characteristic time and inserting the characteristic penetration

distance: t ¼ α4fk h2 ¼ 4fkρc
2

h2
The surface boundary condition yields the following simple relationship
between the penetration distance and surface temperature:

1  θs
δ^ ¼
θs
The rate of change is:

dθs
dδ^ ¼
θ2s
4.3 Thermal Boundary Layers in Thick Walls 131

Eliminating the boundary layer thickness in the energy equation:


 ð1  θ s Þ 1  θs 2 dθs
dθs þ ¼ d^t
θ3s θs 1  θs
!
1 1
 dθs ¼ d^t
θs θ3s

Integrating:

1
ln θs þ ¼ ^t þ c
2θ2s

The initial condition is that the dimensionless temperature equals unity at time
t ¼ 0, and therefore the constant of integration is c ¼ ½.
The final relationship between surface temperature and time is:
!
1 1
ln θs þ 1 ¼ ^t
2 θ2s

The surface temperature cannot be expressed analytically in terms of time, but the
time is expressed in terms of the surface temperature. Therefore, a closed-form
solution is obtained, and the boundary layer thickness and surface temperature are
simply related. A plot of these functions is given in Fig. 4.27. The linear scale
shows the penetration depth increasing with a vertical slope initially, and a gradu-
ally decaying slope. The surface temperature drops quickly at early time, and
continues to drop (i.e., approach the fluid temperature) with time, but with a
decreasing rate. After one time constant (t ¼ t*), the dimensionless surface
temperature is 0.47. After ten time constants, it is 0.20 and after 100 time constants
it is 0.07.
Temperature profiles (temperature vs. position at various times) are shown in
Fig. 4.28. For each series, a straight line is drawn from the surface temperature to
the penetration distance at that point. The penetration layer grows much like the
case of a fixed surface temperature, but in addition, the surface temperature
gradually approaches the fluid temperature. There is a rapid change initially, and
the spacing between fixed temperature lines decreases as time increases. There is no
steady-state solution for the truly infinitely thick case.

4.3.4.2 Application to Cooking Chicken

A spreadsheet was used to demonstrate the response of cooking chicken to the
boiling and baking cooking methods. The input and derived parameter values are
shown in Fig. 4.29. The material properties are the same. The convection
coefficients are different, namely 500 W/m2/K for boiling, 15 W/m2/K for baking,
taken to be representative values for those environments. Also, the fluid
132 4 1-Node Transient Models

Surface temperature and penetration depth

1.8
1.6 ∗)
/δ
t h(δ
1.4
on Dep
et rati
1.2 Pen
1
0.8
0.6 Surface Temperature

0.4
0.2
0
0 0.2 0.4 0.6 0.8 1 1.2 1.4 1.6 1.8 2
Time (t/t*)

1 10
0.9 9
qs = (Tsurface-Tfluid)/(Tinitial-Tfluid)

0.8 8

Penetration Depth(δ/δ∗)
Surface Temperature
0.7 7
0.6 6
0.5 5
0.4 4
0.3 3
0.2 2
0.1 Penetration Depth 1
0 0
0.0001 0.001 0.01 0.1 1 10 100
Time (t/t*)

Fig. 4.27 Surface temperature and penetration distance as a function of time (all dimensionless).
The top figure is on linear axes and the bottom on semi-log x axes

temperatures are different, namely 100  C for boiling and 177  C for baking.
The dimensionless parameters for these cases are the characteristic time
(t* ¼ 2.7 s for boiling and 2,972 s for baking) and characteristic penetration depth
(δ* ¼ 0.8 mm for boiling and 26.7 mm for baking). These differences are due to the
convection coefficient. The higher convection coefficient for boiling means the
heating of the outer layers takes place quicker, but penetrates a shorter distance.
4.3 Thermal Boundary Layers in Thick Walls 133

1-θs = (Tsurface-Tinitial)/(Tfluid-Tinitial)
0.9
0.8
0.7
0.6
t/t*= 10
0.5
0.4
0.3 1
0.2 0.1

0.1
0.01
0
0 0.5 1 1.5 2 2.5 3 3.5 4
Dimensionless Penetration Depth (δ /δ∗)

1
0.9
1-θs = (Tsurface-Tinitial)/(Tfluid-Tinitial)

2.0
0.8 1.8
1.6
0.7
1.4
0.6 1.2
time
0.5 1.0
0.8
0.4
t/t*= 2 0.6
0.3 0.4
0.2 0.2

0.1 0.2

0
0 0.5 1 1.5 2
Dimensionless Distance (x/δ∗)

Fig. 4.28 Temperature profiles (temperature vs. distance from surface) at various times. The top
plot shows times increasing by an order of magnitude. The bottom plot shows lines with equal
increments (0.2) up to two time constants. The dotted line is for t ¼ t*. The surface temperature is
presented as 1-θs to correspond to a dimensional case in which the fluid temperature is higher than
the initial temperature. For each curve, a straight line from the surface to the leading edge is
shown. More detailed models in future chapters will capture the curvature

The surface temperatures of the two cooking methods are shown in Fig. 4.30 for
20 min of cooking. The safe cooking temperature (165  F ¼ 74.2  C) is indicated as
a dotted line. Boiling causes the surface temperature to rise above the safe cooking
temperature in a very short time, and then gradually approach (the fluid tempera-
ture). On the other hand, with baking, the surface temperature increases slowly and
134 4 1-Node Transient Models

Fig. 4.29 Parameter values for boiling (left) and baking (right)

120

100 Boiling
Surface Temperature (oC)

80 Safe Cooking Temperature = 74.2 oC

60

Baking
40

20

0
0 5 10 15 20
Time (min)

Fig. 4.30 Surface temperatures for boiling and baking chicken

after 20 min is near the safe cooking temperature, but still far from the fluid
temperature (177  C). That is, the surface layers are overcooked during boiling,
while the interior is still raw (making the chef cringe).
The corresponding depth of penetration of the surface layer is shown in
Fig. 4.31. The penetration depth grows somewhat faster in the boiling case, but
not dramatically different, since it is the conduction within the solid that is the
primary factor in the propagation of the thermal wave forward. Higher convection
driving the surface does have a noticeable influence on penetration depth, however.
The temperature distribution within the solid is shown in Figs. 4.32 (for boiling)
and 4.33 (for baking) for the early times. Figures 4.34 and 4.35 show the
4.3 Thermal Boundary Layers in Thick Walls 135

0.02 Centerline

0.015
Boiling
Penetration Depth (m)

Baking

0.01

0.005

0
0 5 10 15 20
Time (min)

Fig. 4.31 Penetration depths for boiling and baking chicken. The centerline of a 3.8 cm thick
sample is shown

120

Fluid Temperature
100

80 Safe Cooking Temperature

Temperature (oC)

Centerline

60

40 time

20 60 sec

10 sec
0
0 0.5 1 1.5 2
Distance from surface (cm)

Fig. 4.32 Temperature distributions for boiling for the first minute of cooking (lines at 10 s
intervals)

temperature distribution for later times. For boiling, the surface temperature
exceeds the safe cooking temperature within the first 10 s of cooking, but does
not penetrate very far. For baking, the temperature is more uniform, but cooking
still proceeds by heating an outer layer inward.
136 4 1-Node Transient Models

200
Fluid Temperature
180
160

Centerline
140
Temperature (oC)

120
100
Safe Cooking Temperature
80
60
time
40
6 min
20
1 min
0
0 0.5 1 1.5 2
Distance from surface (cm)

Fig. 4.33 Temperature distributions for baking for the first 6 min (lines at 1 min intervals)

120

Fluid Temperature
100

80
Temperature (oC)

Safe Cooking Temperature

Centerline

60
time

40
6 min

20
1 min

0
0 0.5 1 1.5 2
Distance from surface (cm)

Fig. 4.34 Temperature distributions for boiling for 6 min (lines at 1 min intervals)

These results are all consistent with experience, but hopefully now with a deeper
fundamental grasp. The slower something is cooked, which is controlled by the
convection environment, the more uniform the cooking will be, as conduction to the
interior limits how quickly it cooks in the center.
4.3 Thermal Boundary Layers in Thick Walls 137

200
Fluid Temperature
180
160

Centerline
140
Temperature (oC)

120
100
80 Safe Cooking Temperature

60 time
30 min
40
20
5 min
0
0 0.5 1 1.5 2
Distance from surface (cm)

Fig. 4.35 Temperature distributions for baking for 30 min (lines at 5 min intervals)

The focus of this analysis is on the early cooking times, when the cooking object
exhibits thermally thick behavior. When the penetration depth reaches the center-
line, the object is no longer thermally thick, and the fixed size 1-node model
(or more detailed numerical analysis) is more appropriate.

4.3.4.3 Heisler and Gurnie–Lurie Charts Revisited

Close inspection of the published Heisler or Gurnie–Lurie charts at low Fourier
number (dimensionless time) shows an offset of approximately 0.1. That is, the
centerline temperature remains at the initial temperature for some time before an
exponential decay occurs. This behavior reflects the transition from an initial thick
wall phase, where the penetration layer grows with time, to a lumped heating
(or cooling) stage. That transition occurs when the penetration distance reaches
the centerline of the object.
The difference between the Heisler charts (dimensionless temperature vs. time)
and the closed formula predicted by the fixed size 1-node model (Fig. 4.15) is that
the former is a plot of the centerline temperature, while the latter is a plot of the
average temperature. The Heisler charts, therefore, include a separate plot of the
average temperature (a spatial integral) in the form of a dimensionless total heat
transferred. That plot is not necessary with the 1-node model which predicts the
time dependence of the average temperature. Furthermore, the average temperature
of a finite-sized object during the thick wall phase can be calculated from the
1-node model, even though all the action is in the growing penetration layer.
Therefore, there is really no need to attempt to adjust the 1-node model to account
for the initial phase.
In short, for applications of heating (or cooling) objects subjected to sudden
changes in their environments, the 1-node model can be used to quickly and in
138 4 1-Node Transient Models

closed-form estimate the response. If it is established that the scenario falls into a
thick wall category, that model can be used. The 1-node model is not restricted to
any geometry, or to homogeneous objects. Informed estimates of conductive and
convective/radiative resistances are needed. The chapters on multidimensional
steady-state conduction will provide additional tools for that. For cases where the
change in the environment is not sudden, or there are additional changes, numerical
methods that address time dependency (i.e., Euler method) can be used.
Two examples that extend the method to situations not covered in the Heisler or
Gurnie–Lurie charts are considered in detail next. In one, an internal heat genera-
tion term is added to the cylindrical geometry case to model the effect of
microwaving. In the other, the cooling mug of coffee is revisited, with a focus on
the initial moments after pouring hot coffee into a room temperature mug.
If it is judged that more detail is needed, then the next step is to begin the process
of breaking physical space into more than 1-node. The next chapters develop that
methodology.

4.4 Cooling Mug of Coffee Revisited: Mug Heating Phases

An application that ties together several concepts from this chapter and shows the
versatility of the 1-node models is to analyze the response of a room temperature
mug immediately after hot coffee is poured into it. There is an initial heating phase
during which time the mug acts as a thermally thick wall, followed by a phase
where the mug is simultaneously heated by the coffee and cooled by the air until it
reaches a quasi-steady state temperature.
Imagine it takes no time at all to pour hot coffee into a room temperature mug.
Imagine also that the coffee behaves like a solid, and does not move after it is
poured. The interface between the mug and coffee will immediately attain an
intermediate contact temperature, and a thermal wave will propagate into both
the coffee and the mug (but at different rates, due to different thermal properties).
The outer temperature of the mug remains at its initial temperature during this
phase, and the variable-sized nodal analysis applies. Once the penetration depth
reaches the outer surface of the mug, the fixed size nodal analysis applies. This
behavior is developed in this section.
In this analysis, the average temperature of the coffee is assumed to remain fixed
at the boiling point of water (100  C). This assumption is equivalent to assuming
that the coffee has an infinite capacitance, so the heat transferred to the mug does
not change the coffee temperature. In the next chapter, this assumption will be
relaxed, and the coffee will cool simultaneously with the mug being heated. That
problem, however, cannot be solved in closed form, and a numerical simulation is
developed for that problem.
4.4 Cooling Mug of Coffee Revisited: Mug Heating Phases 139

4.4.1 Phase 1: Thermally Thick Mug

During the initial phase of heating, the thin-wall approximation (mug thickness
much less than the inner diameter) is invoked, so that the mug behaves like a thick
wall of constant-area. A good exercise would be to apply the cylindrical model
instead. Also, an alternative model to treating the coffee as a solid during this phase
would be to assume a representative value of convection (say, 400 W/m2/K).
Table 4.2 lists thermal properties assumed for a typical ceramic and for coffee at
nominal room temperature values (the thermal properties are assumed to not change
significantly with temperature).
The contact temperature, that is, the temperature that is achieved at the instant of
the event is therefore:

ΓA T A þ ΓB T B ð1; 587Þð100Þ þ ð1; 166Þð25Þ

Tc ¼ ¼
ΓA þ ΓB ð1; 587Þ þ ð1; 166Þ

T c ¼ 68:2  C

Notice that this temperature is different than the 87.7  C that was calculated for the
starting temperature of the main cooling event. The latter thermodynamic calcula-
tion depends on the finite amount of coffee and mug and is a fundamentally
different phenomenon than the contact behavior which is independent of
material size.
Immediately after the coffee is poured, the contact temperature of 68.2  C is
established and a thermal boundary layer grows in both the ceramic and the coffee.
The underlying model is valid as long as the material is thermally thick. That
criterion is met as long as the thickness of the growing boundary layer is less than
the physical thickness of the material (w):

δ<w

Inserting the suddenly heated wall result for the boundary layer thickness as a
function of time:
pffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
δ¼ ð16=π Þαt < w

Table 4.2 Property values for coffee/mug problem at nominal room temperature
Property Symbol Unit Coffee (A) Ceramic (B)
Thermal conductivity k W/m/K 0.6 1.0
Density ρ kg/m3 1,000 1,700
Specific heat c J/kg/K 4,200 800
Thermal diffusivity α  k=ρ=c m2/s 1.43(10)7 7.35(10)7
pffiffiffiffiffiffiffi
Thermal inertia Γ ¼ kρc J/s0.5/m2/K 1,587 1,166
140 4 1-Node Transient Models

Solving for the time restriction:

π w2
t<
16 α
For the coffee mug:

 π  ð0:00398 mÞ2
t<
16 7:35ð10Þ7 m2 =s

t < 4:2 s

That seems like a reasonable value. It can be estimated empirically by quickly

pouring hot coffee into a ceramic mug, holding the mug and counting how long it
takes until it starts to feel warm.
Figure 4.36 shows the results for this initial phase of the coffee cooling process.
Temperature is plotted against position at 1 s increments of time for the first 5 s. The
details of construction of this plot are discussed in Appendix 2. During this time, the
mug behaves as if it were infinitely thick. The temperature at the inner wall is
brought instantly to the contact temperature (68.2  C) and it stays there. Rapid
growth of the two penetration layers occurs in the first second, due to the extremely
high temperature gradient, with a thermal boundary layer growing approximately
2 mm in the mug, and about 1 mm in the coffee. The difference between coffee and

100
90 e
m
Ti
80
Contact Temperature
70
Temperature (oC)

60
50
e
m
40 Ti

30
20
10
Coffee Mug
0
4 2 0 2 4
Distance from Inner Wall (mm)

Fig. 4.36 Temperature profiles for the first 5 s of the coffee cooling example. Temperature is
plotted against position at 1 s increments. Each curve is a parabolic fit (explained in Appendix 2)
4.4 Cooling Mug of Coffee Revisited: Mug Heating Phases 141

mug penetration depths is a direct result of the thermal diffusivity being approxi-
mately four times as great in the mug as in the coffee. The leading edge of the
boundary layer advances with time, but the incremental change (i.e., space between
curves) decreases with time (because the temperature gradient at the wall that
drives the heat decreases).
The coffee in this model is treated as a hypothetical solid. Of course, the coffee
flows, introducing another mechanism for heat transport. There are actually two
distinct mechanisms for liquid movement. The coffee was just poured, so it is
“churning.” Even if that effect settles (but in 10 s, that’s not likely to be complete),
there is a natural convective mechanism. The coffee adjacent to the mug will be
colder, and hence slightly more dense, than the coffee in the middle and will tend to
fall along the wall, to be replaced from the top. A mild circulation pattern will be
induced that profoundly changes the thermal behavior. Convection coefficients
allow engineering models to incorporate these effects. If incorporated, these effects
would increase the effectiveness of heat transfer on the coffee side, and result in a
somewhat higher contact temperature. Also, the contact temperature would drift.
Note that during this initial mug heating phase, the outer wall of the mug remains
at ambient temperature, which means that heat is not being lost by the coffee/mug
combination. This initial phase involves redistribution within the original system
boundary.

4.4.2 Phase II: Mug as Lumped Capacity Node

The thermally thick variable capacitance 1-node model that produced these results
breaks down as soon as the leading edge reaches the outer wall of the mug. There is
a shift to a new phase, and a new model, occurs. A conventional fixed size model
is shown in Fig. 4.37. Treating the mug itself as a 1-node model in contact
with coffee at a fixed temperature on one side and air on the other, the mug will
continue to warm to an intermediate temperature between the air and the coffee.
This process establishes a quasi-steady balance between the mug and coffee, and
they subsequently cool together in a third phase, namely, the main cooling event
analyzed in Chap. 3.

Tcoffee Tinside Tmug Toutside T∞

convection
radiation
Rcm Rmug Rmug
2 2 Rma
Fig. 4.37 RC network Cmug
for phase II mug heating
142 4 1-Node Transient Models

The nodal energy balance applied to the mug node is:

dT m T coffee  T mug T air  T mug

Cm ¼ þ
dt Rcm þ Rm =2 Rm =2 þ Rma

The solution to this problem, with the coffee (and air) temperature fixed is:

T mug  T ss t
¼ eRC
T mug, initial  T ss

where the steady-state mug temperature is given by:

T coffee
ðRcm þRm =2Þ þ ðRmaTþR
air
m =2Þ
T ss ¼ 1
ðRcm þRm =2Þ þ ðRma þR
1
m =2Þ

And the time constant is:

C
RC ¼   mug 
1
Rcm þRm =2 þ Rm =2þR
1
wa

Input and derived parameter values for the phase II mug heating are listed in
Table 4.3.
The time constant of 23.2 s is long compared to the 4.2 s it took for the initial
thermally thick boundary layer phase, and short compared to the subsequent
cooling of the coffee/mug assembly. Experimentally, the mug temperature peaked
at about a minute, giving good order of magnitude agreement.
The model predicts the temperature of the mug at its centerline (T). The
temperatures at the inner and outer walls (Ti and To) of the mug are related by a
voltage divider-type calculation. The inner wall temperature is obtained by equat-
ing the heat rate from the coffee to the center with the heat rate from the coffee to
the inner wall:

qcoffee>mugcenter ¼ qcoffee>innerwall

T coffee  T T coffee  T i
¼
Rcm þ Rm =2 Rcm

Solving for the inner wall temperature:


Rcm 
T i ¼ T coffee  T coffee  T
Rcm þ Rm =2

A similar analysis between the mug center and ambient air yields an expression for
the outer wall temperature:
4.4 Cooling Mug of Coffee Revisited: Mug Heating Phases 143

Table 4.3 Parameter values for phase II mug heating

Input parameters
Phase I time 4.22 s

Tmug, initial 46.6 C Average mug temperature (end of phase I)

Tair 25 C Ambient temperature

Tcoffee 100 C Fixed coffee temperature
w 0.003975 m Mug wall thickness
D 0.08255 m Mug inner diameter
H 0.098 m Mug inner height
hma 13.9 W/m2/K Outer convective/radiative coefficient
hcm 406 W/m2/K Inner convective coefficient
kmug 1.0 Thermal conductivity of mug
rhomug 1,700 kg/m3 Density of mug
cmug 800 J/kg/K Specific heat of mug
Derived parameters
Cmug 144.0 J/ C/W Capacitance of mug wall

Rcm 0.097 C/W Inner convective resistance

Rm 0.149 C/W Mug wall conductive resistance

Rma 2.58 C/W Outer convective/radiative resistance

R 0.161 C/W Equivalent resistance
RC 23.2 s Time constant

Tss 95.5 C Steady-state mug temperature


Rm1
To ¼ T1 þ ðT  T 1 Þ
Rm1 þ Rm =2

Figure 4.38 shows temperature profiles during the second phase of heating of the
mug at 10 s increments. The dashed line shows the temperature profile at equilib-
rium. The coffee temperature is fixed in this model at 100  C. Within the mug wall,
straight lines are drawn from the inner surface to the center and from the center to
the outer wall. In moving from phase I to II model, there is a discontinuous jump on
the outer wall surface from ambient (25  C) to the value given by this second phase
of mug heating (38.7  C). There is a similar discontinuous jump from the contact
temperature used in the infinite mug phase (68.2  C) to the value of this model
(69.8  C), but the change there is (coincidentally) less obvious. The thicknesses of
the fluid thermal boundary layers shown are estimates from a future chapter and are
meant to be illustrative. This model indicates that a concave up shape persists for
the temperature profile in the mug wall as the wall heats, and eventually attains a
linear temperature distribution in the mug at equilibrium. The model is focused on
the heating of the mug in the initial stages. The gradual decrease in coffee
temperature with time follows with the next phase of the cooling, which is taken
up further in the next chapter.
144 4 1-Node Transient Models

100
90 equil
60
80
50
70 40
Temperature (oC)

60 30
20
50
10
40 0
30
20
10
Coffee Mug Air
0
-2 -1 0 1 2 3 4 5 6
Distance from Inner Wall (mm)

Fig. 4.38 Phase II heating of mug. Temperatures are plotted against distance from the inner wall
of the mug at 10 s increments of time (from the start of phase II at 4.2 s). The dashed line shows the
temperature profile at quasi-equilibrium (steady-state, with the coffee and air temperatures fixed).
The thermal boundary layer thicknesses in the fluids (coffee and air) are estimates based on models
of future chapters

Appendix 1. Detailed Calculation of Individual Coffee Mug

Thermal Resistances

In this appendix, the formulas used to calculate the individual resistance values are
detailed. Figures 4.1 and 4.9 are repeated here for reference. Input parameter values
were listed in Figs. 4.4 and 4.10. The rationale for choosing values is described,
much of the detail of which is developed in later chapters.

D W at,rad
at,evap
L at,conv top
T(t)
V H
ct at,rad
side
T∞ = T∞w
coffee H0 cs ws as,conv

C bottom
mug cb wb fb
Appendix 1. Detailed Calculation of Individual Coffee Mug Thermal Resistances 145

Top Channel Resistors

Free Surface Convective Resistance, Liquid Side: Rct

For this resistance, the coffee/air interface at the free surface is treated like a
convective resistance. On the liquid side, the resistance is:

1 1
Rct ¼ ¼ 
hcoffee=surface Asurface hcoffee=surface πD2 =4

The value chosen for convection coefficient (300 W/m2/K) is representative of a

natural convection for liquids.

Free Surface Convective Resistance, Air Side: Rat

There are three resistances in series: convective, radiative, and evaporative. Each of
them is modeled as:

1 1
Rat ¼ ¼ 
heffective Asurface heffective πD2 =4

The values chosen (6.8, 6.6, and 10 W/m2/K for convective, radiative, and evapo-
rative) were chosen based on detailed Nusselt number and radiation coefficient
analyses (methods developed in later chapters) for a coffee temperature of 70  C.

Side Channel Resistors

Inner Side Wall Convective Resistance: Rcs

The heat transfer between the coffee and the side wall is by the convective mode,
and radiation is not important for two reasons: the inner surface does NOT fall into
the model for a resistance model because the inner wall “sees” itself. Also, because
the liquid convection coefficient is over an order of magnitude larger than typical
air values, convection dominates radiation. The thermal resistance is therefore:

1 1
Rcs ¼ ¼
hcoffee=side AInnerSurface hcoffee=side ðπDLÞ

The value for hcoffee/side (470 W/m2/K) was obtained from a Nusselt number
calculation for a coffee temperature of 70  C.

Side Wall Conductive Resistance: Rws

This resistance represents a conductive resistance across the side wall. Following
the general outline, two surfaces are identified, and average distance between the
surfaces and average areas calculated. For the side wall, the surface 1 is defined as
the inner submerged surface. Its area is:
146 4 1-Node Transient Models

A1 ¼ πDðH  LÞ ¼ 0:0198 m2

Surface 2 is defined as the total surface area of the outer wall exposed to air,
including the top rim and the inner surface area above the coffee surface. This 2D
conduction problem is discussed in some detail in a future chapter. The area is:

π 
A2 ¼ π ðD þ 2wÞH 0 þ πDL þ ðD þ 2wÞ2  D2 ¼ 0:0372 m2
4
The outer area is 88 % greater than the inner one. The geometric average area is:

A12 ¼ ðA1 þ A2 Þ=2 ¼ 0:0285 m2

The average distance between surfaces 1 and 2 (Δx12) can be taken to be simply the
wall thickness, w, although that underestimates the true distance. The thermal
resistance is:
Δx12 w
Rws ¼ ¼
k12 A12 k12 ðA1 þ A2 Þ=2

The thermal conductivity k12 is something you “look up” for the material in
question, unless there is more specific information available for a material in
question. In this case, an Internet search (January 30, 2011) of “thermal conductivity
of ceramic mug” yielded:
http://www.askmehelpdesk.com/physics/why-ceramic-mug-cools-liquid-faster-
98327.html
k12 ¼ 15 W=m=K

http://www.madsci.org/posts/archives/2004-12/1102383405.Eg.r.html

k12 ¼ 0:5  20 W=m=K

A good website for material property values is “Engineering Toolbox” http://www.

engineeringtoolbox.com/.
Searching in material properties tabs for thermal conductivity. . . http://www.
engineeringtoolbox.com/thermal-conductivity-d_429.html.
The thermal conductivity for porcelain, and the value used here, is published as:

k12 ¼ 1:5 W=m=K

Such a wide range of uncertainty in a material property is not uncommon.

Engineering judgment often comes into play.

Outer Side Wall Convective Resistance: Ras

The exposed outer surface of the side wall, including the inner area above the coffee
level and the top of the rim (A2 from the conduction analysis), exchanges heat by
Appendix 1. Detailed Calculation of Individual Coffee Mug Thermal Resistances 147

convection in parallel with radiation with ambient air. Technically, the inner
portion “sees” the coffee and itself in addition to ambient, but this is a small part
of the total exchange so this effect is not considered. The resistances are:

1 1
Ras, conv ¼ and Ras, rad ¼
hair=sides A2 hradiation A2

The values for hair/sides (6.8) and hradiation (6.6) are obtained from a Nusselt and
Kirchhoff analysis for a coffee temperature of 70  C.

Bottom Channel Resistors

Inner Bottom Wall Convective Resistance: Rcb

The only difference between this resistance and the inner side wall is the exposed
area:

1
Rcm ¼ 
hcm πD2 =4

where hcm is taken as the same natural convection value as between the coffee and
the sides.

Bottom Wall Conductive Resistance: Rfb

For the conductive resistance through the mug bottom, consider the shape to be a
truncated cone, with surface 1 as the inside surface of the mug, and surface 2 to be
the surface area in contact with the floor. The area is taken to be the geometric
average:
" #
1 πD2 π ðD þ 2wÞ2
A12 ¼ þ
2 4 4

The thermal resistance is therefore:

Δx12 w
Rwb ¼ ¼
k12 A12 k12 A12

Conductive Resistance Across the Floor: Rfb

Heat transfer between the floor of the mug and its surroundings is difficult to
precisely define without more information as to the nature of the table on which
the mug rests. There is an air gap trapped beneath the bottom rim. If (a big “if”)
natural convection effects are sufficiently suppressed so that the air is stationary,
then a conductive resistance between the mug bottom and the table surface can be
148 4 1-Node Transient Models

estimated. A radiative channel between flat plates exists between the floor and the
bottom.
Then there is a conductive series resistance from the table surface toward the
interior of the table. To estimate this resistance, assume the table is large and take a
long range view so that the heat transfer emanates outward spherically. The inner
surface (surface 1) is taken to be a hemisphere with a radius such that the surface area
of the hemisphere equals the surface area of the floor underneath the mug. That is:

2πr 21 ¼ π ðD þ 2wÞ2 =4

or
pffiffiffi
r 1 ¼ ðD þ 2wÞ= 8

The rate of heat transfer across a hemispherical shell of inner radius r1 and outer
radius r2 (from 1 to 2) is

dT
q12 ¼ k2πr 2
dr
Separating variables and integrating:

Tð2 ðr2
q12 dr
dT ¼
2πk r2
T1 r1


1 r2


q12 r q12 1 1
T2  T1 ¼ ¼ 
2πk 1 r1 2πk r 2 r 1

ðT 1  T 2 Þ
q12 ¼  
1 1
2πk r1  1
r2

In this case, surface 1 represents the floor surface (subscript “b” for bottom), which
is modeled as a hemisphere of radius r1 (that gives the same area as that of the floor
under the mug), and surface 2 represents infinity (so 1/r2 ! 0 and T2 ¼ T1 is
ambient). The heat transfer rate, defining the conductive resistance, is:

T1  T1
q11 ¼ ¼ 2πkr 1 ðT 1  T 1 Þ
R11
pffiffiffi
1 2
Rb1 ¼ ¼
2πkr 1 πkðD þ 2wÞ

The following table details results for the floor resistance estimate. This analysis
could use a couple of schematics, but the text in the opening paragraph of this
Appendix 2. Parabolic Fits for Plotting Temperature Profiles 149

section contains a word picture. In the end, the result suggests that the total
resistance through the bottom of the mug (40.2  C/W) is much higher than between
the free surface and ambient (8.51  C/W), which is less than that between the side
wall and ambient (2.21  C/W). That means the bottom surface behaves as a nearly
insulated surface.

Detail of floor resistance

D 0.083 m Inner diameter of mug
hrim 0.004 m Height of air gap
wrim 0.001 m Width of rim in contact with floor
kai r 0.023 W/m/K Thermal conductivity of air
kfloor 0.17 W/m/K Thermal conductivity of table (Oak value)
kmug 1.5 W/m/K Thermal conductivity of mug
Aair gap 0.00541 m2 Area of air gap
Arim 0.000264 m2 Area of rim contact
emiss 0.95 Emissivity of both surfaces
hrad 6.8 Radiation coefficient (between 70 and 25  C
plates)

Rbc,air 32.1 C/W Across air gap

Rbc,gap 27.2 C/W Across air gap
radiation
Rrim 10.1 Through rim

Rbinf 31.2 C/W Infinite floor

Rfloor 37.1 C/W Effective floor contact resistance
UAfloor 0.027 W/ C Effective floor contact conductance
Ufloor 4.7 W/m2/K Effective floor contact coefficient

Appendix 2. Parabolic Fits for Plotting Temperature Profiles

This appendix demonstrates a common technique for approximate methods for

boundary layer analysis. If three conditions are specified at two locations (the wall
and the free stream), then the shape of a function can be fit to a parabola, as detailed
here. The three conditions here are that the temperature at the wall is fixed at the
contact temperature, and at the edge of a boundary layer, the temperature is fixed at
the initial value, and the temperature gradient is set to zero.
For this case, the heat transfer rate between the two solids will not be
constrained, and therefore the heat transfer rate between them will not match.
A fourth condition would be required to match the heat transfer rate between the
two solids, and the temperature could be fit to a cubic. A good exercise (with a fair
degree of additional algebra) would be to develop that cubic fit.
For the purposes of demonstrating the temperature profiles (snapshots of temper-
ature plotted against distance from the wall surface at fixed times), a simple
parabolic fit that fixes the temperature as the wall surface, the temperature at the
150 4 1-Node Transient Models

edge of the boundary layer and forces the slope to be zero at the edge of the boundary
layer. That is, with “x” the distance from the wall, an assumed parabola is:

T ðxÞ ¼ a þ bx þ cx2

where the coefficients (a, b, c) are determined from constraints imposed on the fit.
At the wall surface (x ¼ 0), the temperature is fixed at Tc:

T ð0Þ ¼ T c ¼ a þ bð0Þ þ cð0Þ2

or simply, a ¼ Tc.
At the edge of the boundary layer, x ¼ δ, the temperature is set to ambient:

T ðδÞ ¼ T 1 ¼ T c þ bδ þ cδ2

where the result for coefficient “a” has been inserted. Also, the slope at the edge of
the boundary layer is set to zero:

dT 
¼ 0 ¼ b þ 2cδ
dx x¼δ

Multiplying this result by d and subtracting from the previous:

T ðδÞ ¼ T 1 ¼ T c  cδ2

Tc  T1
c¼
δ2
Inserting this into the third constraint:

2ðT c  T 1 Þ
b ¼ 2cδ ¼
δ
A curve fit that yields an approximate shape for the temperature profile therefore is:

x  x 2
T ðxÞ ¼ T c  2ðT c  T 1 Þ þ ðT c  T 1 Þ
δ δ
A somewhat more compact and dimensionless form of this expression is:

T ðxÞ  T 1 xx 
¼ 2
Tc  T1 δ δ
This formula does not give the exact shape of the temperature profile, but an
approximate fit. The time dependency of the profile is contained in the boundary
layer thickness, δ, which grows with time.
The inner wall of the mug is at the contact temperature, and the temperature falls
within the thermal boundary layer, eventually reaching ambient. There is no single
Appendix 2. Parabolic Fits for Plotting Temperature Profiles 151

mug temperature. However, an average mug temperature can be obtained by

integration (whose value would yield the correct total energy stored, assuming
constant specific heat):
ðw
T ave w ¼ Tdx
x¼0

where the coordinate x is measured from the inner wall of the mug outward. Within
the boundary layer (x < δm), the temperature profile was fit to a parabola to generate
the plot of Fig. 4.36, and outside the boundary layer (x > δm), the temperature is
ambient. Therefore, the integration can be broken into two pieces:

δðm ðw
T ave w ¼ T ðxÞ dx þ T 1 dx
x¼0 δm

 2
The parabolic fit for temperature is: T ðxÞ ¼ T c  2ðT c  T 1 Þδxm þ ðT c  T 1 Þ x
δm
Performing the integrations:
 δ
x2 x3  m
T ave w ¼ T c  2ðT c  T 1 Þ þ ðT c  T 1 Þ 2  þ T 1 ½x
jδwm
2δm 3δm 0
δm
Doing the algebra: T ave ¼ T 1 þ 3w ðT c  T 1 Þ
Inserting the expression for the boundary layer thickness as a function of time:
pffiffiffiffiffiffiffi
4 αm t
T ave ¼ T 1 þ pffiffiffi ðT c  T 1 Þ
3w π

Similarly, the average temperature of the coffee can be calculated during this time:

ðR
T ave, coffee πR ¼ 2
T2πrdr
r¼0

where in this case, the cylindrical shape is considered, so the integration is conducted
over a cylindrical shell from the centerline of the coffee (r ¼ 0) to the inner wall of
the mug (r ¼ R). The coordinates r and x are related simply by r ¼ R + x. Breaking
the integral into two parts, inside and outside the thermal boundary layer:

ðc
Rδ ðR
T ave, coffee πR ¼
2
T coffee, init 2πrdr þ T2πrdr
r¼0 Rδc
152 4 1-Node Transient Models

Since the radius (R ¼ 32 mm) is much larger than the maximum thermal boundary
layer thickness (dc,max ¼ 2.7 mm), the average coffee temperature changes very
little during the initial heating, while the mug has increased from ambient (25  C)
to 39.4  C.

Workshop 4.1. Cooking 1-Node Hot Dogs

h
T(t)

D k T∞
ρ
c

A hot dog of diameter D (¼11/1600 ), length L is to be cooked by several methods

(boiling baking, barbequing, microwaving). The hot dog is initially at 5  C and is
placed into a cooking environment with a fluid temperature T1. It is considered to be
cooked when the average temperature exceeds 70  C. Estimate the cooking times
and the surface and center temperatures at that time for each method. Develop a
1-node extended model, where the total thermal capacitance of the hot dog is lumped
into a single node and a thermal channel with a conductive resistance from the
interior to the surface is in series with a convective resistance between the outer
surface and ambient. Use input parameters from the input parameter tables shown.

For the convection coefficient, the value for boiling water (800 W/m2/K) is on the
high end of natural convection in liquid water. There is expected to be significant
Workshop 4.1. Cooking 1-Node Hot Dogs 153

fluid motion induced by the boiling process. For baking (at 350  F ¼ 177  C) and
microwaving, a value of 15 W/m2/K is typical of natural convection in air, in parallel
with a radiative coefficient. The value for barbequing (25 W/m2/K) is taken to be
somewhat higher due to the strong updraft caused by the fire.
For the barbequing and microwaving cases, some of these input parameters were
chosen after the fact, to give cooling times consistent with experience. For
barbequing, the fluid temperature is taken to be 500  C. In practice, raising the
distance between the coals and the food lowers the fluid temperature in contact with
the food. There is an important radiation effect from the coals that is not considered
here as it does not fit into either radiation model. For the microwaving case, the
value of the rate of heat generation is given in Watts absorbed per unit volume, and
the value (1(10)7 W/m3) is chosen to be one which gives a cooking time consistent
with empirical experience. That’s cheating, a little. . . Or, rather, it is using experi-
mental results and a theoretical model to evaluate an approximation for the effec-
tive rate of absorption of microwave energy, for which no attempt to model from
first principles is made here.
Note: A detailed model development follows. Conduct your own analysis before
referring to it.

Model Development
A general 1-node model with a heat source (for the microwave case) and a thermal
channel between the capacity-carrying node and ambient with a conductive and
convective/radiative channel in series is shown in Fig. 4.39.
The energy balance applied to this general RC network is:

dT T1  T
C ¼ Q_ þ
dt Rk þ Rh
The solution (with constant parameter values) has been given in Chap. 1:

T  T SS t
¼ eCðRk þRh Þ
T 0  T SS

where T ss ¼ T 1 þ Q_ ðRk þ Rh Þ is the steady-state temperature and T0 is the initial

temperature. Note that the microwave case is the only one for which there is internal
heat generation, and there is a difference between the steady-state and ambient fluid
temperatures.

.
Q T∞
T(t) Ts

Rk Rh
Fig. 4.39 General 1-node C
model applicable to hot dog
model
154 4 1-Node Transient Models

The surface temperature is directly related to the average hot dog temperature
through the voltage divider channel between the capacity-carrying node and the
outer hot dog surface:

T  T1 Ts  T1 T  Ts
¼ ¼
Rk þ Rh Rh Rk
Rh ð T  T 1 Þ Rk ð T  T 1 Þ
Ts ¼ T1 þ ¼T
R k þ Rh Rk þ Rh
While not essential, it is useful to define Biot and Fourier numbers:

Bi ¼ Rk =Rh Gives SPATIAL INFO Fo ¼ t=CRk Gives TEMPORAL INFO

 t 
T  T SS R
= Rk
¼ eð1þBiÞFo
CRk 1þ h Bi
¼e
T 0  T SS
ðT  T 1 Þ BiðT  T 1 Þ
Ts ¼ T1 þ ¼T
Bi þ 1 Bi þ 1
Consider the Biot number limits:
For Bi  1, there is good convection and Ts ! T1, that is, the hot dog cooks on
the edges first, and the center lags behind.
For Bi  1, there is good conduction, and Ts ! Tave, that is, the hot dog cooks
with a uniform temperature.
Evaluating the capacitance is unambiguous:

C ¼ ρcLπD2 =4

Similarly, evaluating the convective resistance is unambiguous because convection

occurs at a single, specific surface:

Rh ¼ 1=ðhπDLÞ

The convection coefficient is considered to be an average value. For the barbequing

case, it is known from experience that the hot dog cooks more on the bottom side.
That effect is probably due to radiation heat coming from the hot coals, an effect
that could be accounted for by a parallel radiation channel acting on the bottom
surface (but from a different temperature than the hot gases flowing past it,
requiring a new RC network be drawn). But think of the hot dog as being turned
frequently during the heating, and the higher convection coefficient value takes the
nonuniform heating effect into account.
Evaluation of the conductive resistance requires some more thought because
conduction occurs between two surfaces that have different surface area. As
suggested by the sketch in Fig. 4.40, the cylindrical hot dog is modeled as an
equivalent wall, with one “surface” (of zero surface area) being the centerline, and
Workshop 4.1. Cooking 1-Node Hot Dogs 155

D
2

πD
.
Q T
T•
D
Rk Rh

Fig. 4.40 Conceptual model for estimating hot dog conductive resistance

Table 4.4 Summary of conduction models for 1-node hot dog problem
Biot Fourier
Model Node placement Rk number number
 
1 r¼0 D=2
hD 4t k
kð0þπDLÞ=2 ¼ πkL
1 k
D2 ρc
  
2 r ¼ D/4 (midway between D=2D=4
¼ 1:5πkL
1 hD 1 4t k 1
πkLðD=2þD=4Þ=2 k 1:5
D2 ρc 1:5
surface and center)
  
3 r ¼ D/3 (centroid of wedge) D=2D=3
¼ 1 hD 1 4t k 1
kðπLD=3þπLD=2Þ=2 2:5πkL k 2:5
D2 ρc 2:5
D=2   
4 r ¼ D/3 (centroid of wedge, ln D=3 ð1:5Þ
hD 1 4t k 1
cylindrical shell) 2πkL ¼ ln2πkL ¼ 4:93πkL
1 k 4:93
D2 ρc 4:93

the other surface being the outer surface of the hot dog, with surface area πDL (with
L the length of the hot dog), a distance D/2 away. The capacity-carrying node is
placed at some radius between the centerline and the surface.
The general form of a conductive resistance is:

xð2
1 dx
Rk ¼
k A
x1

which has the general form of an average distance between surfaces (Δx) divided
by the thermal conductivity, and an average area. A few models (summarized in
Table 4.4) are considered before obtaining results for the base case hot dog.
156 4 1-Node Transient Models

MODEL 1: The most straightforward conduction model (for conductive resistance

between the hot dog surface and the capacity-carrying node) is to “place” the
capacity-carrying node at the centerline of the hot dog (at r ¼ 0), and use the
basic estimate for Rk as the distance between surfaces dividing by the thermal
conductivity times average surface area.
MODEL 2: The centerline in a cylindrical geometry acts like an insulated boundary
because the surface area approaches zero with the radius. Conceptually, heat
conducted from the surface into the interior does not require it to conduct all the
way to the centerline. Therefore, placement of the capacity-carrying node at the
centerline will overestimate the conductive resistance. This model places the
capacity-carrying node at the midpoint between the surface and the centerline
(at r ¼ D/4). The conductive resistance is based on the distance between the surface
and node, and on the average areas. When executed, the functional dependencies
are the same as Model 1, but the Biot number is reduced by a factor of 2/3.
MODEL 3: This model places the capacity-carrying node at the centroid of a wedge
of the hot dog (at r ¼ D/3), rather than at the midpoint between the centerline and
center. The centroid is more representative of the solid than the geometric center.
The Biot number is 2/5 of the value of Model 1.
MODEL 4: The model places the capacity-carrying node at the centroid of a wedge,
but uses the exact formula for conduction between two surfaces at different radial
positions, that is, across a cylindrical shell. The Biot number is reduced by a factor
of 4.93 compared to Model 1.
Model 4 is considered to be conceptually the most accurate for this case. However,
the important take-home message is that all models give the same functional
dependency. Moving forward, methods for breaking the hot dog into multiple
nodes will be developed, and the differences between models for conduction will
be less important.
MODEL 4 results:
Spreadsheet results using conduction Model 4 are summarized in Fig. 4.41 and
plots of the average hot dog temperature and the surface temperature are shown in
Figs. 4.42, 4.43, 4.44, and 4.45. The conductive resistance and capacitance are
independent of the method of cooking and are calculated in the top section (derived
parameters) of a worksheet. This block refers to the input parameter block of the
problem statement. The convective resistance depends on the method of cooking.
Cooking in boiling water results in a large (but not effectively infinite) Biot number
(4.73) event, which means conduction through the hot dog is slow compared to the
rate at which heat is transferred by convection from the water to the surface. There
is a large spatial gradient, with the surface temperature much closer to the fluid
temperature than the core temperature. The other three cases are low Biot number
cases (but not effectively zero), which means that conduction is effective relative to
convection and the hot dog cooks nearly uniformly. Comparing the boiling and
baking cases, the cooking time is faster in boiling water, despite the fact that the
Workshop 4.1. Cooking 1-Node Hot Dogs 157

Fig. 4.41 Results of 1-node hot dog workshop for conduction Model 4

100
90
Surface Temperature
80
70
Temperature (oC)

Cooked Temperature
60
50
40
Average Temperature
30
20
10
0
0 0.5 1 1.5 2 2.5 3
time (min)

Fig. 4.42 Boiling: Biot number ¼ 4.73, Tinf ¼ 100  C

fluid temperature (100  C) is lower than that in an oven (177  C). But at the time the
hot dog is cooked, the surface temperature in water is 95  C (25  C above the hot
dog average, and only 5  C below the fluid temperature) while the surface temper-
ature is 79  C for baking (9  C above the hot dog, and almost 100  C below the fluid
temperature).
158 4 1-Node Transient Models

100
90
Surface Temperature
80
70
Cooked Temperature
Temperature (oC)

60
50
Average Temperature
40
30
20
10
0
0 2 4 6 8 10 12
time (min)

Fig. 4.43 Baking: Biot number ¼ 0.089, Tinf ¼ 177  C

100
90 Surface Temperature
80
70
Temperature (oC)

Cooked Temperature
60
50
40
30 Average Temperature
20
10
0
0 0.2 0.4 0.6 0.8 1 1.2 1.4 1.6 1.8 2
time (min)

Fig. 4.44 Barbeque, Biot number ¼ 0.148, Tinf ¼ 500  C

For barbequing, the cooking time is shorter than boiling, because the fluid
temperature (500  C) is so high. Meanwhile, the surface temperature is 125  C.
In this case, despite the low Biot number, the hot dog appears to heat nonuniformly,
but that is merely because the fluid temperature is so high. The surface temperature
is closer to the average hot dog temperature than it is to the fluid temperature.
Workshop 4.2. Beverage Chilling Methods 159

100
90
80
Cooked Temperature
70
Temperature (oC)

e
tur
60 mper
e
eT tur
e
50 v e rag per
A e m
eT
40 fac
Sur
30
20
10
0
0 0.1 0.2 0.3 0.4 0.5
time (min)

Fig. 4.45 Microwave, Biot number ¼ 0.089, Tinf ¼ 25  C

The fluid temperature chosen for this case is not well characterized. Nevertheless, it
is well known from experience that it is a function of the vertical distance between
the source of the fire and the hot dog. It burns on the outside if placed too close.
Finally, microwaving yields the shortest cooking times (but to me, a hot dog in a
microwave is too rubbery. . .). The equilibrium (steady-state) temperature of
3,200  C represents the temperature the hot dog would achieve in theory if the
microwave were left on indefinitely. Of course, something would happen to the hot
dog to change the model long before it reached that point.
The surface temperature at the microwave cooking time is lower than surface
temperature. That is, heat flows outward from the core of the hot dog to the surface.
Heat is absorbed uniformly throughout the interior, yet the temperature is not
uniform. In order for heat to flow from the interior to the surface, and ultimately
to the ambient, there must be a temperature gradient driving it. From Fig. 4.45, at
the start of the cooking event, the surface temperature is higher than average. That
is because the hot dog is taken from the refrigerator, and it is being warmed by the
air temperature at the same time it is being heated from within. Once the hot dog
average temperature exceeds the air temperature, the core is hotter than the surface.

Workshop 4.2. Beverage Chilling Methods

Beverages are commonly purchased at room temperature, but are consumed cold.
This workshop develops a means to predict the time it takes to chill a room
temperature beverage to a desirable drinking temperature. A standard beverage
chilling time is defined here as the time it takes to chill a beverage from 25 to 5  C
using a specific cooling method. The Cooper Cooler™ is a consumer appliance
developed at the Cooper Union that is designed to do so rapidly. The user prepares
160 4 1-Node Transient Models

an ice/water slurry in a reservoir and the Cooper Cooler sprays a cold recycled jet of
water on a horizontal beverage while is it rotated along its natural axis. For this
workshop, five different chilling methods are considered: three conventional modes
and two Cooper Cooler modes.

1. Refrigerator: The beverage is placed in a standard household refrigerator.

2. Freezer: The beverage is placed in a freezer.
3. Ice/water bath: The beverage is placed in a cooler filled with ice, and enough
water to completely immerse the beverage.
4. Cooper Cooler™ Spray-only Mode: A jet of cold water is sprayed onto a
stationary, horizontal beverage.
5. Cooper Cooler™ Spray and Spin Mode: A jet of cold water is sprayed onto a
rotating horizontal container.

Model Development
All five chilling methods are modeled with the same RC network which the reader
can draw based on the following description. The capacitance of the beverage and
container are lumped into a single node, Tbev. For heat to be transferred from the
beverage to the coolant (at Tcoolant), it must flow from the beverage to the inner wall
of the container (a convective resistance), across the container wall (conductive),
and from the outer wall to the coolant (convective/radiative). In other words, there
is a single thermal channel from the beverage to the coolant with three resistances in
series. The differences in chilling times are due to different values of the various
capacitances (for different beverage types), thermal resistances, and coolant
temperature.
The governing nodal equation is:

dT T coolant  T bev
Cbev ¼
dt Req

where Req is the equivalent thermal resistance (sum of three series resistances). The
solution to this now familiar governing first-order equation is:

T bev  T coolant t
¼ eRCbev
T initial  T coolant
This solution can be used, once estimates of R and C are made, to prepare plots of
beverage temperature vs. time. To determine the standard chilling time (tchill), the
natural logarithm of both sides is taken and the beverage temperature set to the
desired final temperature:


5  T coolant
tchill ¼ RCbev ln
25  T coolant

A spreadsheet template for a specific beverage type is shown in Fig. 4.46. Blocks for
input parameters, derived parameters, and results are shown in which suggested values
Workshop 4.2. Beverage Chilling Methods 161

Fig. 4.46 Spreadsheet templates for beverage chilling workshop

for the case of a 12 oz. bottled beverage are listed. Input values considered to be
common for different beverage types are listed in the results block. Blank boxes are
left for the formulas needed to implement the method. The beverage is considered to
have a clearly identified outer diameter and treated as an equivalent cylinder to obtain
the length. A thin-wall approximation is suggested in which the inner and outer areas
are treated as approximately equal when thermal resistances are calculated.
The coolant temperatures are taken to be those typical of a refrigerator, freezer,
and the melting point of water for the three conventional cooling methods. For the
Cooper Cooling methods, a coolant temperature of 1  C is suggested for spray only,
and 2  C for spray and spin. In practice, the coolant is recycled from an ice/water
bath, and heat added to the coolant from the beverage warms it somewhat, even as
the ice in the bath cools it. The modeling of this process involves more than one
node and is discussed later. The coolant temperature is warmer for the spray and
spin because of the higher heat transfer rates from the beverage.
162 4 1-Node Transient Models

The value for inner heat transfer coefficient is taken to be a typical value for the
natural convection of fluid with nominal water properties (300 W/m2/K) for all
stationary beverages. When the beverage is rotated, there is mixing inside which
produces a sort of wind chill on the inside and the value used (1,500 W/m2/K) is
representative of a mild forced convection with water. The outer convection/radia-
tive coefficient is taken to be representative of natural convection in air, with a
parallel radiation channel, for the refrigerator and freezer. For the ice/water bath,
the outer coefficient is representative of natural convection in water. A value
representative of a forced convection with water (stronger than for the inside) is
listed for the two cases, where a jet of cold water is sprayed.
For the three conventional cooling methods (refrigerator, freezer, ice/water
bath), the coolant is considered to be in contact with the entire surface area of the
beverage (sides, top, and bottom) whereas for the two Cooper Cooling cases, only
the side surface is considered to be in good thermal contact with the coolant.
It is suggested that the case of a 12 oz. canned beverage be compared with the
12 oz. bottled beverage, in which case the only difference is the thermal resistance
of the container wall. Also, the cases of a 0.5, 1, and 2 L plastic bottles be
considered, where both the thermal capacitance and surface area vary. It is also
suggested that plots of temperature vs. time be prepared with all five cooling
methods on the same set of axes for a given beverage. Include the inner and outer
wall temperatures. A final suggestion is to conduct some experiments and compare
results. Notice that in an experiment, the natural dependent variable to set is the
chilling time, and this analysis can help inform an intelligent choice for that time for
a single experiment. Too short a time, and the temperature difference between
initial and final temperature will be small and experimental uncertainty will be
high. Too long a time, and the beverage will be too close to the final equilibrium to
obtain a reasonable dimensionless temperature.
Few-Node Transient Models
5

Try this experiment. Fill a ceramic mug with boiling water and immediately grab its
side. Close your eyes and concentrate on what it feels like. You’ve just experienced
the mug heating phase, the topic of this chapter.
In the cooling mug of coffee problem, the mug is initially at ambient
temperature when hot coffee is poured into it. Figure 5.1 shows the measured
outside wall and coffee-free surface temperatures (as symbols), and the predic-
tion of the 1-node lumped capacity model for the coffee from Chap. 4 (solid
line), the top surface (dotted line), and the outer side wall (dashed line). The
1-node extended model is capable of distinguishing the temperature of the
coffee (think average bulk coffee temperature) from the various surface
locations (two of which are measured). However, it misses the distinction
between the mug and the coffee during the first few minutes. It predicts that
the outside wall of the mug is hot at the instant the coffee is poured. However, it
takes some time for that to happen.
In the 1-node model (coffee and mug as a single node), the capacitance of the
coffee and mug are lumped together, and their temperatures must rise or fall
together. In essence, the initial thermal mixing of coffee and mug is assumed to
take place instantaneously, and the initial temperature for the ensuing cooling
process is obtained from the energy balance detailed in Chap. 2 (in which only
the capacitances of the mug and coffee come into play). The initial period where the
mug rises as the coffee falls cannot be observed.
In Chap. 4, a variable-sized 1-node model was developed that yielded the basic
response of the mug to being suddenly exposed to a hot fluid. However, the bulk
temperature of the fluid was fixed. The initial cooling of the coffee as the mug is
being heated is not captured in that model.
In order to observe the separate behavior of the coffee and the mug, the capaci-
tance of the coffee and the mug must be treated separately, creating a 2-node model.
The goal of this chapter is to develop a general scheme for doing so, and apply it to
the mug of coffee problem.

# Springer International Publishing Switzerland 2015 163

G. Sidebotham, Heat Transfer Modeling: An Inductive Approach,
DOI 10.1007/978-3-319-14514-3_5
164 5 Few-Node Transient Models

100
90
80
70
Temperature (oC)

60
50
Lines Are Theory:
40 solid = coffee
dotted =top surface
30
dashed = outer side wall
20 Markers are Experimental:
open = top surface
10 filled = outer side wall
0
0 2 4 6 8 10
Time (min)

Fig. 5.1 1-Node lumped capacity model predictions during mug heating phase

The term “Few-Node Model” is loosely defined as one in which space is

discretized into clearly identifiable regions that can be given clear names
(as opposed to numbers). In the next chapter, called “multi-node,” space will be
discretized into finer elements in a way that there is no clear sense of how many
nodes there should be beforehand. Once a spreadsheet is set up to solve for a
particular number of nodes, analyzing the same problem but for a different number
of nodes is not possible without setting up a new sheet. For solution approaches that
set the number of nodes as an input parameter, MATLAB or other programming
code can be written that allows the user to simply change this input parameter, and
run the code right away.
This chapter closes with a 3-node analysis of a cooling mug of coffee into which
Joulies™ are placed (http://www.joulies.com), which involves simulating a solid
absorbing heat from the coffee with an inner core material that changes phase.
Coffee Joulies placed into a cup of hot coffee dramatically reduces the time it takes
the coffee to cool to a suitable drinking temperature range, then releases stored
latent energy, holding the coffee in a suitable range for a while. They represent an
engineering solution to the problem of “why do I have to wait to drink my coffee?”

5.1 A Generic 2-Node RC Network

Figure 5.2 shows an RC network with two nodes that have capacitance associated
with them (T1 and T2). Initially, both nodes are at ambient (T1, held by a battery
above ground). The switch (s) is momentarily closed, long enough to charge
capacitor C1 to a starting point (T0), while T2 is initially undisturbed, and then
released. Subsequently, heat flows directly to ambient through R11 and
5.1 A Generic 2-Node RC Network 165

Fig. 5.2 RC network for a 2-node system

simultaneously to the second node (T2), whose capacitor begins to charge. As soon
as T2 rises above ambient (if T0 > T1), heat flows from T2 to ambient. If T0 is less
than T1, then everything flows in the opposite direction. Both nodes can also
receive heat from separate heat sources.
This model represents a more complicated model of the cooling mug of coffee.
The difference between this model and the previous lumped model is that instead of
lumping the capacitance of the mug and coffee together, each component is given
its own capacitance. The direct channel from T1 to T1 (through R11) represents the
heat transfer from the coffee to ambient through the free surface. The channel
between T1 and ambient through T2 represents heat transfer from the coffee to the
mug through the sides and bottom.
The three thermal resistances are shown as equivalent resistances of a combina-
tion of series and parallel channels. Through the use of voltage divider type
calculations, intermediate temperatures can be related to the nodal temperatures
as desired. For example, T2 represents the average temperature of the mug, while
the inner wall temperature is between the coffee and mug temperatures, and the
outer wall is between the mug and ambient. As with the 1-node lumped model
approach, more spatial detail can be gleaned by appropriate selection of thermal
resistances.
In this model, the radiation and convection ambient temperatures are considered
equal. It is relatively easy to modify the RC network to account for the effect of
different fluid and radiation wall temperatures if needed, or to introduce a correc-
tion to either radiation or convection as detailed previously.
166 5 Few-Node Transient Models

5.2 Nodal Equations

To solve this system, a nodal balance for both capacitor-carrying nodes is required.
Figure 5.3 isolates node 1. The energy balance for this node, being careful about
signs, is:
0 1 0 1 0 1
Rate of Change Heat Flow into Heat Flow into
@ of Energy Stored A ¼ @ Node 1 from A þ @ Node 1 A
in Node 1 Heat Source 1 from Node 2
0 1
Heat Flow into
þ@ Node 1 A
from Ambient

dT 1 T2  T1 T1  T1
C1 ¼ Q_ 1 þ þ ð5:1Þ
dt R12 R11
This differential equation cannot be integrated directly because it depends on T2,
which is an unknown variable, and therefore is shown with an open symbol.
Figure 5.4 isolates T2. The energy balance for this node is:

Fig. 5.3 Isolation of node 1

Fig. 5.4 Isolation of node 2

5.2 Nodal Equations 167

0 1 0 1 0 1
Rate of Change Heat Flow into Heat Flow into
@ of Energy Stored A ¼ @ Node 2 from A þ @ Node 2 A
in Node 2 Heat Source 2 from Node 1
0 1
Heat Flow into
þ@ Node 2 A
from Ambient

dT 2 T1  T2 T1  T2
C2 ¼ Q_ 2 þ þ ð5:2Þ
dt R12 R21
Equations (5.1) and (5.2) represent a coupled system of first-order differential
equations. If all resistance and capacitance values are constants, the system is
linear, and there are analytical techniques for solving them (LaPlace transforms,
for example). However, the objective of this text is to build toward more complex
models for which analytic solutions do not exist. Numerical methods, on the other
hand, can treat these more general complex cases, and are developed immediately.
Two-node problems using the explicit method can generally be solved equally well
with spreadsheets or writing code in MATLAB or other programming language.

5.2.1 Explicit Numerical Formulation

For the explicit method, a forward finite difference approximation is used to

approximate the derivatives with respect to time:

ðpþ1Þ ðpÞ
dT 1 T 1  T1
¼
dt Δt pÞ
ð

ðpþ1Þ ðpÞ
dT 2 T 2  T2
¼
dt ΔtðpÞ
Inserting these approximations into the two nodal equations and rearranging them:
 ðpÞ 
ΔtðpÞ ðpÞ T2  T 1 ðpÞ T 1 ðpÞ  T 1 ðpÞ
Q_ 1 þ
ðpþ1Þ ðpÞ
T1 ¼ T1 þ þ ð5:3Þ
C1 ðpÞ R12 ðpÞ R11 ðpÞ

 ðpÞ 
ΔtðpÞ ðpÞ T1  T 2 ðpÞ T 1 ðpÞ  T 2 ðpÞ
Q_ 2 þ
ðpþ1Þ ðpÞ
T2 ¼ T2 þ þ ð5:4Þ
C2 ðpÞ R12 ðpÞ R21 ðpÞ
On the right hand side, all variables and parameters are evaluated at the present
time. A superscript “p” has been applied to everything (time step, R and C values)
indicating that all these may change as time marches forward.
There are two underlying time constants for this system, one for each node, that
represent a first-order response of each node to an equilibrium value if the
168 5 Few-Node Transient Models

neighboring nodes were fixed (detailed in Appendix 1). For example, in Fig. 5.3, if
T2 is held fixed, T1 will approach a value that is intermediate to T2 and T1, with an
equivalent resistance resembling a parallel path. That is, a time constant for
node 1 is:
ðpÞ
ðpÞ ðpÞ C1
τ1 ¼ ðRCÞ1, equiv ¼
1 1
ðpÞ
þ ðpÞ
R12 R11

For node 2:
ðpÞ
ðpÞ ðpÞ C2
τ2 ¼ ðRCÞ2, equiv ¼
1 1
ðpÞ
þ ðpÞ
R12 R21

Note that the heat sources are not associated with the inherent response time. A
Second Law limit for the system is defined by limiting the time step taken at each
step to the smaller of the two time constants. If a larger step size is used, the
temperature will overshoot this “meta-equilibrium” value. As was seen in Chap. 1,
there is also a numerical stability limit for the step size, which, if exceeded, would
cause the system temperature to grow without bounds.
A common misuse of this algorithm is to use a different step size for each node, an
egregious conceptual error. The same step size must be used for both nodes when
marching in time. On the other hand, it is possible for either the capacitance or
especially one or more of the thermal resistors to be temperature and/or time depen-
dent (hence the superscripts “p”), and the response time of one or both of the nodes
will change with time. A variable step-size algorithm can be used in those cases.
It may sound confusing, but “the maximum allowable step size for the method is
the minimum time constant.” That is:
 
ðpÞ ðpÞ
Δtðmax
pÞ
¼ min τ1 ; τ2

In setting up a computer simulation, it is often useful to set a parameter (called “f”

for example), that is, set to be a fraction of the maximum allowable step size,
and therefore setting the time step equal to a fraction of the maximum allowable
step size:
 
ð pÞ ðpÞ
ΔtðpÞ ¼ f  Δtðmax
pÞ
¼ f  min τ1 , τ2

This approach places a constraint on time (the independent variable). In some

applications, it is advantageous to invoke a variable time step algorithm by placing
a constraint on one or both of the temperatures (the dependent variables), and
calculate the required time. One approach is to fix the maximum change in
temperature. Another is to fix the fractional change in temperature.
5.3 Coffee Mug: Explicit Numerical Solution with Constant Coefficients 169

5.3 Coffee Mug: Explicit Numerical Solution with Constant

Coefficients

To apply this general model to the specific case of a cooling mug of coffee, the three
thermal resistances and the two capacitances must be evaluated. Figure 5.5 shows a
more detailed RC network. The two temperatures (coffee and mug, shown as open
circles) represent the main capacity-carrying nodes. However, using voltage divider
formulas, the temperatures of the free surface, inside wall of the mug and outside
wall of the mug can be determined (shown as small filled circles). The free surface
and outside wall surface are the most closely related conceptually to the measured
temperatures to which the model predictions will be compared. Ambient tempera-
ture is a fixed temperature (large filled circle) that could easily be made to vary with
time in a prescribed way.
Convection and radiation act in parallel at the outer surface of the side wall. For
the free surface, an additional parallel channel is shown, namely, an evaporative
channel. As discussed in Chap. 3, evaporation effects are probably dominant when
the coffee is near the boiling point, but negligible thereafter. A single constant heat
transfer coefficient is chosen to include all these effects. The side and bottom
channels from Chap. 3 are treated as a single channel from the coffee to the outside
surface of the mug. There, a side channel and a parallel bottom channel connect to
ambient. More capacity-carrying nodes would be required to further distinguish the
bottom of the mug from the side.

T∞
evaporation

convertion
radiation

Rsa

Tsurface
coffee Rcs mug
Tinside T2 Toutside T∞
T1 convection
radiation
Rcm Rmug Rmug
2 2 Rma, side
C1 C2
Rma, floor
Fig. 5.5 RC network of mug of coffee detailing thermal channels
170 5 Few-Node Transient Models

Detailed derivations of all these individual thermal resistances are included in

Appendix 2. There are many subtle points and different ways to construct an RC
network such as this one, and there could be debate as to how to define specific formulas.
It is suggested that time be taken to define all these individual resistances as a group
in-class and/or homework exercise before consulting the appendix. The strategy
taken here is, given the constraint of only two capacity-carrying nodes, to model the
problem in some detail, and then compare the results with the experimental data. As
will be seen, there is good general agreement, but some discrepancy. The temptation
to adjust input parameter values to obtain better agreement is resisted, as the
modeling methods developed are intended to be applied to situations where experi-
mental data does not exist.

5.3.1 Spreadsheet Program Description

The explicit method was implemented in an Excel workbook. The construction of

this worksheet is described in some detail, as a case study for a general methodol-
ogy for using spreadsheets for this type of problem.
A worksheet named “params” contains a block of input parameter values
(which are input as numbers, not formulas). Figure 5.6 shows this block of
cells. The values listed are those that apply to the experiment detailed in
Chap. 2. The user can change any of these parameter values to simulate other
scenarios. In this implementation, all heat transfer coefficients (convective, radi-
ative, and evaporative) are treated as constants, and are listed as input quantities.
For the parallel channels at the side wall and free surface, a single heat transfer

Fig. 5.6 Input parameter values for 2-node mug of coffee problem
5.3 Coffee Mug: Explicit Numerical Solution with Constant Coefficients 171

coefficient that represents the sum of the individual coefficients is used. The
values chosen are based on estimates evaluated at a coffee temperature of 70  C
from analyses from future chapters.
In constructing a spreadsheet based on a problem statement, there are generally
fixed input parameters, and then many other parameter values (quantities that do not
change with time) that are obtained from formulas based on them. It is a good
practice to keep the fixed values separate from the derived values, especially for
conducting parametric studies (i.e., changing a fixed input to observe its effect on
the system response). It is easy to tell the difference. Any spreadsheet cell that
contains a number is an input parameter. Any cell that contains a formula is a
derived parameter.
In the same worksheet, quantities derived from the input quantities that do not
change with time are shown in Fig. 5.7 in three different blocks. There are
dimensions, capacitances, and surface areas that will be used as inputs to the
individual thermal resistances, and the individual resistances are combined into
series and parallel channels to form the three main equivalent resistances in the
2-node simulation. Detailed explanations of all these calculations are given in
Appendix 2.
The conductive resistance across the mug wall is taken to be between the surface
area of contact between the coffee and the mug and the surface area of the mug
exposed to ambient (including the bottom, through a different mechanism). As was

Fig. 5.7 Derived parameter values for 2-node mug of coffee problem
172 5 Few-Node Transient Models

done with the 1-node model, the heat transfer from the inside surface area of the
mug above the free surface to ambient is included in the side channel (not the top
channel). The conductive resistance across the mug wall could be further improved
by considering a parallel channel through the rim (a future chapter on heat transfer
fins). Either approach could be used in either the 1-node or 2-node model and
highlights some of the art of heat transfer modeling.
An important feature of the detailed RC network is how the total conductive
resistance across the mug wall is split into two, with the mug capacitance placed in
the middle. This placement allows for spatial detail in the temperature field. It also
adds numerical stability, although that may not be clear at this point.
Much can be inferred from a consideration of the relative magnitudes of the
three model resistances. Heat flows easily between the coffee and mug due to the
low value of R12 compared to the other two resistors. Heat leaves the side wall more
easily than it does through the top, despite the higher net coefficient (reflecting the
inclusion of evaporation), a trend attributed to the larger surface area of the
side wall.
The basic structure of the 2-node simulation is shown in the worksheet “2-
nodeMug” (Fig. 5.8). The top block includes the main inputs to the general 2-node
simulation. These cells refer to the appropriate cells in the “params” worksheet,
except for the cell that contains the user-defined parameter “f”, which has the value
0.2 chosen for this implementation, a choice based on experience from Chap. 1 with
the general consideration of the Euler method for a 1-node problem. This numerical
parameter is used to determine the (constant) step size used in the simulation. An
input value of 1.0 would represent the maximum Second Law step size of the system.
The time constants for nodes 1 and 2 are calculated in a separate block below the
inputs. It is apparent that the characteristic response time of the coffee in this case
(241 s) is about eight times as large as that of the mug (35.6 s). An “if-then-else”
statement is used to calculate the step size (Dt ¼ 7.12 s) used in the simulation, which
is the numerical parameter “f” multiplied by the smaller of the two time constants.
The simulation is implemented in the next block. The first row contains the initial
time (0), and initial coffee and mug temperatures (92 and 21). The next row contains
the recursion formulas for time (add the fixed time step to the previous time and
convert from seconds to minutes), coffee temperature and mug temperature (T1 and
T2, from Eqs. (5.3) and (5.4)), which implement the explicit method. Once written,
this second row is copied down as many additional rows are needed to simulate the
desired total elapsed time. For this case, 2,020 rows are needed to simulate 4 h of
elapsed time. If a smaller step size were chosen (by choosing f ¼ 0.1 instead of 0.2,
for instance), that number of rows would simulate less time (2 h instead of 4).
Another worksheet was written to extract additional information used to analyze
the response of the system. The structure of this worksheet (“2NodeMugDetailed”)
is shown in Fig. 5.9. The top block (not shown) is the same as the top of Fig. 5.8, but
several additional quantities are calculated from the two main capacitive-bearing
nodes (coffee and mug) in the simulation. A row of spatial locations (x) is added.
For the two fluids (coffee and air), a thin boundary layer is added for the purposes of
demonstrating the relationship between surface temperature and fluid temperature
5.3 Coffee Mug: Explicit Numerical Solution with Constant Coefficients 173

Fig. 5.8 Structure of

worksheet that implements
the fixed step-size explicit
numerical method (Euler
method)

Fig. 5.9 Structure of worksheet in which additional detail is extracted (i.e., intermediate
temperatures and heat transfer rates)
174 5 Few-Node Transient Models

100

Top
Coffee
90
Top
80 Side
70 Top Data
Temperature (oC)

Side Data Side

60 coffee
50 mug
40
30
20
10
0
0 30 60 90 120 150 180 210 240
Time (minutes)

Fig. 5.10 2-Node model prediction for entire cooling event on linear axes

across a convective thermal boundary layer. Columns for the inner and outer mug
surface temperatures are added and the voltage divider formulas implemented.
A column for the free surface temperature is added. Finally, the individual heat
transfer rates between nodes are calculated (individual terms from governing
equations).

5.3.2 Model Prediction Results

Figures 5.10 (linear axes) and 5.11 (semi-logx axes) show the temperature vs. time
predictions from the 2-node model (shown as lines) compared to the experimental
data (shown as markers) for the entire cooling event using the parameter values
from the previous section. Figure 5.12 focuses on the mug heating phase (early
time) on linear axes. The line for coffee (solid line) shows node 1 at a slightly higher
temperature than the surface and outer wall. The dotted line which corresponds with
the filled circles shows the free surface temperature. The dashed line which
corresponds with the open circles shows the outer mug wall temperature. The
mug temperature itself (node 2) is not shown in these plots, but only the two non-
capacity-carrying nodes that correspond to the measured quantities.
The linear axes (Fig. 5.10) show the entire event, and the initial mug heating
period is not clearly visible. The coffee, top (free) surface of the coffee, and mug
side wall all fall together. The model prediction lags the experimental for the first
hour or so, but then the final approach to equilibrium gives good agreement
between theory and experiment. However, this model assumes constant
coefficients, and the effect of a variable coefficient cannot be seen. Recall that
the constant values chosen were based on a temperature of 70  C, and more
5.3 Coffee Mug: Explicit Numerical Solution with Constant Coefficients 175

100
90 Coffee
80 Top
Side
70
Temperature (oC)

Top Data
60 Side Data
50
40
30
20
10
0
0.1 1 10 100 1000
Time (minutes)

Fig. 5.11 2-Node model prediction for entire cooling event on semi-logx axes

100
90
80
70
Temperature (oC)

60
50 Coffee
40 Top
Side
30
Top Data
20 Side Data
10
0
0 1 2 3 4 5
Time (minutes)

Fig. 5.12 2-Node prediction for the cooling mug of coffee for the first 5 min

sophisticated models would show the resistance values increasing with time
(as the temperature falls).
The semi-logx axes (Fig. 5.11) show the entire event, but the behavior during the
initial mug heating phase is apparent in this view as well. The focus on the first
5 min of the event on linear axes (Fig. 5.12) shows the undistorted shape of the
temperature response, with the same trends observed on the semi-logx scale. In the
first minute, the model predicts a slower rise of the outer side wall than measured,
and then excellent agreement between 1 and 10 min. The predicted top (free)
176 5 Few-Node Transient Models

1000
Coffee to mug
Mug to Ambient
Heat Transfer Rate (Watts) 100

10

Coffee to Ambient (top)

1

0.1

0.01
0.1 1 10 100
Time (min)

Fig. 5.13 Heat transfer rates between nodes for the entire event (log–log axes)

surface of the coffee is noticeably higher than the experimental value throughout.
Use of a higher value for thermal resistance between the coffee and the free surface
(Rcs) would give better agreement (but not the “dip” in the data, which may or may
not be a repeatable trend, and only a single experiment was conducted). Inclusion of
this resistance represents a deviation from the classic modes of heat transfer
because the free surface represents a boundary between two immiscible fluids,
not between a fluid and a solid. Nevertheless, application of Newton’s Law of
Cooling with a convection coefficient as if the fluid/fluid boundary behaved as a
fluid/solid boundary (on either side) is a plausible modeling tool to reflect a
mechanism for heat transfer from a bulk fluid to an interface.
The heat transfer rates between nodes are shown as a function of time in
Fig. 5.13. During the mug heating phase (the first minute), the rate of heat transfer
from the coffee to the mug is very high. The heat transfer rate from the mug to
ambient rises as the outside wall temperature rises during the first minute or two.
After that, the rate of heat transfer from the mug to ambient is slightly higher than
from the coffee to the mug (the difference is the change in energy stored in the mug
wall). The rate of heat loss through the top surface is always small, and this effect
can be attributed to the smaller surface area, despite the higher effective heat
transfer coefficient value input there. The heat transfer rate between the side and
bottom channel are not distinguished, but the thermal resistance of the bottom
channel is much larger than that of the side wall (due to smaller area and lower
effective heat transfer coefficient).
A main point of presenting this heat transfer plot is to demonstrate how much
more information can be gleaned from a theoretical analysis of a problem (if one
chooses to look for it) compared to experiment. The experiment has the advantage
of being a measurement (keeping experimental errors always in mind), but
5.3 Coffee Mug: Explicit Numerical Solution with Constant Coefficients 177

100 Time (min)

90 0.952
Time 0.846
80 0.740
Time
70 0.634
Temperature (oC)

0.529
60
0.423
50 0.317
40 0.211
0.106
30 0.000
20
Coffee Mug Wall Ambient
10
0
3.9 4 4.1 4.2 4.3 4.4 4.5 4.6 4.7
Radial Position (cm, from centerline)

Fig. 5.14 Temperature as a function of side wall position for the first minute of the simulation
(mug heating phase)

generally limited information. Theory and experiment make a great team. For
example, an IR detector was used that can only yield surface temperatures, and
the model shows the side and surface temperatures with excellent agreement. This
experiment does not show that the bulk of the coffee is actually slightly higher the
temperature at the free surface and at the mug surface. A thermocouple array might
show that detail (a great project for a course on experimentation). The heat transfer
rates cannot be measured directly at all.
The spatial temperature variation in the side wall as time elapses is demonstrated
in Figs. 5.14 (for the mug heating phase) and 5.15 (for the coffee/mug cooling
phase), which show temperature as a function of radial position at fixed times. The
fluid temperatures (coffee and ambient air) are shown with somewhat arbitrary
0.05 cm thermal boundary layer. Straight lines are drawn between points with
markers that indicate specific locations from the model. Arrows are used to indicate
the trends with time.
The first plot (Fig. 5.14) shows these temperatures for the first ten discrete times
of the simulation which focuses on the first minute of the event. At the initial time,
where the coffee and mug temperatures are set to their initial values (92  C and
21  C, respectively), the model predicts (through the voltage divider calculation)
that the initial temperature on the inner wall surface does not equal the initial mug
temperature, but is between the mug and the coffee. Its value (49.95 from Fig. 5.9)
is roughly halfway between the mug and coffee temperatures because the convec-
tive resistance between the coffee and mug (0.0844  C/W) is coincidentally of the
same order as that of the conductive resistance across the mug wall (0.116/
2 ¼ 0.0553  C/W). After one step in time, the mug temperature rises to about
35  C and the outer surface wall temperature rises with it. The outer surface
178 5 Few-Node Transient Models

100
Time (min)
90
1.069
80 15.0
70 30.1
Temperature (oC)

45.0
60
60.0
50
75.0
40
30
Time
20
10 Coffee Mug Wall Ambient

0
3.9 4 4.1 4.2 4.3 4.4 4.5 4.6 4.7
Radial Position (cm, from centerline)

Fig. 5.15 Temperature as a function of side wall position at 15 min time intervals (after initial
mug heating)

temperature is between the mug and ambient, but is much closer to the mug because
the convective resistance to ambient (2.06  C/W) is much larger than the conduc-
tive resistance across the mug (0.0553  C/W), a Biot number concept. During this
time, the inner wall temperature rises about 5  C as the coffee temperature drops
about 1  C. Each step in time reveals a similar qualitative change, but with reduced
magnitude temperature jumps.
Experience from the variable-sized 1-node model from Chap. 4 would suggest
that it takes approximately 4 s for the initial thermal wave to traverse the mug wall,
and that time is short compared to the step size used in this simulation. Therefore,
the use of a fixed size mug node is justified.
The concavity in the temperature profile across the mug is an indication of the
degree to which it is important to assign separate capacitance to the mug
(as opposed to lumping it together with the coffee). If heat transfer rate from
the inner wall to the mug exceeds that from the mug to the outer wall, the
temperature profile will be concave up. The difference is the rate of change of
the energy stored in the mug. After 1 min, the concavity essentially vanishes (the
temperature is approximately linear across the mug wall), meaning that the NET
heat transfer rate to the mug is negligible. They are not exactly equal, as shown in
Fig. 5.13. A high heat transfer rate enters from the inside and exits the outside at
almost exactly the same rate. This behavior is an example of “quasi-steady state”
behavior, and why it is justified to lump the capacitance of the mug and coffee
together in order to simulate the response AFTER the mug attains this quasi-
equilibrium. The heat transfer rate at any time (after 1 min) is the same as it would
be if the slowly changing coffee temperature were maintained by some means at a
constant value. Recalling Fig. 5.13, these trends are also made apparent by
5.3 Coffee Mug: Explicit Numerical Solution with Constant Coefficients 179

considering the heat transfer rates between the coffee and mug, and between the
mug and ambient.
Once this quasi-steady state equilibrium for the mug is established, the tempera-
ture remains approximately linear across the mug as the coffee and mug cool
(Fig. 5.15). During this time, the coffee and mug respond together, and this
2-node model provides justification for lumping their capacitance together during
the cooling phase. However, the 1-node lumped model is not capable of revealing
the key physics during the mug heating phase. In this particular case, the coffee/
mug cooling phase dominates the entire event, and therefore a 1-node model does a
good job of simulating the main event, while missing detail at the beginning.
A really good problem statement would be to conduct a similar experiment
and analysis with a Styrofoam cup. The main difference would be a lower
mug capacitance and a higher conductive resistance across the mug wall. The
mug outer side wall would be much lower than the coffee-free surface in this case.
Keeping the same geometry, the mug mass (from a measurement), specific heat
and the mug thermal conductivity (from a literature search) could be changed to
their appropriate values. The time constant for the mug will be much smaller
than the ceramic mug, and this problem is becoming numerically “stiff” where
the time step restriction on a fast responding node requires many more time
steps to simulate the overall response than would be required for the slower
responding node.

5.3.3 Numerical Accuracy and Stability Issues

An important distinction to be made in evaluating heat transfer models is between

the physics of the model and the mathematics of its simulation. The discussion of
the previous section was entirely about the physics of the model, and it was
assumed (without formal statement) that the plots presented were from an accurate
implementation of the model. An inherent challenge and potential danger of
numerical analysis is generating output that “looks right” but may in fact be
unacceptably and unnecessarily inaccurate.
Figure 5.16 shows the effect of step size used in the simulation defined as a
fraction (f) of the maximum value allowed by the Second Law of Thermodynam-
ics. Three plots show the predicted coffee and mug temperatures for different
values of “f”. The bottom right compares the mug temperature directly for these
cases, plus an additional two values. Symbols with straight lines are used to
indicate the discretization of time. A value f ¼ 1 (bottom left) shows the mug
jumping up and overshooting the coffee temperature, which at face value, appears
to violate the Second Law. However, the value f ¼ 1 represents a step size in
which the mug will come to thermal equilibrium with its neighbors (coffee and
ambient) based on their present values, not their future values. Values of f ¼ 0.1
and 0.5 reveal similar qualitative predictions, with the smaller value yielding
smoother behavior, at the expense of more calculations. However, when
180 5 Few-Node Transient Models

100 100
90 f = 0.1 90 f = 0.5
80 80
70 70
Temperature (oC)

Temperature (oC)
60 60
50 50
Coffee Coffee
40 40
Mug Mug
30 30
20 20
10 10
0 0
0 0.5 1 1.5 2 2.5 3 0 0.5 1 1.5 2 2.5 3
Time (min) Time (min)

100 100
90 f=1 90
80 80
70 70
Temperature (oC)

Temperature (oC)
60 60
50 50
Coffee
40 40 f = 1.773
Mug
30 30 f=1
f = 0.5
20 20
f = 0.2
10 10 f = 0.1
0 0
0 0.5 1 1.5 2 2.5 3 0 0.5 1 1.5 2 2.5 3
Time (min) Time (min)

Fig. 5.16 Effect of step size (f ¼ fraction of maximum Second Law step size) during the mug
heating phase. The bottom right graph compares the mug temperature for 5 different values of the
step size (as a fraction, f, of the maximum allowable)

compared directly (bottom right), the f ¼ 0.5 case shows a noticeable difference
from f ¼ 0.1. An intermediate value (f ¼ 0.2) is included which shows good
agreement with f ¼ 0.1.
This analysis shows a somewhat informal or brute force test for “grid insensitiv-
ity.” To accurately simulate a given mathematical model numerically, a typical
Question and Answer series might be

Q1. “What step size should be used?”

A1. “As small as it needs to be to accurately solve the mathematical problem?”
Q2. “How do you know what’s small enough?”
A2. “When the prediction of two values of step size agree acceptably, the bigger
one is small enough”

In this case, a value f ¼ 0.2 yields a good compromise between computational

expense and numerical accuracy. More formal test methods can be developed,
especially as the models become spatially more complex.
Notice for this system that even the case f ¼ 1 yields excellent agreement with
the smaller step-size cases at the end of the mug heating phase. So that choice
is acceptable from the point of view of the simulation of the entire cooling event.
5.3 Coffee Mug: Explicit Numerical Solution with Constant Coefficients 181

200
f = 1.773
150

100
Temperature (oC)

50

0
0 10 20 30 40 50 60
-50

-100 Coffee
Mug
-150

-200
Time (min)

Fig. 5.17 Coffee and mug temperatures for the step size just above a critical value for numerical
stability

It just misses the detail at the beginning. It can be argued that the reason for that
agreement is because it merely finds the right quasi-equilibrium in a few steps.
For a problem for which that quasi-equilibrium is not a good model, all bets
are off.
Numerical stability limits are demonstrated in Fig. 5.17, which shows the mug
and coffee temperatures for an hour with a value f ¼ 1.773. This value (also one of
the cases in the bottom right plot of Fig. 5.16) was chosen by trial and error, and is
considered near, but above the stability limit of this explicit formulation. The mug
temperature would eventually grow without bounds (numerical instability) if
allowed to continue.

5.3.4 Implicit Numerical Formulation

There may come a time when the numerical stability limit of an explicit formulation
becomes so restrictive as to make it untenable. For example, in a 2-node system, if
the response time constant of one node is 1,000 times smaller than that of the other,
a minimum of 1,000 time steps would need to be taken just to simulate an amount
of time corresponding to one time constant of the slower responding node. An
implicit method represents another numerical tool that overcomes this problem. It
is introduced in this section, but not implemented, as the explicit method is
considered adequate for this particular job. This method is not readily solved
using a spreadsheet.
As was demonstrated in Chap. 1, a backward difference for the time derivative
brings a quickly responding node to equilibrium if an excessive step size is used. In
182 5 Few-Node Transient Models

this case, the two backward difference approximations (for nodes 1 and 2, coffee
and mug) are:

ðpÞ ðp1Þ
dT 1 T 1  T 1
¼
dt ΔtðpÞ
ðpÞ ðp1Þ
dT 2 T 2  T 2
¼
dt ΔtðpÞ
Updating present to future (p ! p + 1), inserting these approximations into the two
nodal equations yields:
!
ðpþ1Þ ðpÞ ðpþ1Þ
T1  T1 ðpþ1Þ T2  T 1 ðpþ!Þ T 1 ðpþ1Þ  T 1 ðpþ1Þ
C1 ðpþ1Þ ¼ Q_ 1 þ þ
Δtðpþ1Þ R12 ðpþ1Þ R11 ðpþ1Þ

!
ðpþ1Þ ðpÞ
T2  T2 ðpþ1Þ T 1 ðpþ1Þ  T 2 ðpþ!Þ T 1 ðpþ1Þ  T 2 ðpþ1Þ
C2 ðpþ1Þ
¼ Q_ 2 þ þ
Δtðpþ1Þ R12 ðpþ1Þ R21 ðpþ1Þ

All temperature variables and parameters with a superscript “p” are evaluated at the
present time. If the initial condition is known, these are considered to be known
quantities. A superscript “p + 1” is a future value, and is unknown. If the numerical
time step (Δt), R and C parameters are functions are variable, their (unknown)
future value is required. The two nodal equations constitute a system of two
algebraic equations for the future values of T1 and T2.
If the parameters are constants, it is a linear system, and can be solved using
linear algebra. The two governing equations can be written in matrix form for the
two future temperatures, and then solved using Cramer’s rule or Gaussian elimina-
tion, or easily inverted in a MATLAB program.
If the parameters are variables, the system is nonlinear, and an iterative numeri-
cal technique is required (as opposed to a “stepping” technique as with time). The
Jacobi Method (Successive Substitution) and closely related Gauss–Seidel method
are powerful methods, and involve rearranging the governing equations by “solving”
for each nodal future value in terms of its neighbors, making an initial guess, and
iterating until either convergence or failure of the method. If the method fails, then
Newton’s method is a good technique to try next. The recursion equations in the form
for Gauss–Seidel are:
5.4 Coffee JouliesTM: A 3-Node Model 183

ðpþ1Þ ðpþ1Þ ðpÞ

ðpþ1Þ T ðpþ1Þ T 2 C T
Q_ 1 þ 1ðpþ1Þ
þ ðpþ1Þ þ 1 ðpþ11Þ
ðpþ1Þ R11 R12 Δt
T1 ¼ ðpþ1Þ
1 1 C1
ðpþ1Þ
þ ðpþ1Þ
þ
R11 R12 Δtðpþ1Þ

ðpþ1Þ
ðpþ1Þ T ðpþ1Þ T 1 ðpþ1Þ ðpÞ
Q_ 2 þ 1
C2 T2
ðpþ1Þ
þ ðpþ1Þ þ Δt ðpþ1Þ
ðpþ1Þ R21 R12
T2 ¼ ðpþ1Þ
1 1 C2
ðpþ1Þ
þ ðpþ1Þ
þ
R21 R12 Δtðpþ1Þ

Application of the implicit method, then, involves solving a system of algebraic

equations for each step in time taken. It is that extra work, requiring a nested
loop (which cannot be done in a spreadsheet without using macros) that makes it
more difficult to implement than the explicit method. It can be advantageous
(or necessary) to use an implicit method if the value of the two time constants
differ by orders of magnitude, in which case the problem is said to be “stiff.” To test
for stiffness, the ratio of the two time constants can be evaluated:
2 3
R
  1 þ 12
τ1 C1 6
6 R21 7
7
¼ 4
τ2 C2 R12 5
1þ
R11
If this ratio is less than, say 0.001, then the response time of node 1 is much shorter
than that of 2. In this case, the response of the system is controlled by the slower
moving 2, and node 1 will be in a quasi-equilibrium between 2 and the environment.
On the other hand, if the ratio is greater than, say 1,000, then node 2 quickly
responds, and the system is controlled by node 1.
Alternatively, if a 2-node problem is shown to be stiff, the original model can be
simplified by appropriately lumping the two capacitances together, and thereby
reducing the problem to a 1-node problem. Detail is compromised, and if that detail
would help make engineering decisions, then lumping would not be a good simpli-
fication of the model. The coffee/mug problem is a case is point. Lumping the
coffee and mug prevents investigation of the mug heating phase, but does not
significantly affect the overall cooling behavior.

5.4 Coffee JouliesTM: A 3-Node Model

Coffee Joulies represent an engineering solution whose objective is to shorten the

time before pouring hot coffee and being able to drink it, and lengthen the time
coffee stays in an acceptable drinking range. Joulies fit comfortably in your hand,
and can be placed into a hot mug of coffee, or placed in a mug, and then coffee
184 5 Few-Node Transient Models

T∞

evaporation

convection
radiation
Rsa

Tsurface mug
coffee
Rcs
Tjoulies T1 Tinside T2 T T∞
outside convection
radiation
RCJ Rcm Rmug Rmug
2 2
Rma
Cjoulies C1 C2

Fig. 5.18 3-Node model with Coffee Joulies in a single-stage phase of heating or cooling

poured over them. They are constructed with a shell of stainless steel, and filled
with a “phase change material” that is called “wax” in this text. The requirement of
the wax is that its melting point temperature be within the suitable temperature
range. The Joulies store energy in the form of “latent heat.”
To model the cooling of coffee with Joulies, a 3-node model consisting of coffee,
mug, and Joulies captures the key effects (Fig. 5.18). The coffee and mug are the
same as the 2-node model, except that the coffee has an additional thermal partner,
namely, the Joulies. The Joulies are treated as a lumped element. During single
phase stages (i.e., completely solid or liquid), they are modeled with a capacitance,
and a thermal resistance (RCJ) that consists of three resistances in series: conduction
through the wax, conduction across the shell, and convection between the Joulie
and the coffee.
When undergoing phase change (Fig. 5.19), the temperature remains fixed at the
melting point, but the fraction of wax that is liquid and the fraction that is solid
changes. There is a “quality” (x) of the solid/liquid mixture that changes. The
quality, defined as the mass fraction of liquid, is analogous to the quality used in
liquid/vapor phase change in traditional thermodynamic analysis. The quality is
only defined during phase change, when the Joulie contains a mixture of saturated
solid (further heating would cause melting) and saturated liquid (cooling would
cause freezing) in equilibrium. If the Joulie is below the melting point, the solid is
not saturated and the quality is not applicable (x ¼ NA). Similarly, for temperatures
above the melting point the liquid is not saturated (and x ¼ NA).
The Coffee Joulie node is represented in Fig. 5.19 as a “rechargeable battery”
given the symbol of a battery with an arrow through it. The node is filled, indicating
that the temperature of the node is fixed at the melting point of the wax, but the
arrow indicates that the degree of battery “charge” varies. When the wax is
5.4 Coffee JouliesTM: A 3-Node Model 185

T∞

evaporation

convection
radiation
Rsa

Tsurface mug
coffee
Tjoulies, MP Rcs T2
T1 Tinside T outside convection
T∞
radiation
RCJ Rcm Rmug
2
Rmug Rma
C1 2
C2

Fig. 5.19 3-Node model with Coffee Joulies undergoing phase change (melting or freezing)

completely solid, the battery is discharged, and when liquid it is fully charged. The
quality “x” monitors the degree of charge.
When placed in sufficiently hot coffee, the Joulies will go through five distinct
stages.

1. Heating stage: The Joulies will be heated (by the coffee) to the melting point of
the wax. Meanwhile the coffee is also heating the mug wall and is being cooled
itself. If the Joulies reach the melting point with the coffee still above it, the wax
will become a saturated solid and stage 2 will begin.
2. Melting stage: The Joulies will continue to absorb heat from the coffee but
remain at the melting point temperature of the wax. The quality (mass fraction of
vapor) will increase until all the solid is melted, and the wax is a saturated liquid.
If melting is complete, the coffee temperature remains above the melting point,
and stage 3 will begin.
3. Superheating stage: The temperature of the Joulies rises above the melting point,
and will tend to approach the coffee temperature, then fall back with it as the
coffee cools. When the Joulies cool back to the melting point, the liquid will be
saturated and stage 4 will begin.
4. Freezing stage: The coffee temperature drops below the Joulies and the wax
begins to freeze. Heat is now transferred from the Joulies to the coffee, helping
delay the coffee from cooling below a suitable drinking temperature. Once the
wax has completely solidified, the final stage begins.
5. Cooling stage: The coffee, mug, and Joulies cool together, eventually attaining
thermal equilibrium with ambient. Unless the coffee has been drunk by now. . .
186 5 Few-Node Transient Models

5.4.1 Incorporating a Node with Phase Change

The general nodal equation for the Coffee Joulie(s) is:

dEJoulie T coffee  T Joulie

¼
dt RCJ
Table 5.1 summarizes the model for the energy stored in the Joulie and its rate of
change. The Joulie is considered to be constructed of a central spherical core of
wax, and a spherical shell of steel. In this lumped model, the temperature of the wax
and shell are not necessarily equal, but they rise and fall together.
If the Joulie is below the melting point (TMP), the wax is completely solid, and its
internal energy, relative to a reference temperature (Tref, below the melting temper-
ature) has been handled previously. The term “quality” hasno relevance (x ¼ NA). 
The total capacitance (sum of wax plus shell) is Csolid ¼ ðmcÞshell þ ðmcÞwax, s .
There is a similar expression for when the wax is completely liquid, the third row in
Table 5.1, except the liquid wax could have a different specific heat than the solid.
In the energy expression for single phases, only the temperature changes.
If the temperature equals the melting temperature and it is completely solid, the wax
in the Joulie has properties of a saturated solid, and the quality has a value of 0.
The internal energy
(relative to the reference temperature) is us ¼
Csolid T MP  T ref . If the wax is at the melting point and is completely liquid,
the wax has properties of a saturated liquid, and the quality has a value of 1.0.
If the internal energy is between a saturated solid and liquid, the quality has a
value between 0 and 1, and the internal energy of the wax can be expressed as
Ewax ¼ mwax ½ð1  xÞus þ xuL  ¼ mwax ½us þ xuLS  where the heat of fusion is
uLS ¼ uL  uS .
In the phase changing region, the only variable is the quality, so the rate of
change of energy depends on the heat of fusion and the rate of change of quality.

Table 5.1 Joulie energy, relative to a reference temperature below the melting point, stored in a
Joulie and its rate of change for different temperature ranges
Temperature dEJoulie
range EJoulie ¼ Eshell þ Ewax dt
  
TJoulie < TMP ðmcÞshell þ ðmcÞwax, s T Joulie  T ref ¼ Csolid T Joulie  T ref dT Joulie
Csolid
 dt
TJoulie ¼ TMP ðmcÞshell T MP  T ref þ mwax ½us þ xuLS  dx
mwax uLS
j k dt
TJoulie > TMP ðmcÞshell þ ðmcÞwax, l ðT Joulie  T MP Þ þ mwax uL Cliquid dTdt
Joulie
5.4 Coffee JouliesTM: A 3-Node Model 187

5.4.2 Nodal Equations

The nodal equations (in the form for the Euler method) are therefore:

ðpÞ
!
ðpÞ ðpÞ
ΔtðpÞ T 1  T mug T coffee  T mug
T ðmug
pþ1Þ
¼ T ðmug
pÞ
þ þ
Cmug RM1 RCM

ðpÞ ðpÞ ðpÞ ðpÞ

!
ðpþ1Þ ðpÞ ΔtðpÞ T 1  T coffee T ðmug
pÞ
 T coffee T Joulie  T coffee
T coffee ¼ T coffee þ þ þ
Ccoffee RC1 RCM RCJ

The nodal equation for the Joulies depends on whether it is in the single or mixed
phase. When in the single phase (solid or liquid):

ðpÞ ðpÞ
!
ðpþ1Þ ðpÞ ΔtðpÞ T coffee  T Joulie
T Joulie ¼ T Joulie þ
CJoulie RCJ

When in the mixed phase, the temperature of the Joulies is fixed at the melting
point, and an Euler formula for the quality is:

ðpÞ
!
ΔtðpÞ T coffee  T MP
xðpþ1Þ ¼ xðpÞ þ
mwax uLS RCJ

As always with the Euler method, there is a maximum time step for numerical
stability that must be carefully monitored.
A variable time step algorithm is used during the melting and freezing stages, as
indicated by a superscript (p) for the time step. Rather than input a time step to
march, a step in quality (Δx) is made. Rearranging the governing equation for
quality gives the required time step (which must be less than the maximum time
step for the coffee and mug).

Δxm u R

ðpÞ
wax LS CJ

Δt ¼
ðpÞ

coffee  T MP

Absolute values are used to prevent negative time steps.

All the resistance values are equivalent resistances. A detail for the resistor
between the Joulies and the coffee (RCJ) is shown in Fig. 5.20, which consists of
three thermal resistances in series. For the purposes of modeling, the Joulies are
treated as spherical, with a diameter that reproduces the total volume of a
manufactured Joulie, and an average density equal to the total mass divided by the
total density (detailed in Appendix 3). The actual shape is not spherical so the surface
188 5 Few-Node Transient Models

Fig. 5.20 Detail of RCJ, the thermal resistance between the Joulies and coffee

to volume ratio of a manufactured Joulie is different than the one used in this
simulation. The total mass (mass of the wax plus the mass of the shell) and volume
are easily measured, and the equivalent shell thickness can be estimated given
assumed density values. The mass-averaged specific heat is similarly estimated.
The outer resistance is modeled as a convective resistance:

1
Ro ¼ 
ho πd2

An outer convection coefficient typical of a natural convection in water is assumed,

acting on the outer surface area (4πr2 ¼ πd2) of the Joulie.
The resistance across the shell wall is a conductive resistance. Using a thin-wall
approximation:

ðd  d wax Þ=2
Rw ¼ 
kshell πd 2

The thermal conductivity of the shell is required.

The inner conductive resistance (between the inner shell surface and the wax) is
intriguing and a good heat transfer modeling case study. A first approach is to
assume a pure conduction during the Joulie heating stage, in which case a first
approximation would be:

d=4
Rw ¼  2 ða first approximation, not used in simulationÞ
kwax πdwax =2

Here, the numerator is the average distance between the center and the inner
shell surface, and the denominator is taken to be the average area of those surfaces.
5.4 Coffee JouliesTM: A 3-Node Model 189

When evaluated (Appendix 3), the value of this resistance is much larger than the
two others in this series channel. Therefore, in practice, the shell temperature will
change first, and the surface of the wax will follow, then quickly enter into a stage
where the wax melts. This behavior requires the shell and wax to be treated as
separate nodes. However, that approach will not be pursued here (a good workshop
assignment!) in the interest of gaining a reasonable prediction of the system
response with Joulies. The modeling compromise is to treat the inner thermal
resistance as a convective resistance:

1
Ri ¼ 
hi πd 2wax

An inner convection coefficient typical of natural convection with a liquid is input.

Actually, during phase change, the effective heat transfer coefficient would be
much larger than that, as convection coefficients during phase change tend to be
larger than single phase convection due to latent heat effects. During those periods,
the total resistance would be controlled by the sum of the shell and outer
resistances.

5.4.3 Spreadsheet Implementation

The worksheet used to calculate various parameter values that are inputs to the
simulation is shown in Fig. 5.21. Many of these parameters were determined from a
separate spreadsheet detailed in Appendix 3 based on basic measurements (volume
and mass) and assumptions about a single Joulie. A total of 3 Joulies are added to
the same mug. The mass of the wax (40.6 g for 3 Joulies) is approximately half of
the total Joulie mass (86 g), the remainder being the shell. The 3 Joulies displace
62 mL of volume. Note that the capacitance of the Joulies during the heating phase
(Csolid ¼ 123 J/K) is less than half that of the mug (Cmug ¼ 268 J/K), so it is expected
that the main effect of the Joulies on coffee temperature will be due to phase change
(latent heat), not the sensible energy effect that adding more thermal mass would have.
Five different “simulation” spreadsheets are written, one each for each stage of
the event (Joulie heating, melting, superheating, freezing, cooling), and these sheets
are linked so that the time and temperatures at the end of one stage are referenced as
the initial condition of the subsequent stage. The structures of two of these sheets
are shown in Figs. 5.22 and 5.23 for the heating and melting stages, respectively.
Each sheet marches forward a total of 100 steps (arbitrarily chosen, and not tested to
yield a time step smaller than that required of the other two nodes, mug and coffee).
During heating (Fig. 5.22), the quality is set to a value “NA,” and the Joulie
temperature varies. The numerical simulation marches in time, with a fraction of
the maximum allowable step size (f ¼ 0.0218. . ., small enough for excellent Euler
accuracy) determined using “goal seek” to the value that results in the temperature
190 5 Few-Node Transient Models

INPUT PARAMETERS
n 3 Number of Joulies
Djoulie 0.034 m Diameter of Joulie
rhojoulie 1360 kg/m3 density of Joulie
csolid 1461 J/kg/K specific heat of solid Joulie
cliquid 1461 J/kg/K specific heat of liquid Joulie
rhocoffee 1000 kg/m3 coffee density
ccoffee 4200 J/kg/K coffee specific heat
uLS 210000 J/kg heat of fusion of Joulie
Tmp 60 C melting point of Joulie
Tinf 25 C ambient
Vtotal 4.10E-04 m^3 total volume (coffee plus Joulies)

DERIVED
Ajoulie 0.0109 m^2 total surface area of Joulies
Vjoulie 6.17E-05 m^3 total volume of Joulies
mjoulie 0.0840 kg total joulie mass
mwax 0.0406 kg mass of wax
Esat,Solid 4294 J Joulie energy at saturated solid
Esat,Liquid 21926 J Joulie energy at saturated liquid
Coffee Volume 3.48E-04 m^3 coffee volume

RC VALUES
Rcj 0.86 oC/W bet. Coffee and Joulie
Rcm 0.15 bet Coffee and mug
Rminf 1.74 bet. Mug and ambient
Rcinf 6.63 bet. Coffee and ambient
Cmug 268 J/K mug capacitance
Ccoffee 1463 J/K coffee capacitance
Cjoulie,solid 123 J/K Joulie solid capacitance
Cjoulie,liquid 123 J/K Joulie liquid capacitance

Fig. 5.21 Worksheet of input parameters for the 3-node simulation

at the last step equal to the melting point. The maximum step size (inherent time
constant) is controlled by the mug (dtmax,mug ¼ 37.2 s) which is smaller than the
time constant for the Joulies (dtmax,Joulies ¼ 105.2 s). Even though the capaci-
tance of the Joulies is half that of the mug, the thermal resistance is almost six times
as large between the Joulies and coffee (RCJ ¼ 0.86  C/W) as it is between the
coffee and the mug (RCM ¼ 0.15  C/W). This effect is one of contact area.
During melting, the Joulie temperature is set to the melting temperature and the
quality varies with time. The time step (dt) varies according to the governing
equation for quality and is confirmed to be lower than the inherent time constant
of the mug at all times.
This workbook is not completely automatic. Any time a parameter value is
changed, it is necessary to adjust the time step (through “f”) at each stage to
transition from stage to stage.
JOULIE HEATING PHASE VALID SIMULATION
dtmax,Joulie 105.1924716 sec
dtmax,Coffee 184.0242627 sec
dtmax,mug 37.18518815 sec
f= 0.021782786 fraction of max step size
dt 0.80999699 sec
time (min) Tcoffee Tmug Tjoulie x dTc/dt dTmug/dt dTJ/dt dx/dt Ejoulie
0 100 25 25 NA -0.40755 1.855774 0.712979 NA 0
0.01349995 99.66988171 26.50317 25.57751 NA -0.39849 1.807182 0.704351 NA 70.84444
0.0269999 99.34710938 27.96698 26.14803 NA -0.38964 1.75983 0.695859 NA 140.8315
0.040499849 99.03150161 29.39244 26.71168 NA -0.38101 1.713687 0.6875 NA 209.9748
0.053999799 98.72288158 30.78052 27.26855 NA -0.3726 1.668721 0.679272 NA 278.2876
0.067499749 98.42107692 32.13218 27.81876 NA -0.36439 1.624904 0.671173 NA 345.7828
0.080999699 98.12591956 33.44835 28.36241 NA -0.35639 1.582206 0.663199 NA 412.4733
0.094499649 97.83724568 34.72993 28.89959 NA -0.34858 1.540598 0.655348 NA 478.3713
0.107999599 97.55489558 35.97781 29.43042 NA -0.34097 1.500053 0.647617 NA 543.4893
0.121499548 97.27871353 37.19285 29.95499 NA -0.33354 1.460544 0.640005 NA 607.8392
0.134999498 97.00854776 38.37589 30.47339 NA -0.32629 1.422045 0.632509 NA 671.4327
0.148499448 96.74425028 39.52774 30.98572 NA -0.31923 1.384529 0.625126 NA 734.2813
0.161999398 96.48567683 40.6492 31.49207 NA -0.31233 1.347972 0.617854 NA 796.3963
0.175499348 96.23268675 41.74105 31.99253 NA -0.30561 1.312349 0.610692 NA 857.7888

Fig. 5.22 Spreadsheet simulation during the initial Joulie heating phase. Similar sheets are used
for Joulie superheated and cooling phases. A fixed time step based on a fraction of the maximum
allowable step size is used. “Goal seek” is used to calculate the correct time step so that the Joulie
temperature equals the melting point at the end of the stage (100 steps chosen). An “if-then-else”
statement is placed in the first row that checks, and returns “VALID SIMULATION” if the last
row for Joulie temperature matches the melting point. If not, the formula returns “SIMULATION
NOT VALID”

JOULIE MELTING PHASE VALID SIMULATION

dtmax,Joulie NA sec
dtmax,Coffee 184.0243 sec
dtmax,mug 37.18519 sec
dx 0.01 step in x

time(min) Tcoffee Tmug Tjoulie x dTc/dt dTmug/dt dTJ/dt dx/dt Ejoulie dt

1.349994983 85.79192 77.1434 60 0 -0.06605 0.101952 NA 0.003528 4293.523 2.834817
1.397241934 85.6047 77.43241 60 0.01 -0.06372 0.089547 NA 0.003502 6393.523 2.855546
1.444834368 85.42275 77.68812 60 0.02 -0.06157 0.078168 NA 0.003477 8493.523 2.875983
1.492767416 85.24568 77.91293 60 0.03 -0.05959 0.067741 NA 0.003453 10593.52 2.896155
1.541036665 85.0731 78.10912 60 0.04 -0.05776 0.058195 NA 0.003429 12693.52 2.916089
1.589638147 84.90467 78.27882 60 0.05 -0.05608 0.049464 NA 0.003406 14793.52 2.935811
1.638568332 84.74004 78.42404 60 0.06 -0.05452 0.041485 NA 0.003384 16893.52 2.955347
1.687824115 84.5789 78.54664 60 0.07 -0.05309 0.034201 NA 0.003362 18993.52 2.974722
1.737402809 84.42097 78.64838 60 0.08 -0.05177 0.027557 NA 0.00334 21093.52 2.99396
1.787302135 84.26596 78.73088 60 0.09 -0.05056 0.021503 NA 0.003319 23193.52 3.013084
1.837520202 84.11364 78.79567 60 0.1 -0.04943 0.015991 NA 0.003298 25293.52 3.032118
1.888055505 83.96374 78.84416 60 0.11 -0.0484 0.010978 NA 0.003278 27393.52 3.051084
1.9389069 83.81607 78.87765 60 0.12 -0.04745 0.006424 NA 0.003257 29493.52 3.070002
1.990073602 83.67041 78.89737 60 0.13 -0.04656 0.002289 NA 0.003237 31593.52 3.088894
2.041555164 83.52658 78.90444 60 0.14 -0.04575 -0.00146 NA 0.003218 33693.52 3.107778
2.093351468 83.3844 78.89991 60 0.15 -0.045 -0.00486 NA 0.003198 35793.52 3.126674

Fig. 5.23 Spreadsheet simulation during melting phase of Joulie. A similar sheet is used for the
Joulie freezing stage. The numerical step is taken with a fixed change in quality, “x”, and the
required time step “dt” is calculated from the governing equation for the quality. The initial
conditions (time, temperatures) refer to the last row from the heating phase
192 5 Few-Node Transient Models

100
Coffee without Joulies
90
Coffee with 3 Joulies
80
Temperature (oC)

70 Mug

60

50

40 Joulies

30

20
0 10 20 30 40 50 60 70 80 90
Time (min)

Fig. 5.24 Temperatures as a function of time for the case of a mug of coffee with 3 Joulies. A
2-node simulation (with additional coffee replacing the volume occupied by the Joulies) was
conducted as a control with the same conditions

100
Coffee without Joulies

90
Coffee with 3 Joulies
80
Mug
Temperature (oC)

70

60

50
Joulies
Superheating

40
Freezing
Heating

Cooling
Melting

30
STAGE:

20
0.01 0.1 1 10 100
Time (min)

Fig. 5.25 The same data as Fig. 5.24 plotted on semi-logx axes. The boundaries for the stages
undergone by the Joulies are shown as vertical dotted lines

Plots of temperature vs. time are shown in Figs. 5.24 (linear axes) and 5.25
(semi-logx axes). The 2-node coffee/mug model is shown as the control case. The
mug curve is shown only during the Joulie heating and melting stages to reduce
clutter in the plots. The mug follows the coffee (a few degrees lower) after that.
Appendix 1. Nodal Response Time 193

During the Joulie heating phase, the mug and Joulie temperatures rise quickly, the
mug rising faster owing to its smaller time constant. The Joulies then reach
the melting point and stay there until melting is complete. Thereafter, in the
superheating phase, the Joulie temperature increases (as a liquid) toward the
coffee, and then decay as the coffee continues to cool. The peak in the Joulie
temperature coincides with the point where the coffee temperature equals the
Joulie temperature, as there is no heat transfer during this time. After the peak, the
coffee is colder than the Joulie, and heat flows from the Joulie to the coffee,
returning energy stored to the coffee. Once the Joulie cools to its melting
temperature, the liquid freezes to a solid until phase change is complete. The
Joulie cools with the coffee after that.
Compared to the control curve (coffee without Joulies), the coffee cools quickly
with Joulies, reaching a drinking temperature in a noticeably shorter time. The heat
is retained for a slightly longer time, yielding a much longer time window.
For me, it’s being able to drink hot coffee much sooner that sold me on the
product. As I said “I like Joulies so much, I wrote a book about it!”

Appendix 1. Nodal Response Time

Figure 5.3, reproduced here, shows a node with a capacitance, two neighbors and
heat generation. The inherent response time of this node is established by deter-
mining its response assuming its neighbors are fixed, not variable. In this case, T1
will approach an equilibrium value that will depend on its two neighbors and the
heat generation.

Repeating the governing nodal equation:

dT 1 T2  T1 T1  T1
C1 ¼ Q_ 1 þ þ
dt R12 R11
A dimensionless temperature (ψ 1) that starts at a value of unity (1) and approaches a
value of zero at equilibrium, and a dimensionless time defining a response time for
node 1 (τ1 ¼ RC1equiv, to be determined) are defined:
194 5 Few-Node Transient Models

T 1  T 1eq
ψ1 ¼
T 0  T 1eq
t t
^t ¼ ¼
τ1 RC1equiv

Inserting into the nodal equation:

     
C1 d T 1eq þ ψ 1 T 0  T 1eq T 2fixed  T 1eq þ ψ 1 T 0  T 1eq
¼ Q_ 1 þ
τ1 d^t R12
  
T 1  T 1eq þ ψ 2 T 0  T 1eq
þ
R11
Note that T2 is given a “fixed” superscript. In the system, T2 is a moving target, so
this calculation is a hypothetical “what if.” With T2 and therefore T1eq fixed, the
first term under the time derivative is zero. Rearranging:
  
C1 T 0  T 1eq dψ 1 _ T 2fixed  T 1eq T 1  T 1eq
¼ Q1 þ þ
τ1 d^t R12 R11
 
 1 1
 T 0  T 1eq ψ 1 þ
R12 R11

The term in brackets [ ] is identically zero because the equilibrium value for T1 is
defined by setting the time derivative of the nodal equation to zero:

T 2fixed  T 1eq T 1  T 1eq

0 ¼ Q_ 1 þ þ
R12 R11

Then the factor (T0  T1eq) cancels and the dimensionless equation is:
 
dψ 1 τ1 1 1
¼ þ ψ
d^t C1 R12 R11 1

The time constant can now be defined as:

C1
τ1 ¼   ¼ R1equiv C1
1
R12 þ R11
1

and the governing equation becomes the first-order model:

Appendix 2. Detailed Thermal Resistance Formulas for Coffee Mug Problem 195

dψ 1
¼ ψ 1
dt
with solution:

ψ 1 ¼ e^t

It is reemphasized that this temperature response is a hypothetical one, with T2

fixed. The aim of this derivation is to determine the inherent response of a given
node to its immediate environment. In a real simulation, T1 would not approach this
equilibrium value because T2 is changing with time, unless the system itself has
reached equilibrium.

Appendix 2. Detailed Thermal Resistance Formulas for Coffee

Mug Problem

Figures 5.6 and 5.7 (repeated side by side) shows the input parameters
(left) and derived quantities (right) from the “params” worksheet from
“Ch5_2Node_Transient.xls.” The formulas used in the tables on the right side
(that reference the cells in the input quantities block) are detailed in this appendix.
196 5 Few-Node Transient Models
Appendix 2. Detailed Thermal Resistance Formulas for Coffee Mug Problem 197

The basic nomenclature of the geometry is repeated here, introduced previously

in Chap. 2.
D W
L

V H

coffee H0

mug

The coffee depth is:


V
HL¼
πD2 =4

The mug and coffee capacitances are:

Cmug ¼ mmug cmug and Ccoffee ¼ ρcoffee ccoffee 

V

The area of the coffee-free surface is:

Afree surface ¼ πD2 =4

The area of contact between the coffee and the mug is the sum of the bottom and
sides:
Ainside ¼ πD2 =4 þ πDðH  LÞ

The total outside surface area of the mug (exposed to ambient) includes the outside,
bottom (neglecting the bottom ridge area), the top rim surface, and the inside mug
surface above the coffee level:

Aoutside ¼ Outside þ Bottom þ Rim þ Inside

h i
Aoutside ¼ π ðD þ 2wÞHo þ π ðD þ 2wÞ2 =4 þ π ðD þ 2wÞ2  D2 þ πDL

The average mug surface area (used in the mug-conductive resistance) is the
geometric average of the inside and outside areas:

Aaverage ¼ ðAinside þ Aoutside Þ=2

198 5 Few-Node Transient Models

Individual Resistances

The RC network is repeated here for reference.

The convective resistance between the coffee and the mug is:

T∞
evaporation

convection
radiation

Rsa

Tsurface
coffee Rcs mug
Tinside T2 Toutside T∞
T1 convection
radiation
Rcm Rmug Rmug
2 2 Rma, side
C1 C2
Rma, floor

1
Rcs ¼
hcoffee=surface Atop

The value chosen for the equivalent heat transfer coefficient (hcoffee/surface ¼ 470 W/m2/K)
is representative of a natural convection between a liquid with water properties
and a solid surface.
The equivalent convective/radiative/evaporative resistance between the coffee
surface and ambient is:

1
Rsa ¼
hair=top Atop

The value chosen for the equivalent heat transfer coefficient is built on values of
6.8, 6.6, and 10 W/m2/K for convection, radiation, and evaporation, respectively.
The convective resistance between the coffee and the mug inner surface is:
1
Rcm ¼
hcm Ainside
The value chosen for the convection coefficient (470 W/m2/K, with radiation
considered negligible in comparison) is representative of natural convection
between a liquid with water properties and a vertical surface.
Appendix 3. Joulie Property Estimation 199

The conductive resistance across the mug wall is:

w
Rm ¼
kmug Aaverage

This resistance is split into two equal parts between the capacity-carrying mug node
and the mug surfaces. This model is considered reasonable for nearly full mugs. A
detailed analysis of nearly empty mugs is deferred to a later chapter.
Heat transfer from the outer surface of the mug to ambient is split into two
parallel channels: side and bottom. For the side, a convective/radiative equivalent
resistance for those portions exposed to air is:

1
Rma, sides ¼  
hair=sides Aoutside  π ðD þ 2wÞ2 =4

The area is the outside area subtracted by the surface area of the bottom. The
equivalent heat transfer coefficient is the sum of the convective and radiative
portions (8.8 and 8.6 W/m2/K, respectively). For the heat transfer through the
bottom:

1
Rma, floor ¼
hfloor π ðD þ 2wÞ2 =4

The effective heat transfer coefficient (hfloor ¼ 4.7 W/m2/K) is detailed in Chap. 4,
and reflects heat that flows from the bottom of the mug to the floor it is placed on,
and then conducted outward into the floor.

Appendix 3. Joulie Property Estimation

Resistance and capacitance value estimation requires assumptions about the con-
struction of the Joulie, detailed in this appendix, and calculated in a separate
worksheet (Fig. 5.26). The mass and volume of Joulies are easily measured
(34.0 g and 20.6 mL). Thermal properties are obtained online (accessed 1 February
2014), with convection coefficients inside and outside entered as representative
values.
The Joulies are modeled as equivalent spheres. That is, the equivalent outer
diameter is calculated from the measured volume as:
 1=3
6V
d¼
π
200 5 Few-Node Transient Models

MEASURED VALUES
mass 0.028 kg Single Joulie mass
Vtotal 2.06E-05 m^3 Single Joulie volume

THERMAL PROPERTIES (STEEL SHELL, PARAFFIN WAX CORE)

kshell 20 W/m/K http://www.engineeringtoolbox.com/thermal-conductivity-metals-d_858.html
rhoshell 8000 kg/m^3 http://www.engineeringtoolbox.com/metal-alloys-densities-d_50.html
cpshell 490 J/kg/K http://www.engineeringtoolbox.com/specific-heat-metals-d_152.html
kwax 0.25 W/m/K http://www.engineeringtoolbox.com/thermal-conductivity-d_429.html
rhowax 721 kg/m^3 http://www.engineeringtoolbox.com/density-materials-d_1652.html
cpwax 2500 J/kg/K http://en.wikipedia.org/wiki/Paraffin_wax
hCJ 250 W/m2/K convection bet. Coffee and Joulie
hwax 200 W/m2/K convection bet. Wax and shell

DERIVED PROPERTY VALUES

d equivalent diameter of single Joulie
0.034 m
dwax diameter of wax core
0.0330 m
Atotal 3.63E-03 m^2 surface area
Awax 3.42E-03 m2 surface area of wax
Vwax 1.88E-05 m^3 volume of wax core
tshell 5.13E-04 m shell thickness
mwax 0.0135 kg mass of wax
mshell 0.0145 kg mass of shell
rhoave 1361 kg/m^3 average density
cpave 1462 J/kg average heat capacity
THERMAL RESISTANCES OF SINGLE JOULIE
Rwax 19.31 oC/W conductive resistance across wax
Rwax 1.46 oC/W Convective inner wax layer
Rshell 0.01 oC/W conductive resistance across shell
Rconv 1.10 oC/W convective resistance
Rcj 2.57 oC/W equivalent resistance between single Joulie and coffee

Fig. 5.26 Input spreadsheet focused on property values for a single Joulie used as inputs to the
simulation input sheet

The diameter of the wax (and thereby the wall thickness) is determined by summing
the total mass from wax core plus shell:

mTotal ¼ ρshell V shell þ ρwax V wax ¼ ρshell ðV Total  V wax Þ þ ρwax V wax
¼ ρshell V Total  V wax ðρshell  ρwax Þ

Solving for the wax volume:

ρshell V total  mtotal

V wax ¼
ρshell  ρwax

 
6V wax 1=3
dwax ¼
π
Workshop 5.1. A Batch Mode Solar Water Heater 201

The total specific heat (during single phase stages) is:

CJoulie ¼ ðmcÞTotal ¼ ρshell V shell cshell þ ρwax V wax cwax

Yielding effective specific heat:

ρshell V shell cshell þ ρwax V wax cwax

cave ¼
ρshell V shell þ ρwax V wax

For the resistances, there are two separate entries for Rwax, representing the thermal
resistance between the wax and the inner shell wall. The first is a pure conduction
value and the second is assuming a convective mode. The conduction value is much
larger than the convective one and also large compared to the outer convection plus
conduction across the shell wall (which is negligible). That means that the Biot
number of the wax is large, and during heating, the center of the wax will respond
much more slowly than its outer layers. Use of the convection resistance is
considered a better way to estimate the rate at which heat is absorbed by the Joulie.
A more detailed spatial model would be the way to address this issue (and a great
assignment!).
These calculations are for a single Joulie. Note that when multiple Joulies are
used, the total capacitance is proportional to the number of Joulies, while the
thermal resistance is inversely proportional. Adding more Joulies increases the
total surface area, reducing the resistance of heat flow. That is:

1
C3 Joulies ¼ 3Csingle Joulie while R3 Joulies ¼ Rsingle Joulie
3

I’d bet that lots of people would not think that through and assume R3 Joulies ¼
3R1 Joulie, and in so doing confuse a parallel heat transfer concept with a series
effect.

Workshop 5.1. A Batch Mode Solar Water Heater

An alternative to burning fossil fuels to produce hot water for domestic needs is to
heat water using solar energy. It is a common experience that placing a garden hose
in direct sunlight can produce hot water. It is also a common experience that
sunlight can be focused using a magnifying glass. Alternatively, a reflecting mirror
can be used to focus sunlight. This workshop applies these principles to a specific
arrangement in which a “solar pipe” is filled with initially cold water and placed in
focused sunlight. The water is heated (without a flow of water) until the water
reaches a target temperature, at which point a pump is activated (say by a solar
202 5 Few-Node Transient Models

photovoltaic cell) and the hot water is pumped out of the pipe into a separate tank,
and the pipe is filled with a fresh batch of cold water, completing the batch cycle.
The time it takes to complete a cycle depends on the angle and intensity of sunlight
which varies throughout the course of a day.
An alternative mode, called continuous mode, is to use the same arrangement but
with a continuous flow of water through the solar pipe. That problem will be studied
much later, after the effects of advective flows are developed.
Problem statement: In a solar thermal application (Fig. 5.27), black polypropylene
tubing (kp ¼ 0.15 W/m/ C, ρp ¼ 1,400 kg/m3, cp ¼ 1,600 J/kg/ C) of length L ¼ 3 m,
outer diameter (D0) of 0.040 m, and wall thickness t ¼ 0.0028 m is subjected to
focused sunlight distributed evenly on the outer surface (concentrated with a
parabolic mirror). The intensity of sunlight that enters the top plane (the solar
insolation) has a maximum value of I0,max ¼ 800 W/m2 (on a clear day at noon)
and a solar collection width to rod diameter ratio (n) to be determined as a design
parameter. An optical efficiency ηopt ¼ 0.9 (defined as the ratio of the solar radiation
that strikes the pipe to the collected radiation) accounts for collected solar energy
not absorbed by the pipe. In the end view, the angle of the sun remains perpendicu-
lar to the collecting plane through proper orientation of the mirror (and changes
gradually through the course of a year). From the side view, the angle between the
incoming solar rays and the perpendicular to the pipe varies through the course of a
day, from θ ¼ 90 at sunrise to 0 at noon to 90 at sunset. The absorption
coefficient of incident radiation is α ¼ 0.85 (85 % of the radiation that strikes the
pipe is absorbed, 15 % is reflected). At the start of a cycle, the tube is filled with
liquid water (ρw ¼ 1,000 kg/m3, cw ¼ 4,200 J/kg/K) at a temperature T0 ¼ 15  C.
The tube is exposed to ambient at T1 ¼ 25  C. The objective is to raise the
temperature of the water to a target of Tfinal ¼ 50  C before pumping it into a
separate collection tank and starting a new cycle.

Fig. 5.27 Schematic of a solar pipe placed at the focal point of a parabolic mirror
Workshop 5.1. A Batch Mode Solar Water Heater 203

Write an Excel spreadsheet or program code that implements an explicit numer-

ical formulation of a 2-node model (tube and water). Assume the inner surface is
exposed to the water with a constant convection coefficient of hi ¼ 300 W/m2/ C,
and the outer surface is exposed to ambient with a constant convection coefficient
of ho ¼ 15 W/m2/ C. Use the program to determine the size of the solar collection
area (nDL) that would cause the outer surface of the pipe to reach its melting point
(Tmelt ¼ 160  C) at the same time the water reaches its target of 50  C when the
solar flux equals its maximum value at noon. Then determine the time it takes to
heat a fresh batch of water to its target as a function of solar angle (which can be
related to time of day) for fixed values of solar flux. Report a minimum intensity
required to heat the water to the target temperature, the maximum intensity before
the outer surface of the pipe reaches its melting point (Tmelt ¼ 160  C), and the size
of the solar collection area (nD), assuming the maximum intensity of direct sunlight
is 800 W/m2 (a typical maximum value on a clear day at noon). Prepare a table of
relevant quantities and plots of

• required heating time as a function of intensity (above the minimum required)

• all nodal temperatures (including non-capacity-carrying nodes) vs. time with
maximum intensity
• temperature vs. radial position at fixed times with maximum intensity

Note: A specific 2-node model is developed on the next page. It is advised that you
develop your own model first before consulting that model. Then decide which
model to implement.

Model Development
The 2-node thermal RC network shown in Fig. 5.28 treats node 1 as the water and
node 2 as the pipe. The pipe node (2) is placed in the middle of the pipe wall. The
solar absorption is modeled as a heat source absorbed on the surface (not directly on
the capacity-carrying node). The surface node is in thermal contact with ambient
(through a parallel convective/radiative channel with an assumed constant effective
heat transfer coefficient) and the pipe center (through a conductive channel). The
pipe is in thermal contact with the water (through a series conductive/convective
resistance).

T1 Ti T2 Tabs T∞
R1i Ri2 R2a Ra∞
C1
C2
.
Qabs

Fig. 5.28 2-Node RC network for a solar pipe in batch mode

204 5 Few-Node Transient Models

The nodal equations are (water ¼ 1, pipe ¼ 2, pipe outer surface ¼ abs):

dT 1 T2  T1
C1 ¼
dt R1i þ Ri2
dT 2 T 1  T 2 T abs  T 2
C2 ¼ þ
dt R1i þ Ri2 R2a
The solar flux is absorbed at the outer surface, not at the center, and a nodal balance
on the surface node is:

T 2  T abs T 1  T abs
0 ¼ Q_ abs þ þ
R2a Ra1
Expressing the time derivatives with a forward difference approximation and
rearranging for the recursion relations, the future values (superscript “p + 1”, after
a time step Δt) are explicitly expressed in terms of present values (superscript “p”):
!
ðpÞ ðpÞ
ðpþ1Þ ðpÞ Δt T 2  T 1
T1 ¼ T1 þ
C1 R1i þ Ri2
" ! !#
ðpÞ ðpÞ ðpÞ ðpÞ
ðpþ1Þ ðpÞ Δt T1  T2 T abs  T 2
T2 ¼ T2 þ þ
C2 R1i þ Ri2 R2a

After rearrangement, the temperature at the absorbing surface is expressed as:

!  
ðpÞ
ðpÞ _ T2 T1 1 1
T abs ¼ Q abs þ þ þ
R2a Ra1 R2a Ra1

The time constants for the two nodes are given by:

τ1 ¼ ðR1i þ Ri2 ÞC1

and
C2
τ2 ¼
1 1
þ
R1i þ Ri2 R2a þ Ra1
The time step can be chosen to be a fraction of the smallest of these.
The rate of heat absorption is defined by:
0 10 10 1
fraction of fraction of Rate of Solar
Q_ abs ¼ @ collected radiation A@ radiation that strikes A@ Radiation A
that strikes pipe pipe that is absorbed Collected

Q_ abs ¼ ηopt ðαÞðnDLI 0 cos θÞ
Workshop 5.2. Cooper Cooling with a Finite Reservoir 205

Workshop 5.2. Cooper Cooling with a Finite Reservoir

In Workshop 4.2, various methods of chilling a beverage were considered, the

fastest of which involved the process of Cooper Cooling™ in which a horizontal
beverage is rotated along its axis while being sprayed with a cold jet of water. In
that problem statement, the temperature of the water jet was a fixed input value, a
result that would occur if the source of the water was a large (infinite) reservoir of
ice and water, and/or the water heated by the beverage was not returned and
recycled. However, the first marketed consumer product that incorporated the
Cooper Cooling process has a finite reservoir. Jet temperature data and a schematic
of the process are shown in Fig. 5.29, taken from Jaime Cachiero’s Master’s Thesis.
The cold water jet is drawn from an ice/water bath and is pumped onto the
beverage. Heat transferred from the beverage is added to the jet before the jet is
returned to the ice water bath. Furthermore, a large portion of the electrical input to
the pump is added to the recirculating water. In the reservoir, the water is in contact
with ice, which removes heat from the water jet. Since the beverage temperature
changes with time, the rate at which heat is added to the jet changes. In practice,
during the first four or so beverages, the jet temperature fluctuates around approxi-
mately 2  C (the input value used for Workshop 4.2). The goal of this workshop is
to develop a 3-node model (beverage, reservoir, ice) that predicts this behavior and
can be used to explore other scenarios.

25

Jet
20 Temperature
Bev
Q
Temperature (C)

15
Ice

10

Cold Water No Ice

5
Cold Water

0
0 100 200 300 400 500 600 700 800
Time (seconds)

Fig. 5.29 Schematic of Cooper Cooling with a Finite Reservoir. The data shows the measured
temperature of the water jet before it is used to chill a horizontal rotating beverage. A warm
beverage is added every 60 s (copyright Wei Dai, reproduced with permission)
206 5 Few-Node Transient Models

Note: A specific 3-node model is developed on the next page. It is advised that you
develop your own model first before consulting that model. Then decide which
model to implement.

Model Development
It is good practice to draw a thermal resistance network for this problem in which
capacitance is assigned to the beverage and reservoir, and a rechargeable battery at
the melting point of the reservoir (Tmp) is used to represent the ice. The mass of the
ice, but not its temperature, changes as it melts. Notice that if a reservoir fluid other
than water is used (like salt water), the melting point can differ from that of water.
Also, the capacitance of the reservoir changes with time in a predictable way as ice
melts. The composition of the reservoir might change if a reservoir other than water
were used.
The governing equation for the beverage is the same as that in Workshop 4.2,
namely:

dT bev T res  T bev

Cbev ¼
dt Rbev
The difference in this model is that the reservoir temperature is not a fixed input,
and this differential equation cannot be integrated directly.
The liquid water in the system is considered to be at an average temperature
(Tres) that exchanges heat with the beverage (at Tbev), the ice (at Tmp, the melting
point of the reservoir), and ambient (at T1) and picks up electrical energy that is
converted to thermal energy in the pump (VIpump):

dT res T bev  T res T mp  T rex T 1  T res

Cres ¼ þ þ þ VI pump
dt Rbev Rice R1
The energy stored in the ice is in the form of latent (phase) energy, and an energy
balance on the ice is given by:

dMice T res  T mp
Lf ¼
dt Rice
Lf is the heat of fusion of ice (330 kJ/kg) and the thermal resistance between the ice
and the reservoir (Rice) is one of external flow forced convection on a surface area
that changes with time.

Explicit Formulation

The nodal equations written in the form of an explicit formulation are with
superscripts used to indicate all values that change with time:
Workshop 5.2. Cooper Cooling with a Finite Reservoir 207

!
ðpÞ
ðpþ1Þ ðpÞ ΔtðpÞ T ðres
pÞ
 T bev
T bev ¼ T bev þ
Cbev Rbev
!
ðpÞ
ΔtðpÞ T bev  T ðres
pÞ
T mp  T ðres
pÞ
T 1  T ðres
pÞ
T ðres
pþ1Þ
¼ T ðres
pÞ
þ ðpÞ þ ðpÞ
þ þ VI pump
Cres Rbev Rice R1
!
ðpþ1Þ ðpÞ ΔtðpÞ T ðres
pÞ
 T mp
Mice ¼ Mice þ ðpÞ
Lv Rice

The reservoir capacitance equals the specific heat of the reservoir times the mass of
the reservoir, which equals the initial mass of the reservoir (Mres,0) plus the mass of
ice that has melted (the initial mass of ice less the remaining ice, Mice,0 – Mice):
 
ðpÞ
Cðres
pÞ
¼ cres Mres, 0 þ Mice, 0  Mice

The thermal resistance that models the melting of the ice requires an assumption
about the shape of the ice, and how that changes. For example, if ice cubes (actual
cubes of dimension “w”) are used, and it is assumed that they remain as cubes (even
though in practice the edges become rounded), then the surface area can be related
to the mass. That is, if a total of ncubes are added initially of dimension w0, the initial
mass of ice is given by:

Mice, 0 ¼ ncubes w30

As the ice melts, the number of cubes does not change, but their size does, so that
the remaining mass of ice can be related to the ice cube size by:

Mice ¼ ncubes w3

The thermal resistance between the ice and the reservoir is given by:

ðpÞ 1 1
Rice ¼ ðpÞ
¼ 2
hice Aice hice ncubes 6ðwðpÞ Þ

Notice that the thermal resistance increases as the ice cubes become smaller. A value
of 1,500 W/m2/K would be reasonable for the convection coefficient in this case.
The values for the beverage capacitance and thermal resistance can be taken from
Workshop 4.2, and simulations with different beverages can be conducted. The
electrical input to the pump can be varied, but VIpump ¼ 20 W is a good starting place.
There are a variety of scenarios that can be used to run a simulation. For
example, a prechilling period can be conducted in which case a number of ice
cubes are placed into an input mass of reservoir at an initial temperature, without
actively chilling a beverage, and letting the initial ice/water reservoir come to
208 5 Few-Node Transient Models

equilibrium before adding beverages. A decision about how long to chill each
individual beverage can be made and that simply involves resetting the temperature
of the beverage. Chilling consecutive beverages can be simulated by either fixing
the chilling time or allowing the system to operate until the beverage reaches a
target temperature. In the former case, the final temperature of each beverage will
be different. In the latter case, the total time required to chill each beverage will be
different. Eventually, the reservoir will not be able to chill a beverage to that target.

Additional Workshop Ideas

Workshop 5.2: Solve the 2-node mug of coffee problem using a variable step-size
algorithm that fixes the maximum change in temperature (absolute value) of either
the mug or coffee.
Workshop 5.3: Solve the 2-node mug of coffee problem using a variable step-size
algorithm that fixes the maximum fractional change in temperature of either the
mug or coffee.
Workshop 5.4: Solve the 2-node mug of coffee problem using the backward
implicit method with a fixed time step.
Workshop 5.5: Solve the 2-node mug of coffee problem using a Styrofoam cup.
Workshop 5.6: Solve the 2-node mug of coffee problem using the explicit numeri-
cal scheme with variable thermal resistances on the side channel. Assume an
emissivity of 0.9 for the radiation coefficient, and use the following formulas
(derived in a future chapter) for the convection coefficients. For convection
between the outer surface of the mug and ambient:


0:25

T mug  T ambient

hair ¼ 1:5646  0:0015T film, air

H 0:25
0

where Tfilm,air is the average of the mug and ambient. For convection between the
inner surface of the mug and the coffee:

 

T
0:25

coffee  T mug
hcoffee ¼ 32:38T film, coffee
0:441
0:25
H0

where Tfilm,coffee is the average of the coffee and mug. Compare this model with the
model any of the constant convection coefficient models (and the data).
Workshop 5.7: A hybrid implicit method
For a stiff problem (τ2  τ1), a legitimate approach is to use a forward difference
approximation for the node with the larger response time (i.e., coffee) and a
backward difference for the node with the shorter response time (mug). The future
value of T1 is obtained (from before) as:
Additional Workshop Ideas 209

 ðpÞ 
ΔtðpÞ ðpÞ T2  T 1 ðpÞ T 1 ðpÞ  T 1 ðpÞ
Q_ 1 þ
ðpþ1Þ ðpÞ
T1 ¼ T1 þ þ
C1 ðpÞ R12 ðpÞ R11 ðpÞ
The backward difference equation for T2 is rearranged as:
!
C2 ðpþ1Þ 1 1 ðpþ1Þ C
ðpþ1Þ ðpÞ
T T 1 ðpþ1Þ T 1 ðpþ1Þ
¼ Q_ 2
ðpþ1Þ
T2 þ þ þ 2 ðpþ12Þ þ þ
Δtðpþ1Þ R12 ðpþ1Þ R21 ðpþ1Þ Δt R12 ðpþ1Þ R21 ðpþ1Þ

ðpþ1Þ ðpÞ
Since T1 changes slowly compared with T2, then for one time step, T 1  T1 .
Furthermore, when the coefficients vary with time, they will do so slowly, and the R
and C values can be evaluated at the present, not the future time. Therefore, a
legitimate explicit prediction of the future temperature of T2 that is inherently stable
numerically (let Δt ! 1) is:
ðpÞ ðpÞ
ðpÞ C2 T 2 T 1 ðpÞ T 1 ðpÞ
¼ Q_ 2 þ
ðpþ1Þ
T2 þ þ
Δt ð pÞ
R12 ðp Þ
R21 ðpÞ
C2 ðpÞ 1 1
þ þ
ΔtðpÞ R12 ðpÞ R21 ðpÞ
For this scheme, the time step can be chosen to be a value greater than the Second
Law restriction on a full explicit scheme.
Workshop 5.8: 2-node Electric Water Heater:
An electrically resisting element made of nichrome (thermal conductivity
kelement ¼ 11.0 W/m/K, density relement ¼ 8,400 kg/m3, specific heat
celement ¼ 450 J/kg/K) of diameter d ¼ 0.015 m with a total length to be determined
is designed to be immersed in 1.0 L of liquid water in a cylindrical container of
inner diameter D ¼ 0.15 m with a thin wall (negligible thermal capacity and
conductive resistance). The water is initially at 15  C and electrical power at the
rate of Q ¼ 1,000 W is delivered to the heating element.
Conduct a 2-node analysis (element and water). Assume that the heat transfer
coefficient between the element and the water, and between the water and the
container is helement,water ¼ hwater,container ¼ 400 W/m2/K, the convective/radiative
coefficient between the outer wall and ambient is hcontainer,air ¼ 15 W/m2/K.
Determine:

1. The length of the heating element needed so that the water temperature is 90  C
at the point the water starts to boil (i.e., reaches 100  C) in the thermal boundary
layer between the element and the water and the corresponding time required to
boil. Note: if heated beyond this point, most of the heat will go toward boiling
the water in that boundary layer, and not to raising the temperature of the bulk of
the liquid. That is to say, it is difficult to obtain water hotter than 90  C for this
heating element at this heating rate.
2. The time to start to boil and the final water temperature obtained for this length
heating element, but with a heating power of 500 W.
3. The minimum power that would yield 90  C water and the corresponding time
and the temperature of the element at that time.
Multi-Node Transients
6

This chapter builds on the few-node numerical approach, detailing the side wall of
the mug of coffee problem, which is where the majority of heat is lost (60 % of the
surface area exposed to ambient). Results are generated only for the mug heating
phase, as computationally, simulation of the entire cooling event would require
several large spreadsheets. Conceptually, multi-node problems involve a simple
extension of the principles developed for the 2-node mug of coffee model. For
example, adding Coffee Joulies™ added a third node. This type of problem is
generally solvable with a spreadsheet approach for an explicit formulation (but not
for an implicit one).
However, a subtle distinction is made between a few-node and a multi-node
problem. In programming a multi-node simulation, some or all of the recursion
formulas can be written in terms of an input numerical parameter, say “n”. For
example, the mug side wall will be split into “n” nodes. The governing equations
will be derived in terms of “n”. However, for spreadsheet solutions, as detailed in
this chapter, a commitment to a specific number must be made at the outset (set up
the number of columns). Technically, then, this is a “few-node” approach, but a
spreadsheet implementation is developed here to demonstrate the details of the
method.
Numerical modeling in heat transfer involves discretization of both space and
time. In the 1-node model, space was discretized by lumping all the mass of the
coffee/mug system into a single node. If the thermal resistances are taken as
constants, the governing ODE is easy to integrate and it is not necessary to
discretize time. For variable resistances (natural convection and radiation
boundaries, developed in future chapters), time can be discretized by turning the
time derivative into an approximate algebraic expression using finite difference
approximations, and the transient problem can be simulated by appropriately
stepping forward in time. In the 2-node model, space was discretized into two
lumped pieces; coffee and mug. Time was discretized using a finite difference
approximation. This chapter introduces a more general approach by breaking space
into many discrete nodes.

# Springer International Publishing Switzerland 2015 211

G. Sidebotham, Heat Transfer Modeling: An Inductive Approach,
DOI 10.1007/978-3-319-14514-3_6
212 6 Multi-Node Transients

The model developed is spatially one-dimensional and is focused primarily on

the heat transfer process across the mug side wall. Edge effects near the rim and
base are not modeled. The heat loss through the free surface and bottom are
modeled as they were for the extended lumped model. An extension to 2D and
3D is relatively straightforward, conceptually. In this model, the convection and
radiation coefficients depend on temperature, and therefore vary with time. The
convection formulas are developed in a future chapter, but those results are used
here, much the same way that constant values of h have been used throughout. The
radiation coefficients have been developed already.
Results for the coffee/mug problem are presented. The model implementation
was done with a rather large Excel file, written so that all parameters could be
changed to simulate other problems that fit the same basic model of a cylindrical
container holding a liquid at a temperature different from a constant ambient
temperature. There are good modeling lessons at play. This more detailed model
requires more considerable effort, both in setup and in computation, than the 2-node
model, yet the key results are virtually identical. That is not to say more detailed
models are never needed. It is just demonstrated here that they are overkill for the
cooling mug of coffee example.

Problem Statement
The first part of the problem statement, the description of the mug of coffee and its
dimensions, is exactly the same as the original one. The objective stated is different:
A volume V ¼ 0.41 L of coffee at 1 atm, 92  C is quickly poured into a room
temperature (23  C) ceramic mug with mass of mmug ¼ 0.335 kg, inner diameter
D ¼ 0.083 m, inner height H ¼ 0.098 m, and nominal wall thickness w ¼ 0.0040 m.
The mug of coffee is placed on a table in a still room. Determine the temperature
distribution in the mug side wall as a function of time.

6.1 Multi-Node Model Development

Figure 6.1 shows an RC network for the mug of coffee. The coffee itself is modeled
with a single node, with three parallel channels for heat flow: top (free) surface,
bottom, and side wall. The side wall has been broken up into a total of “n” nodes,
where n will be set to 8 for the case study of the mug of coffee implemented in an
Excel file. Each node represents a cylindrical shell and has a capacitance associated
with it. The sum of all “n” capacitances is the total capacitance of the side wall of
the mug. Heat transfer out of the top rim and bottom of the side wall is neglected.
Thermal resistances between each node represent a conductive thermal channel
between them. The coffee (at temperature T(t)) has a capacitance C that consists of
the capacitance of the coffee plus the capacitance of the bottom wall lumped into
it. This portion of the mug capacitance is easy to account for by lumping it with the
coffee, rather than including it with the mug capacitance as was done in the 2-node
mug of coffee example.
6.2 Defining the Grid (i.e., Discretization of Space) 213

Fig. 6.1 RC network for the mug side wall discretized into n-nodes

The two end nodes (1 and n) are not placed on the mug surface, but rather are
placed inside the volume of the section of the mug it represents. Therefore, the inner
and outer surfaces of the mug are shown as filled circles. These are surfaces, not
volumes, and therefore have no capacitance associated with them. Node 1 has two
parallel channels: one between the node and the coffee and the other between the
node and the air inside the rim. Each of these channels consists of a convective and
a conductive resistance in series. This detail is a result of a “define volumes, then
place nodes” or finite volume approach, as opposed to a “place nodes, then define
volume” approach typical of a finite difference method. It gives additional infor-
mation (really the accuracy of n + 2 nodes) plus added numerical stability for the
explicit method to be used. Notice that the distance between the inner surface and
node 1 is approximately half the distance between node 1 and 2, so these resistance
values differ by a factor of about 2. It is not exactly 2 due to the radial geometry.
The last node (n) has a conductive resistance to the outer surface in series with a
convective/radiative parallel channel.

6.2 Defining the Grid (i.e., Discretization of Space)

There are different ways that the nodal volumes can be defined. There is not
necessarily a “right” way to do it, although the method should be self-consistent.
The approach taken here is to define all the nodes as cylindrical shells of equal
height (from the floor to the rim, H0) and equal capacitance. As a result, the spacing
between nodes decreases from the inside to the outside due to the radial geometry.
An alternative approach would be to space the nodes equally, and then their
volumes would increase with radial position. In this control volume approach to
the spatial discretization, the locations of the element boundaries are first defined by
the chosen volumes, and then each node is placed in the center of the volumes
214 6 Multi-Node Transients

Fig. 6.2 Top view of a

cylindrical shell and its
equivalent as a rectangular
wall of variable width

inside. Conceptually, the node could be placed at the centroid of the nodal element
(for non-square elements).
Figure 6.2 shows a top view of a cylindrical shell imagined to be cut and
opened up to as a wall of uniform thickness (Δr) and a width that varies from 2πr
to 2π(r + Δr). The height (into the page) is also uniform (H0).
Figure 6.3 defines the geometry for node 1. The distances to the radial position of
the boundary of the element are shown above (r1 and r1+), and the distance to the
node is shown below (r1). The distance to the inner boundary is simply the radial
distance from the centerline to the inner surface of the mug:

D
r 1 ¼
2
In this particular approach, the volumes of each element are constrained to be equal
to each other, namely the total volume of the mug side wall divided by the number
of nodes (n):
h i
π ðD þ 2wÞ2  D2 H 0
V node ¼
4n
The height of the node is taken to be the measured outer height of the mug (H0).
Conceptually, that would be the inner height plus the mug thickness if the mug were
uniformly thick. However, the actual mug under investigation has a bottom rim, and
the exterior height is therefore a little larger. The distance to the outer boundary of
node 1 is also the inner boundary of node 2 and is determined by defining its
volume:
6.2 Defining the Grid (i.e., Discretization of Space) 215

T∞ Ras,i R1s,a

r1−

r1+ = r2−

T Rcs R1s 1
R12 2

C1

r1

Fig. 6.3 Geometry of node 1, the inner surface node

 
V node ¼ π r 22  r 21 H o

Solving for the radial position of the boundary between node 1 and 2:
rffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
V node
r 2 ¼ r 21 þ
πHo
The node itself is placed at the centroid of the volume, defined by:

rð
2 rð
2 2
r3  2π  3 
r1 A ¼ rdA ¼ r ð2πrdr Þ ¼ 2π  ¼ r 2  r 31
3 r1 3
r1 r1
 
2 r 32  r 31
r1 ¼  2 
3 r 2  r 21

Alternatively, simply placing the node halfway between would actually be just
as good. . .
216 6 Multi-Node Transients

Consider next node 2. Its volume is the same as node 1 (Vnode) and is defined by:
 
V node ¼ π r 23  r 22 H o

Inserting the result for r2 from node 1 analysis:

V node V node V node

r 23 ¼ þ r 22 ¼ þ þ r 21
πH o πH o πH o
rffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
2V node
r 2þ ¼ r 3 ¼ þ r 21
πH o
The location of node 2 (the centroid) is:
 
2 r 33  r 32
r2 ¼  2 
3 r 3  r 22

The results for node 3 are:

rffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
3V node
r 3þ ¼ r 4 ¼ þ r 21
πH o
and
 
2 r 34  r 33
r3 ¼  2 
3 r 4  r 23

For the general interior node “i”:

rffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
iV node
r iþ ¼ r ðiþ1Þ ¼ þ r 21
πH o
The location of the centroid is:

2 r 3ðiþ1Þ  r 3i
ri ¼ 
3 r 2ðiþ1Þ  r 2i

For the last node, n, the distance to the outer boundary is the outer radius of
the mug:

D
r nþ ¼ þw
2
6.3 Nodal Equations for Explicit Method (Euler) 217

1 2 3 4 5 6 7 8

0.04 0.041 0.042 0.043 0.044 0.045 0.046

Distance from Mug Center (m)

Fig. 6.4 Locations of nodes and boundaries for an 8-node model applied to mug of coffee
problem

The distance to the centroid is:


2 ðD=2 þ wÞ3  r 3n
rn ¼ 
3 ðD=2 þ wÞ2  r 2n

Figure 6.4 shows the precise radial locations of all boundaries of each nodal
element (to scale, horizontally) as dotted lines, except for the inner and outer
surfaces (shown as solid lines) for the 8-node case applied to the mug of coffee
problem. The length of each vertical line is proportional to the surface area of the
boundary. The area increases with radial position because of the increasing circum-
ference. The height of each “wall” is the same, however. To the naked eye, the
nodes may appear equally spaced, but they are in fact closer together as the distance
from the mug center increases. In this case, the mug wall thickness (0.004 m) is
small compared to the inner radius (0.041 m), making a thin-wall approximation
(not taken here) reasonable. The surface area of inner boundary (line length)
increases slightly with radial distance.

6.3 Nodal Equations for Explicit Method (Euler)

In general, the transient thermal energy balance for a fixed volume element “i” in
thermal contact with neighbors “j” is:

dT i X Tj  Ti
Ci ¼ Q_ i þ
dt j
Rij
218 6 Multi-Node Transients

Q_ i represents heat generation, which could come from an internal chemical reaction
(conversion of chemical to thermal energy), from the absorption of radiation, from
I2R heating (conversion of electrical to thermal energy), from microwaving, etc. In
the problem at hand, there are no heat sources. Rij represents the thermal resistance
between node “i” and its neighbor in question “j”. The detailed engineering work is
in estimating the resistance values (which may be variable).
Discretizing time using a forward finite difference approximation:

Ti
ðpþ1Þ
 Ti
ðpÞ
ðpÞ X T ðj pÞ  T ði pÞ
¼ Q_ i þ
ðpÞ
Ci
ΔtðpÞ j
ðpÞ
Rij

Rearranging for the future temperature of node “i”:

!
ΔtðpÞ ðpÞ X T ðj pÞ  T ði pÞ
Q_ i
ðpþ1Þ ðpÞ
Ti ¼ Ti þ ðpÞ
þ ðpÞ
Ci j Rij

For a system with n total capacity-carrying temperature nodes (as opposed to nodes
fixed by a battery, or a surface node), there are n equations. Starting with an initial
temperature, a numerical step size is chosen, and all future temperatures are calculated.
Recall from Chap. 1 that the explicit formulation places a strict limit on the
numerical step size chosen. Each node has a different time constant associated with
it which represents how quickly it responds to its immediate neighbors. From
Chap. 1, the time constant for a node toward an equilibrium value with its neighbors
at a fixed temperature is given by:

ðpÞ
ðpÞ C
τi ¼ Xi
1
ðpÞ
j Rij

The maximum step size for the system that can be used in an explicit formulation is
the smallest of these:

ðpÞ
Δtðmax
pÞ
¼ minτi

If a step size larger than this limit is used, the node with the minimum time constant
will overshoot its equilibrium value.
The nodal equations for all nine variable temperature nodes (coffee plus node
1, six interior nodes, and node 8) are:
For the coffee:

ðpÞ ðpÞ ð pÞ ðpÞ

!
ðpþ1Þ ðpÞ ΔtðpÞ T 1  T coffee T ð1pÞ  T coffee T ð1pÞ  T coffee
T coffee ¼ T coffee þ þ þ
ðpÞ
Ccoffee Rcs þ Rs1 Rtop Rbottom
6.3 Nodal Equations for Explicit Method (Euler) 219

For node 1:

ðpÞ ðpÞ
!
ðpÞ ðpÞ ðpÞ
ðpþ1Þ ðpÞ ΔtðpÞ T coffee  T 1 T  T1 T1  T1
T1 ¼ T1 þ þ 2 þ
C1
ðpÞ Rcs þ Rs1 R12 Ras, i þ Rs1, a

For interior nodes i (nodes 2–7):

ðpÞ ðpÞ
!
ðpÞ ðpÞ
ðpþ1Þ ðpÞ ΔtðpÞ T i1  T i T  Ti
Ti ¼ Ti þ ðpÞ þ iþ1
C R i, i1 Ri, iþ1
i

For node 8:
!
ðpÞ ðpÞ ðpÞ
ðpþ1Þ ðpÞ ΔtðpÞ T7  T8 T1  T8
T8 ¼ T8 þ þ
C8
ðpÞ R78 R8s þ Rs1

The resistance Rs1 is the equivalent resistance of the parallel convective/radiative

channel between the outer surface and ambient. The two surface temperatures
(surface nodes 0 and 9, which have no capacitance) are determined from a voltage
divider analysis. For the inner surface:

T coffee  T 1 T coffee  T 0
qcoffee, 1 ¼ ¼
Rcs þ Rs1 Rcs
Solving for the surface temperature:

Rcs  
T 0 ¼ T coffee  T coffee  T 1
Rcs þ Rs1
For the outer surface:

T8  T1 T9  T1
q81 ¼ ¼
R8s þ Rs1 Rs1
Rs1
T9 ¼ T1 þ ðT 8  T 1 Þ
R9s þ Rs1
The final step before implementing the method in a computer simulation is to define
the capacitances and thermal resistances in terms of system parameters, detailed in
the following paragraphs. Results for the coffee/mug problem are listed in Table 6.1,
and the various entries are discussed as they arise.
220 6 Multi-Node Transients

Table 6.1 Nodal geometry, capacitance, resistance, and step-size restriction at the start of the
simulation

Quantity Symbol unit Node

coffee "0" 1 2 3 4 5 6 7 8 "9"
inner boundary radius r- m NA NA 0.0413 0.0418 0.0423 0.0428 0.0433 0.0438 0.0443 0.0448 NA
node radius r m NA 0.0413 0.0415 0.0420 0.0426 0.0431 0.0436 0.0440 0.0445 0.0450 0.0453
Inner Boundary Area A- m2 NA NA 0.0199 0.0278 0.0282 0.0285 0.0288 0.0292 0.0295 0.0298 0.0301
o
Capacitance C J/ C 1757 0 29.2 29.2 29.2 29.2 29.2 29.2 29.2 29.2 0
o
Resistance (to the right) R+ C/W VARIES 0.0131 0.0185 0.0180 0.0176 0.0172 0.0168 0.0165 0.0161 0.0079 VARIES
Maximum step size dtmax sec 22.871 NA 0.413 0.266 0.260 0.254 0.248 0.243 0.238 0.469 NA

Varies with time. Initial value is shown, which yields a minimum step size

The shaded cells indicate that these step sizes vary with time and must be confirmed to not control
the system at a later time

6.3.1 Capacitances

The capacitance of each node is the same in this particular implementation:

Cnode ¼ ρmug cmug V node

Each node has a capacitance of 29.2 J/ C, and the total capacitance of the mug
side wall is 8(29.2) ¼ 234 J/ C, which is 87.3 % of the total capacitance of the
mug (the remainder at the bottom which is lumped with the coffee).
The capacitance for the coffee is that calculated for the 1-node cases, plus the
capacitance of the bottom of the mug:


Ccoffee ¼ ρcoffee ccoffee V coffee þ mmug cmug V mug  n  Cnode

The capacitance of the coffee, 1,757 J/ C, is (not surprisingly) much larger than the
capacitance of each node in the mug wall. Inclusion of the bottom of the mug has a
small (2.9 %) effect on the coffee capacitance for this case study and could be
neglected. However, it is easy to incorporate it, and it might be more important for
other applications, so it is included.

6.3.2 Resistances

All the resistance values are shown in Table 6.1 (except for the resistance between
node 1 and the air inside the rim of the mug, which is treated separately). They are
derived one at a time, from left to right. Starting with the convective thermal
resistance between the coffee and the inner wall of the mug:

1 1
Rcs ¼ ¼
hcs Ar1 hcs πDðH  LÞ

In this expression, H is the depth of the coffee on the inside and L is the vertical
distance of the air column on the inside of the mug, making H  L the vertical
dimension of the contact area between the coffee and the inner surface of the mug.
6.3 Nodal Equations for Explicit Method (Euler) 221

The convection coefficient, hcs, results from a natural convection process between
the coffee and the inner wall of the mug. From the Nusselt number correlation of a
future chapter:

 T 0:25

coffee  T 1
hcs ¼ 32:38T film
0:441
0:25
ðH  LÞ

The film temperature is the average of Tcoffee and T1.

The thermal resistance between the inner surface (node “0”) and the first
elemental node (node 1) is a conductive resistance and is given by:

r 1  r 1 r 1  r 1
Rs1 ¼  ¼
kmug Ar12þAr1 πkmug ðr 1 þ r 1 ÞðH  LÞ

The numerator is the spacing between nodes 0 and 1, and the area is the geometric
average between the two. This thermal resistance is in series with the convective
resistance, and it is conceptually half that of a thermal resistance between two
interior nodes (from Table 6.1, R01 ¼ 0.0094  C/W, while R12 ¼ 0.0185  C/W).
For the resistance between node 1 and ambient through the inner surface of the
mug rim:

1 r 1  r 1
Ras, i þ Rs1, a ¼ þ
heff πDL πkmug ðr 1 þ r 1 ÞL

In this expression, heff represents the sum of the convective and radiative coefficients
between the surface and the ambient. The same value as calculated for the outer
surface will be used. There are admittedly several inconsistencies in how this thermal
channel is treated. The philosophy is that the exposed surface area in question is a
significant (but not dominant) channel for heat loss, and it is accounted for in a
reasonable approximation by the approach taken. Modeling is an art.
Continuing left to right, the conductive resistance between nodes 1 and 2 is:

r2  r1
R12 ¼
kmug 2πr 2 H 0

Generally, between nodes “i” and “i + 1”:

r iþ1  r i
Ri, iþ1 ¼
kmug 2πr ðiþ1Þ H 0

It would be just as easy to use the exact conductive formula for a cylindrical
geometry, but the more general approach is taken here. From Table 6.1, the
conductive resistance between adjacent interior nodes decreases with radial posi-
tion. There are two reasons for this trend. First, the nodes are closer together, easing
conduction. Second, the surface area across which heat is conducting is larger,
again easing conduction.
222 6 Multi-Node Transients

The conductive thermal resistance between the last element node (“n” in general,
“8” in this implementation) and the surface node (“s” in general, “9” for this
implementation) is:

r9  r8 r9  r8
Rs , n ¼ R9 , 8 ¼ Ar8þ þAr8  ¼
kmug 2
πkmug ðr 9 þ r 8 ÞH 0

Finally, the convective resistance between the outer wall surface (“s” ¼ “9”) and the
ambient air is:

1 1
Rs, 1 ¼ ¼
hs1 Ar9 hs1 π ðD þ 2wÞH 0

The 1-node models taught that the overall system response is very sensitive to the
choice of convection and radiation coefficients between the outer wall and ambient.
This sensitivity is due to the large surface area (a parallel resistance concept), and
because this combined resistance is the largest resistance in the side channel
(a series resistance concept). Therefore, extra attention is warranted in modeling
these parameters. In this numerical simulation, it is relatively easy to account for
heat transfer coefficients that vary with temperature (and therefore time) by calcu-
lating them at each time step. Conceptually, the temperature used should be for the
slightly different node 9, not 8, but that would require an iterative loop, making a
spreadsheet simulation impossible.
For parallel convective and radiative resistances acting on a surface of area A,
with coefficients hconv and hrad, the equivalent resistance is:

1 1 1
¼ þ ¼ hconv A þ hrad A ¼ heff A
R Rconv Rrad
That is, the convective and radiative coefficients add together to form an effective
heat transfer coefficient. If the ambient air immediately adjacent to the mug and
the walls of the room are at significantly different temperatures, then the model
would have to be adjusted by adding a separate channel for radiation and
convection.
The convective coefficient in this case is modeled as a natural convection of air
for a vertical cylinder of height H0, with fluid properties evaluated at the film
temperature. The simplified formula for has from the Nusselt number correlation
(with temperature in  C and height in m) is:

  jðT 8  T 1 Þj0:25
hair ¼ 1:5646  0:0015T film
H 0:25
o

The leading term represents the combined fluid properties and their relatively mild
temperature dependency.
6.3 Nodal Equations for Explicit Method (Euler) 223

The radiative coefficient, with an emissivity ε as derived in Chap. 3, is:

h i
hrad ¼ εσ ðT 8 þ 273Þ2 þ ðT 1 þ 273Þ2 ½ðT 8 þ 273Þ þ ðT 1 þ 273Þ

where the temperatures are in Celsius. Again, technically speaking, the surface
temperature (T9) should be used. However, that would require an iterative solution
for each time step, and that would make a spreadsheet implementation problematic.

6.3.3 Step-Size Restriction

In an explicit formulation, each capacitive node has a step-size restriction equal to

its characteristic response time, ReqC, where Req is the equivalent parallel resis-
tance of all its neighbors. The step size used to march forward in time cannot exceed
the smallest of these. That is, the maximum step size that can be used in an explicit
formulation is the minimum time constant of any individual node.
For nodes that have variable resistances, the maximum step size will also
vary with time. In Table 6.1, the shaded cells contain maximum step sizes that
vary throughout the simulation. The values shown are minimum values, based on
early time, when the convection and radiation coefficients are highest (giving
lowest resistances, and shorter response times). It is apparent that these variable
values are not going to control the step size for this application. It is conceivable
that they could, however, for other input parameter values, and could be checked at
each step. It is relatively easy to implement a variable step-size algorithm, in which
the step size chosen at a given step is a user input-defined fraction of the maximum
step size within the system of equations.
In this case, node 7 clearly limits the step size used in the system formulation to a
value of 0.238 s. Contrast that restriction with the 2-node numerical simulation, in
which case the mug as a lumped node yielded a step-size restriction of 31.8 s.
A comment on the computational requirements is in order. In order to simulate a
fixed total time (tsimulation), the computations per simulation can be expressed
verbally as:
  
Computations Computations Steps time
¼  
Simulation Step time Simulation

The computations per step is a fixed number of times the total number of nodes.
That is:

Computations
n
Step

The steps per simulation is the inverse of the step size:

   
Steps Step time 1
¼  ¼  tsimulation
Simulation time Simulation Δt
224 6 Multi-Node Transients

The step size ¼ Reff,i Ci

C ~ 1/n (the total capacitance divided by number of elements)
R ~ 1/n (the distance between nodes is the total distance divided by number of
elements)
Therefore:
Δt ~ 1/n2

Computations  
 ðnÞ  n2  ðtsimulation Þ
Simulation
That is, if the number of nodes is doubled, the computational requirement increases
by a factor of 8.

6.4 Spreadsheet Implementation

This model was implemented into a spreadsheet simulation (Table 6.2) using a
9-node model (eight nodes in the mug, plus the coffee) and stepping in time with the
maximum step size of 0.238 s. Initial conditions and fixed parameters are shaded.
The maximum step size (dtMax) refers to the appropriate cell in Table 6.1, and the
numerical input parameter “f”, set to 1, is the fraction of the maximum step size
used to define dt, the step size taken. It will be demonstrated why a smaller step
size than that is not required for numerical accuracy for this case. In this model,
three heat transfer coefficients depend on surface temperature (hcoffee, hconv, air, and
hrad, air) and therefore must be calculated at each time step. They are placed to the
right, and then the appropriate thermal resistance values are calculated. In the
second row, the Euler formulas are entered (for coffee and nodes 1–8), and then
all equations are copied down for as many rows as is desired.
Temperature/time plots are presented followed by a series of plots of tempera-
ture vs. radial position at fixed times. The heat transfer coefficients are plotted to

Table 6.2 Structure of spreadsheet simulation showing the first 3 s

dtMax= 0.2375
f= 1
dt (sec) = 0.2375 AIR SIDE
time(s) time(min) coffee "0" 1 2 3 4 5 6 7 8 "9" hcoffee Rcs hconv hrad heff Rinf Tsurface
0.00 0.0000 92.00 36.29 21.00 21.00 21.00 21.00 21.00 21.00 21.00 21.00 21.00 1058 0.048 0.00 5.19 5.19 6.40 87.05
0.24 0.0040 91.84 43.72 30.53 21.00 21.00 21.00 21.00 21.00 21.00 21.00 21.00 1057 0.048 0.00 5.19 5.19 6.40 86.90
0.48 0.0079 91.70 46.81 34.55 25.20 21.00 21.00 21.00 21.00 21.00 21.00 21.00 1053 0.048 0.00 5.19 5.19 6.40 86.77
0.71 0.0119 91.57 49.50 38.07 27.42 22.89 21.00 21.00 21.00 21.00 21.00 21.00 1048 0.048 0.00 5.19 5.19 6.40 86.65
0.95 0.0158 91.45 51.35 40.51 30.07 24.06 21.88 21.00 21.00 21.00 21.00 21.00 1043 0.048 0.00 5.19 5.19 6.40 86.54
1.19 0.0198 91.34 52.99 42.67 31.96 25.76 22.47 21.41 21.00 21.00 21.00 21.00 1038 0.048 0.00 5.19 5.19 6.40 86.43
1.43 0.0238 91.23 54.29 44.38 33.88 27.04 23.49 21.71 21.20 21.00 21.00 21.00 1034 0.049 0.00 5.19 5.19 6.40 86.33
1.66 0.0277 91.12 55.45 45.93 35.42 28.49 24.29 22.31 21.35 21.10 21.00 21.00 1029 0.049 0.00 5.19 5.19 6.40 86.23
1.90 0.0317 91.02 56.43 47.23 36.92 29.67 25.29 22.78 21.69 21.17 21.05 21.05 1025 0.049 1.30 5.19 6.48 5.12 86.14
2.14 0.0356 90.93 57.32 48.42 38.19 30.92 26.13 23.44 21.96 21.37 21.11 21.11 1021 0.049 1.59 5.19 6.77 4.90 86.05
2.38 0.0396 90.83 58.09 49.45 39.41 31.99 27.07 24.00 22.38 21.53 21.24 21.24 1017 0.049 1.92 5.19 7.11 4.67 85.96
2.61 0.0435 90.74 58.80 50.41 40.48 33.06 27.89 24.67 22.74 21.80 21.39 21.39 1014 0.050 2.16 5.20 7.36 4.51 85.88
2.85 0.0475 90.65 59.44 51.26 41.51 34.02 28.76 25.26 23.21 22.06 21.60 21.60 1010 0.050 2.41 5.20 7.61 4.36 85.79
3.09 0.0515 90.57 60.03 52.05 42.42 34.97 29.54 25.92 23.63 22.40 21.83 21.83 1007 0.050 2.61 5.21 7.82 4.24 85.71

The shaded cells contain initial conductions and formulas are embedded in all other cells
6.6 Radial Temperature Profiles 225

observe both their values as well as their change over time. Finally, a consideration
of the numerical issues is presented, demonstrating both spatial and temporal grid
effects.
The workbook written is rather large (15 MB) and it only simulates 36 min of the
entire event. The number of time steps required to simulate 36 min is:
 
Time elapsed 36 min  60 s 
Number of steps ¼ ¼  ¼ 9, 075
Step size 0:238 s min

63,025 steps would be required to simulate the entire 250 event, resulting in a
spreadsheet of over 100 MB. In short, this simulation taxes the limit of spreadsheet
applications, and a shift to a computer program may be warranted.
The input and derived parameters are listed in Fig. 6.5.

6.5 Temperature/Time Plots

Figure 6.6 shows the predicted temperature vs. time for locations that correspond to
the measurements at the mug-free surface and mug side wall during the first 10 min.
The predicted rise of the outer side wall temperature and fall of the free surface both
lag from the measured values, somewhat, but in general, the agreement is consid-
ered to be quite good. The 2-node model (which used fixed resistance values) gave
similar results. However, the 9-node model gives more detail, as shown in the
following paragraphs.
A plot of all nine nodal temperatures, plus the inner and outer surface
(non-capacitive nodes) vs. time is presented in Fig. 6.7, without the experimental
data. The closer each node is to the inner surface, the earlier its temperature rises.
The same plots are shown in Fig. 6.8 on semi-logx axes. This plot more clearly
shows how it takes time before the first “signal” that the inner wall has been
exposed to hot liquid reaches a given location. That is, a thermal “wave” propagates
outward from the inner wall. The initial conditions cannot be shown on the latter
plot (time equals zero cannot be plotted on a logarithmic scale).

6.6 Radial Temperature Profiles

A series of similar plots that show the temperature vs. radial position at fixed times
is presented in Figs. 6.9, 6.10, 6.11, 6.12. On the coffee and air side, hypothetical
thermal boundary layers are shown using several columns and an exponential fit for
illustrative purposes.
The first plot shows the first 3 s, during which time the outer wall never
experiences a change. During this period of time, the wall of the mug behaves as
if it were infinitely thick. In traditional heat transfer texts, analytic solutions of heat
transfer into an infinitely thick wall are presented. Those solutions are for flat walls
with constant convection coefficients, whereas this one is for a radial geometry with
226 6 Multi-Node Transients

Fig. 6.5 Input and derived fixed parameter values

variable convection coefficients (which would be impossible to solve analytically).

This simulation shows the basic response, and gives a sense that it is in fact possible
to have an infinitely thick wall response, provided the elapsed time is sufficiently
short to justify it. So there is a relationship between space and time in transient
conduction problems. Short time is like long distance.
The first 30 s sees the temperature of the outer wall increasing with time, and
retaining a general concave up shape. This problem would be extremely hard, if not
impossible, to solve analytically. The first 3 min shows the mug wall reaching a
maximum, with a subsequent gradual falling of the coffee and mug wall together
6.6 Radial Temperature Profiles 227

100
90
80
Temperature (oC)
70
60
50 coffee
40 top
side
30
Top Data
20 Side Data
10
0
0 1 2 3 4 5
Time (min)

Fig. 6.6 9-Node model predictions during mug heating phase

100
90
80
70
Temperature (oC)

coffee
60 top
side
50
Inner Surface
Radial Position Node 1
40
Node 2
Node 3
30
Node 4
Node 5
20
Node 6

10 Node 7
Node 8

0
0 0.5 1 1.5 2
Time (min)

Fig. 6.7 Nodal temperatures vs. time for the first 2 min

during the first 30 min. This behavior maintains itself until the entire event is over,
when the mug reaches equilibrium. During this period, there a very mild tempera-
ture drop from the coffee to the inner surface, another mild drop across the mug
wall, and then a large drop from the outer surface to ambient. This behavior is due
to the relative resistance values of the side channel.
What this behavior suggests, rather strongly, is that using eight nodes in the
coffee mug is total overkill to do this particular job. The main objective here is to
228 6 Multi-Node Transients

100
90
Temperature (oC) 80
70
coffee
60 top
side
50
Inner Surface
Node 1
40
Node 2
Node 3
30
Radial Position Node 4
Node 5
20
Node 6

10 Node 7
Node 8

0
0.001 0.01 0.1 1 10
Time (min)

Fig. 6.8 Nodal temperatures vs. time for the first 10 min on semi-logx axes

100 Time (seconds)

90 Nodes
2.85
80
2.38
Temperature (oC)

70 1.90
60 1.43
50 0.95
0.48
40
0.00
30
20
10
0
0.03 0.035 0.04 0.045 0.05
Radial Position (m)

Fig. 6.9 Temperature vs. radial position for the first 3 s. Symbols shown for the latest curve to
indicate nodal positions

teach the method, but good engineering judgment would suggest that the 2-node
model (with the finite volume approach to yield surface behavior) is adequate for
this application. And the extended lumped model is sufficient for the entire event
(but not for the early mug heating phase).
6.6 Radial Temperature Profiles 229

100 Time (seconds)

Nodes
90
32.07
80 26.60
70 21.38
Temperature (oC)

16.15
60
10.69
50
5.46
40 0.00
30
20
10
0
0.03 0.035 0.04 0.045 0.05
Radial Position (m)

Fig. 6.10 Temperature vs. radial position for the first 30 s. Symbols shown for the latest curve to
indicate nodal positions

Time (minutes)
100
Nodes
90
3.19
80 2.66
70 2.13
Temperature (oC)

1.60
60
1.06
50 0.53
40 0.00

30
20
10
0
0.03 0.035 0.04 0.045 0.05
Radial Position (m)

Fig. 6.11 Temperature vs. radial position for the first 3 min. Symbols shown for the latest curve to
indicate nodal positions
230 6 Multi-Node Transients

Time (minutes)
100
5.32
90
10.63
80 15.94
70 21.25
Temperature (oC)

26.57
60
31.88
50 Nodes
40 0.00

30
20
10
0
0.03 0.035 0.04 0.045 0.05
Radial Position (m)

Fig. 6.12 Temperature vs. radial position for the first 30 min. Symbols shown for the latest curve
to indicate nodal positions

20
Heat Transfer Coefficient (W/m2/ oC)

18 Net
Radiation
16 convection
14
12
10
8
6
4
2
0
0 1 2 3 4 5
Time (min)

Fig. 6.13 Heat transfer coefficients on the outer surface of the mug vs. time for the first 5 min. If
simulated for the entire cooling event, the convection coefficient will go to zero, while the
radiation coefficient will remain finite

6.7 Heat Transfer Coefficients

The heat transfer coefficients are shown as functions of time for the first 3 min on
the air side in Fig. 6.13 and on the coffee side in Fig. 6.14. At time zero, when
the outer wall is at ambient, the natural convection coefficient is zero, according to
6.8 Numerical Stability and Accuracy 231

1200

Heat Transfer Coefficient (W/m2/oC)

1000

800

600

400

200

0
0 1 2 3 4 5
Time (min)

Fig. 6.14 Heat transfer coefficients on the inner surface of the mug vs. time for the first 5 min

the model. However, the radiation coefficient is not zero initially, and would control
the overall heat transfer rate during periods of time where the solid surface
temperatures are nearly equal to the adjacent fluid. This behavior emphasizes the
importance of including radiation when natural convection predominates. As the
outside surface wall temperature rises in the first 30 s, the convection coefficient
rises and becomes almost equal to the radiative. The effective heat transfer coeffi-
cient is the sum of the two and equals approximately 14 W/m2/ C.
In contrast, the convection coefficient on the water side starts high because the
mug and coffee are at very different temperatures early. The inner mug wall surface
temperature rises rapidly initially, and therefore the temperature difference driving
the natural convection inside decreases. The convection coefficient inside therefore
starts at a value just above 1,000 W/m2/ C and falls rapidly to a value of approxi-
mately 500 W/m2/ C.
As time proceeds, the two natural convection coefficient values gradually fall as
both the mug and coffee approach ambient. The model should really build in a
lower cap to the convection coefficients.

6.8 Numerical Stability and Accuracy

Figures 6.15 and 6.16 show the same plots as Figs. 6.8 and 6.9, respectively, but
with the step size chosen to be only 10 % larger than the minimum step size (0.262 s
as opposed to 0.238 s). It is apparent that an instability arises, and it starts at node
7, and emanates from there. This instability reveals the biggest problem with an
explicit formulation, especially as the spatial grid is made finer.
On the other hand, a comparison of the model prediction for node 7 is made
when the step size is reduced by an order of magnitude in Fig. 6.17. There are
232 6 Multi-Node Transients

100

90

80
coffee
70 top
Temperature (oC)

side
60 Inner Surface
Node 1
50 Node 2

Radial Position Node 3

40 Node 4
Node 5
30 Node 6
Node 7
20 Node 8

10

0
0 0.5 1 1.5 2
Time (min)

Fig. 6.15 Simulation with a step size of 1.1 times the maximum allowable step

Time (seconds)
100
Nodes
90
35.27
80 29.26
70 23.51
Temperature (oC)

17.77
60
11.76
50
6.01
40 0.00
30

20

10

0
0.03 0.035 0.04 0.045 0.05
Radial Position (m)

Fig. 6.16 Radial profile for first 30 s with a step size 1.1 times the minimum

actually two indistinguishable curves plotted, one when f ¼ 1 and the other for when
f ¼ 0.1, a reduction in step size of an order of magnitude. This plot demonstrates
temporal grid insensitivity exists when the maximum allowable step size is used in
this case.
6.8 Numerical Stability and Accuracy 233

100
90
80
Node 7 Temperature (oC)

70
60
50
f=1
40
f = 0.1
30
20
10
0
0 0.2 0.4 0.6 0.8 1
Time (sec)

Fig. 6.17 Effect of step size on node 7. The two curves are indistinguishable

6.8.1 Implicit Method

The general governing equation (assuming constant capacitances, but variable

thermal resistances) for node “i” in thermal contact with nodes “j” written as a
backward-difference approximation, and upgrading (p) to (p + 1) is:

ðpþ1Þ
Ti  Ti
ðpÞ
ðpþ1Þ X T ðj pþ1Þ  T ði pþ1Þ
Ci ¼ Q_ i þ
ΔtðpÞ j
ðpþ1Þ
Rij

Isolating all terms for node “i” evaluated in the future (p + 1):

ðpþ1Þ X T ðj pþ1Þ Ci T i
ðpÞ
Q_ i þ ðpþ1Þ
þ
j Rij
ΔtðpÞ
ðpþ1Þ
Ti ¼
Ci X 1
þ
Δt j R
ðpþ1Þ
ij

This expression can be used as a recursion relation for an iterative procedure,

called, in general the Method of Successive Substitution. The method is developed
more formally in later chapters, and outlined here.
An initial “guess” for all the unknown future temperatures (and the associated
thermal resistances) is made, which for transient problems is logically the present
value, and then an iterative loop that calculates an “improved guess” for all the
temperatures is made, tests for convergence, and then either executes another itera-
tion or exits the iteration loop and saves the temperatures. Testing for convergence
234 6 Multi-Node Transients

requires the change in all temperatures between subsequent iterations be calculated,

and setting the largest change to be less than a defined criterion.
This procedure requires a nested loop that cannot be executed in a spreadsheet
(without using macros). The outer loop steps forward in time, the inner loop is an
iterative loop needed to solve for the temperatures one step into the future.

Workshop 6.1. Transient Heat Transfer Fins

A “heat sink” (Fig. 6.18) is mounted onto a 6 cm  6 cm (w  w) electronic

component that generates 250 W (q) when operating. The heat sink has a 0.6 cm
thick (b) base plate that fits exactly onto the component. Sixteen (M) rectangular
“fins” of width 6 cm (w), thickness 0.10 cm (t), and length 4 cm (L) are mounted
onto the base plate with an air gap of 0.275 cm (g) between each fin. The fins are
cooled by a fan (not shown) that provides sufficient ventilation to maintain the air
temperature in the gaps at ambient temperature (T1 ¼ 25  C). Compare the tran-
sient performance of heat sinks of copper, aluminum, and steel, with thermal
properties listed in the following table:

Material k (W/m/K) ρ (kg/m3) c (J/kg/K)

Copper 400 8,900 390
Aluminum 200 2,700 910
Steel 20 7,900 490

Write a spreadsheet or program code to simulate the thermal response when the
component is powered on, allowed to reach an effective steady-state, and then
turned off and allowed to cool until it is effectively back to ambient. Treat the base
plate as a uniform temperature lumped node (neglect its conductive resistance) that
receives the generated heat and transfers it to the fins, which are assumed to behave
identically. Analyze a single fin by breaking it into “n” equal sized nodes and

g/2 g/2
t

air fin air L

L Fin
W

Base plate
base plate b

W electronic component
b

Fig. 6.18 Schematic of heat sink (left) and an end view of an isolated fin (right)
Workshop 6.1. Transient Heat Transfer Fins 235

performing a transient analysis using the explicit method. Assume the fin (and base
plate) is exposed to ambient with a fixed heat transfer coefficient h. Radiation is
neglected in this case because the fins “see” each other.
The main objectives of this exercise are to successfully implement an “n-node”
transient analysis, and to gain an understanding of the transient and steady-state
response of heat transfer fins. At a minimum, prepare and discuss well-formatted
plots of base plate temperature vs. time (with all materials on the same plot), steady-
state temperature vs. position (distance from the base plate, all three materials).
Also, consider temperature vs. position at fixed times for aluminum. Conduct a
study of the effect of h (range from 10 to 100 W/m2/K to simulate natural
convection to a “strong crosswind” using aluminum.
Note: A specific 5-node model is developed on the next pages. It is advised that you
develop your own model first before consulting that model. Then decide which
model to implement.

Model Development
A 5-node RC network (base plate segment plus n ¼ 4 equal sized nodes of the fin) is
shown in Fig. 6.19 for a single fin. The term “longitudinal” will be used to designate
a distance along the fin length (L or Δx), and “lateral” will refer to across the fin (t).

Fig. 6.19 Schematics for 5-node transient fin problem. The top figure defines the nodal volumes
used. The bottom shows the RC network (the light lines define the nodal volumes)
236 6 Multi-Node Transients

The fin is broken into “n” (4 in this schematic) equally sized longitudinal elements, with
a recognized symmetry so that only half of the fin need be considered. The base plate
node is just that section associated with the fin being analyzed. Capacitances are
assigned to each node and thermal channels between them are shown in the RC network.
The symmetry has been used on the fin, accounting for the “q/2” being delivered to
each fin, with M being the total number of fins and q the total heat dissipated.
The equations will be developed with a general form with the fin being broken
into “n” equally sized nodes. Therefore, the length of each node, which also equals
the distance between them is given by Δx ¼ L=n. The distance between the base
plate (node 0) and node 1 of the fin is Δx/2.
The capacitances are all equal to
t  t L
Ci ¼ ρcw Δx ¼ ρcw
2 2 n

The resistance between the base plate directly to the air is by:

1
R01 ¼
hwg=2

The conductive resistance between the base plate and node 1 is:

Δx=2
R01 ¼
kwt=2

Notice that the distance from the base plate to node 1 is half of element length. The
conductive resistances between the other nodes are:

Δx
Ri, iþ1 ¼
kwt=2

The resistances between each node and ambient is given by a series conduction and
convection resistance. For interior node “i”:

1 t=4 1 ht=4 1
Ri, iþ1 ¼ þ ¼ 1þ ¼ ð1 þ BiÞ
hΔx kΔx hΔx k hΔx

where a Biot number has been introduced (the ratio of lateral conductive to
convective resistance). The end node (“n”) has a different surface area, and its
resistance consists of two parallel channels (lateral and longitudinal) each of which
has a conductive and convective resistance in series:

1 1 1
¼ þ
Rn1 1
hΔx þ t=4
kΔx
1
ht=2 þ Δx=2
kt=2
Workshop 6.1. Transient Heat Transfer Fins 237

This resistance could be expressed in terms of the Bi defined above and the ratio of
fin thickness to element length. It is common in fin analysis to neglect the heat
going out the end in comparison to that going out the sides, but is included for
completeness, and practice in defining resistance values.
Using the explicit numerical method, the nodal equations are:
Base Plate:

" #
ðpÞ ðpÞ ðpÞ
ðpþ1Þ ðpÞ Δt q T1  T0 T1  T0
T0 ¼ T0 þ þ þ
C0 2M R01 R01

Node i ¼ 1 to n1:
" ðpÞ ð pÞ
#
ðpÞ ðpÞ ðpÞ
ðpþ1Þ ðpÞ Δt T i1  T i T iþ1  T i T1  Ti
Ti ¼ Ti þ þ þ
Ci Ri, i1 Ri, iþ1 Ri1 ð1 þ Bii Þ

Node n:
" #
ðpÞ
Δt T n1  T ðnpÞ T 1  T ðnpÞ
T ðnpþ1Þ ¼ T ðnpÞ þ þ
Cn Rn1, n Rn1 ð1 þ Bin Þ

In these expressions, the Biot numbers represent the ratio of the conductive to
convective resistance in the direction perpendicular to the direction of the primary
heat flow and can be shown to be negligible for these cases. The time constants are:

1 1
τ0 ¼ C0 þ
R01 R01

1 1 1
τi ¼ Ci þ þ
Ri, i1 Ri, iþ1 Ri1 ð1 þ Bii Þ

1 1
τ n ¼ Cn þ
Rn, n1 Rn1 ð1 þ Bin Þ

The maximum step size for numerical stability is the smallest time constant. Choose
the step size to be a fraction (f) of this limit.
 Part III
Steady-State Conduction
Heat Transfer Fins (and Handles)
7

In this chapter, a lumped capacity approach to steady-state conduction problems is

developed, using the important practical application of heat transfer fins (and
handles). The approach is applied to the coffee/mug problem, this time with half
a mug, and the heat transfer into the rim and out to ambient is detailed.

7.1 Motivation for Fins

Extended surfaces are widely used as either heat transfer fins (devices designed to
increase the overall rate of heat transfer between a solid and a fluid) or handles
(something you can grab without burning your hand). For fin applications, since the
rate of heat transfer by convection (and possibly radiation in parallel) is propor-
tional to the exposed surface area, the basic idea is to increase that exposed surface
area, as suggested in Fig. 7.1. However, because the temperature in the fin decreases
with the distance from the wall (x), there is a limit to how far the fin can extend and
still be effective. A “perfect” fin is one in which the temperature of the fin is the
same everywhere as its base. In a real fin, the average temperature of the fin exposed
to the fluid will be less than that, and therefore the fin transfers heat at a lower rate.
If the fin is long enough, the temperature at the end of the fin approaches the
ambient temperature. That’s a handle you can grab.
An exact analysis of the steady-state 1D heat transfer associated with an
“extended surface” is included in virtually all heat transfer textbooks for the case
where the cross-sectional area is constant with distance from the wall (i.e., the area
of the fin base (t*d) is projected outward from the base). The result for the more
general transient case (the steady-state being a limiting case of the transient
solution) is summarized in Appendix 1. This analysis also includes the inherent
2D lateral heat transfer that is required for heat to flow from the interior of the fin
toward the surface. Appendix 2 develops the governing differential equation that
could be the starting point for a finite difference numerical method for the more

# Springer International Publishing Switzerland 2015 241

G. Sidebotham, Heat Transfer Modeling: An Inductive Approach,
DOI 10.1007/978-3-319-14514-3_7
242 7 Heat Transfer Fins (and Handles)

Fig. 7.1 Motivation for heat

transfer fins. The bare object
is a rectangular rod of length
d (left). Rectangular heat
transfer fins (length L,
thickness t, and depth d) are
added to increase the exposed
surface area

general case with a non-constant-area fin. This method is not formally developed
here, however.
The approach taken in the main body of the chapter is to develop a finite element
approach. First, the fin is considered as a single lumped element, a 1-node model.
This approach yields a simple closed-form solution that is surprisingly versatile and
captures all the key functional dependencies of the exact solution. Next, a multi-
element approach (“few node”) is developed considering the fin broken into three
equal elements. This method is easily extended to two or three dimensions. Finally,
a detailed methodology of the 1-node model is applied to the case of the rim of a
mug of coffee.

Engineering Problem Statement

An aluminum (k ¼ 200 W/m/K) rectangular fin with a thickness t (¼0.001 m),
length L (¼0.1 m), and width w (¼0.25 cm) is attached to a wall (Fig. 7.2). The base
of the fin is maintained at a temperature (To ¼ 100  C). The fin is exposed to a fluid
at T1 (¼0  C) with an effective convection/radiation coefficient h (¼50 W/m2/K).
Determine

• The rate of heat transfer from the base wall, qss.

• The fin efficiency, η, defined as the rate of heat transfer normalized by that of a
perfect fin (defined shortly).
• The fin effectiveness, ε, defined as the rate of heat transfer divided by the rate of
heat transfer that would occur from the base if there were no fin attached.

By specifying k and especially h and implying that the fluid temperature is not
affected by the fin (outside a thermal boundary layer), the accuracy of this model to
a “real life” fin will always be suspect. The main goal here is to compare the
accuracy of the numerical methods on a problem statement that is well posed so a
comparison with an exact solution is possible. The main advantage of the numerical
method will be in addressing problems that do NOT have an exact solution.
7.2 1-Node Fin Analysis 243

Fig. 7.2 Schematic of a

constant-area fin attached to
a wall. A rectangular fin is h, T¥ P=2(t+w)
shown, but the shape of the
cross-section can be
anything, as long as it does
k t w
not vary with distance from
the wall (if it is to be a To L
constant-area fin) A = tw

7.1.1 Perfect Fin

The heat transfer rate from a perfect fin, in which the temperature of the fin equals
the base temperature everywhere, is governed by Newton’s Law of Cooling:

qperfect ¼ hðPL þ AÞðT 0  T 1 Þ

P is the perimeter of the fin, PL is the area of the sides, and A is the area of the
exposed end (and is the base area projected a distance L from the wall). Note that a
constant-area fin can be any shape (area and perimeter) and is not restricted to a
rectangular geometry.

7.1.2 Corrected Length

In a fin, heat flows out the sides and out the end. In most cases, the heat out the sides
is dominant and inclusion of the end complicates the mathematics. Therefore, it is
common to define a corrected length in which the heat out the end is accounted for
in the term PL, as suggested in Fig. 7.3. The added length adds the same area as the
end. That is, PΔL ¼ A, and the corrected length is Lc ¼ L + ΔL ¼ L + A/P. Therefore,
the heat transfer rate from a perfect fin with a corrected length is
qperfect ¼ hðPLc ÞðT 0  T 1 Þ.

7.2 1-Node Fin Analysis

Two 1-node models will be developed; one for fins, and one for handles. The
concepts of “thermally short” for fins and “thermally long” for handles will
naturally emerge.
244 7 Heat Transfer Fins (and Handles)

Fig. 7.3 Schematic of corrected length. The top fin has heat transfer out the end. The bottom fin
has an added length (ΔL) that adds surface area to the sides and has an insulated (adiabatic) end.
It’s not that insulation is placed on the end; it is that heat out the end of the longer fin is neglected

T∞
qfin Rh
T0
Tfin
Rk
Cfin

qfin (L/2) 1
T0 Tfin T∞
kA hPL

Cfin

Fig. 7.4 Schematic and thermal resistance network of a thermally short fin represented by a single
node. This model is called the “thermally short” fin model, for reasons which will become
apparent. The top network is expressed in general terms of conductive and convective resistances.
The bottom network is defined for a constant-area fin

7.2.1 Thermally Short Fin Model

The resistance network of Fig. 7.4 considers the fin as a single element, represented
by a node. This simple network captures the fundamental heat transfer process
associated with fins. Heat (shown as a current source) must flow from the base (at a
temperature T0) into the fin by conduction and from the fin to the fluid by
convection (and radiation in parallel). Alternatively, a battery could be used to
maintain a fixed To, as is conventionally presented. However, since the heat transfer
rate into the fin is the key objective of the analysis, a current source that supplies
that heat is more appropriate. Also, fins are often called “heat sinks” and a current
source fits that language better than a battery.
7.2 1-Node Fin Analysis 245

The average temperature of the fin is represented by Tfin. In this model, the
distance used in the conductive resistance is the half length (the distance between
the two nodes), while the surface area of the convection is based on the full length
(times the perimeter P). To be complete, the corrected length could be used in the
convective resistance (but conceptually, not in the conduction), but is excluded here
to keep the final result cleaner. A thermal capacitance is shown, but at steady-state,
the net heat flow rate into the capacitor is zero. A good review exercise is to analyze
a transient fin using a 1-node lumped transient. For a fin undergoing a transient
heating or cooling (an important practical case when there is a cycling on and off of
some device, for example), a closed-form transient solution using the previous
1-node methods works fine. The objective here is to introduce steady-state analysis
techniques, however.
At steady-state, the rate of heat transfer is determined in one step, namely:

T0  T1
qss ¼
Rk þ R h
If a given application is posed with a specified base temperature (the conventional
approach), then this form can be used to determine the heat transfer rate. If a given
problem is posed with a specified heat transfer rate (for example, to “dissipate” heat
from a computer chip), then this equation can be rearranged to solve for the required
operating temperature of the base:

T 0 ¼ T 1 þ ðRk þ Rh Þqss

A perfect fin will have no conductive resistance, and therefore the fin efficiency can
be expressed as:

T 0 T 1
qss R þR Rh 1
η¼ ¼ T 0kT 1h ¼ ¼R
qperfect Rh
Rk þ R h k = Rh þ 1

The ratio of the conductive and convective resistances is shown to be the key to the
thermal parameter of a fin. The average fin temperature can be expressed using a
voltage divider analysis:

T Fin  T 1 Rh
¼ ¼η
T0  T1 Rh þ Rk
For fins with high efficiency, the fin temperature is close to the wall temperature.
For fins with low efficiency, the fin temperature is close to ambient.
The thermal resistances can be estimated for a wide variety of complicated
shapes, generally by placing the node at the centroid of the fin. For the case of
the constant-area fin defined by a base area A, and base perimeter P:

T 0  T 1 hPLðT 0  T 1 Þ hPLðT 0  T 1 Þ hPLðT 0  T 1 Þ

qss ¼ ¼ ¼
2 ¼
2kA þ hPL
L 1 hPL2
2kA þ 1
L
 þ1 L
L^ 2 þ 1
246 7 Heat Transfer Fins (and Handles)

where a thermal length L* has been defined as:

rffiffiffiffiffiffiffiffi
 2kA
L ¼
hP
and a dimensionless fin length as the ratio of the actual fin length to the thermal
length:
L
L^ ¼ 
L
The fin temperature is expressed as:
 
L^ 2
T fin ¼ T 0  ðT 0  T 1 Þ
L^ 2 þ 1
Notice the limits:

For L^  1, T fin ¼ T 0

that is, if the actual length of the fin (L) is much less than the thermal length of the
fin (L*), then the fin temperature equals that of the base. Such a fin can be called a
“thermally short” fin, and will behave well as a fin, but lousy as a handle (you’ll
burn your hand if you grab the end). Conversely:

For L^  1, T fin ¼ T 1

That is, if the actual length is much longer than the thermal length, the fin will be at
the ambient fluid temperature. Such a “thermally long” fin will be a lousy fin, but an
excellent handle. An alternative 1-node model will be developed for thermally long
fins shortly.

For L^ ¼ 1, T fin ¼ ðT 0 þ T 1 Þ=2

That is, the fin temperature is the average of the base and ambient temperature. In a
revised model tailored to thermally long fins, this result is used to determine the
distance over which the temperature actually changes.
The fin efficiency is:


qss hPLðT 0  T 1 Þ= L^ 2 þ 1 1
η¼ ¼ ¼ 2
qperfect hPLðT 0  T 1 Þ ^
L þ1

The fin efficiency equals 1 for thermally short fins and decreases monotonically
with fin length.
7.2 1-Node Fin Analysis 247

The fin effectiveness is:


qss hPLðT 0  T 1 Þ= L^ 2 þ 1 PL=A
ε¼ ¼ ¼ 2
qNo Fin hAðT 0  T 1 Þ ^
L þ1
For thermally short fins, the fin effectiveness increases with length. For thermally
long fins, the fin effectiveness decreases with length. Conceptually, once the fin is
thermally long, added extra length does not change the heat transfer rate, and
therefore this model does not capture that effect. The alternative long fin model,
to which attention now turns, will address that deficiency.

7.2.2 Thermally Long Fin Model for Handles

The fin model just developed can be applied to a fin of any length. However, if the
fin is long compared to its thermal length (L > L*), then the area near the end is not
thermally involved, and defining the convective resistance based on the total
surface area is problematic, conceptually. An alternative thermal resistance
model, shown in Fig. 7.5, recognizes that there is a region adjacent to the base
whose average temperature is known to be halfway between the base and ambient
temperature. The size of that region is not known a priori, however, and is to be
determined.
In this model, it is recognized that the actual temperature at the end of the fin is
closer to the fluid temperature than the base temperature. Therefore the fin is broken
into two segments. The segment adjacent to the wall (on the left) is fixed at a
temperature midway between the wall and fluid temperatures. The boundary
between this segment and the rest of the fin is considered to be adiabatic.
That is, heat flows into the fin from the base, and out to the fluid through
the sides of the first segment, but not further out along the fin. The distance from
the wall to the first node is left unknown, and expressed as a fraction, f, of the

T∞
Rh
T0
Tave
qfin Rk
Fig. 7.5 Schematic and fL
thermal resistance network
for a thermally long fin at T0+T∞
(fL/2) 1
steady-state. The factor “f” is qfin T0 T∞
a fraction of the total length, kA 2 hP(fL)
whose value is solved for
248 7 Heat Transfer Fins (and Handles)

total length. The temperature of the fin will be midway between the wall and the
fluid when the convective and conduction resistances are equal. That is:

fL 1
¼
2kA hPfL
Solving for “f”:
rffiffiffiffiffiffiffiffi
1 2kA L 1
f ¼ ¼ ¼
L hP L L^

The heat transfer rate, recognizing the two resistances are equal, is:
pffiffiffiffiffiffiffiffiffiffiffi
T0  T1 T0  T1 hPkA
qss ¼ ¼ pffiffiffiffiffiffi ¼ pffiffiffi ðT 0  T 1 Þ
2 2 hP 2
hPfL pffiffiffiffiffiffiffiffi
hP 2kA
The exact solution for a long fin has exactly the same functional dependency, but
without the factor 21/2 in the denominator. It would be relatively easy to “adjust” the
1-node model to make the agreement exact.
The 1-node model captures the right physics, and the characteristic length L* is
the same as that in the exact solution. Furthermore, if the basic logic is understood,
it is easy to solve for fins with non-constant-area or other complicated geometry.
Notice that the heat transfer rate is independent of the actual fin length. Adding
length to a thermally long fin will merely add material that is already close to
ambient. That’s why a thermally long fin is a good handle. The fin efficiency is:
pffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi rffiffiffiffiffiffiffiffi
hPkA=2ðT 0  T 1 Þ 1 2kA 1
η¼ ¼ ¼
hPLðT 0  T 1 Þ 2L hP 2L^
Both thermally short and thermally long models yield a fin efficiency of 0.5 when
the dimensionless fin length equals unity, so the two models smoothly transition
with length.
The exact solution (Appendix 1) for a constant-area fin, with a corrected
length, is:

pffiffiffi
tanh L^ 2 tanhðmLÞ
ηExact ¼ pffiffiffi ¼
L^ 2 mL

where the factor “m” found in traditional heat transfer literature has been
introduced and is the inverse of L* (with a factor of 21/2).
The efficiency for both the long and short fin model results are plotted versus
dimensionless length and compared to the exact solution in Fig. 7.6. The functional
dependency (shape of the curve) is basically right, although the predicted heat
transfer rates are somewhat less than the exact result (lower by 21/2 to be precise).
7.2 1-Node Fin Analysis 249

1.0

0.9

0.8

0.7
exact
0.6
Fin Efficiency

0.5

0.4

0.3 Long Fin Model

0.2

0.1
Short Fin Model
0.0
0 1 2 3 4 5

L/L*

Fig. 7.6 Fin efficiency as a function of dimensionless fin length for the two 1-node models
compared to the exact solution

Of most importance, the emergence and meaning of “m” is apparent in the simple
1-node model. A sound understanding of this parameter is all that is really needed
for fin design. It is tempting to use this feedback to alter the model to give better
quantitative agreement. However, the definition of Rk and Rh used here seems the
most conceptually correct.
Figure 7.7 shows a spreadsheet programmed to solve for the 1-node rectangular
fin. The numerical results are those of the problem statement given at the start of
this section. An “if-then-else” statement is used for the heat transfer rate and fin
efficiency to select the appropriate model (thermally short or thermally long). The
user can change any of the seven input parameters, and an approximate solution for
any rectangular fin follows.

7.2.3 Thick Fins

The basic flow path of heat in a fin is by conduction from the base of the fin to the
interior of the fin, and lateral convection to the environment. However, a thick fin is
defined as one in which the lateral conduction from the interior to the convective
surface is important. In this case, a series resistance can be added to the convective
fin resistance. That is, for a rectangular fin:
250 7 Heat Transfer Fins (and Handles)

Fig. 7.7 Spreadsheet for

1-node model

   
1 ðt=4Þ 1 1 ht=4 1
Rh ¼ þ ¼ þ1 ¼ ð B þ 1Þ
2 kwL hwL 2Lw k 2Lw

A lateral Biot number (B ¼ ht/4/k) has been defined and shown to give a correction
to the convective resistance. A similar procedure can be conducted for alternative
shapes. Note that the lateral conduction distance of t/4 has been used. Conceptually,
the centerline of the fin is an insulated surface, by symmetry, and the node is placed
at the centroid of one half of the fin. The factor ½ reflects that there are two surfaces.

7.2.4 Orientation of Fins

When a fan or some other means of forcing a fluid flow across a fin, the orientation
of the fin is not generally important. However, when there is no imposed flow
(natural convection), the performance of a fin can be greatly affected by its
orientation. For example, for a heat sink with many rectangular fins attached to a
base plate, there is only one orientation that will work well without a forced
convection, namely, the one with the base plate vertical and the fins oriented
vertically, as shown in Fig. 7.8. For all the other orientations, the flow will not
effectively be generated for the interior fins (Fig. 7.9).
7.3 Half a Mug of Coffee 251

Fig. 7.8 Heat sink in an

effective orientation with
natural convection. Vertically
induced buoyant flows are
generated by all inner fins

Fig. 7.9 Ineffective orientations of heat sinks with natural convection. Vertically induced flows
are blocked for inner fins

7.3 Half a Mug of Coffee

Here’s another coffee story twist. My mom not only likes her coffee as hot as you
can get it, she usually says “just half a cup,” or “just an inch.” This section considers
the effect of coffee volume on the cooling time with a 1-node lumped transient
analysis, but with a fin analysis applied to that portion of the mug wall that is above
the coffee level.
The spreadsheet used for the 1-node lumped capacity model from Chap. 4 gives
a first approximation by simply changing the input coffee volume. However, the
model for the side channel was developed with the tacit understanding that the
surface area exposed on the inside of the mug was small compared to the total
exposed surface area. Therefore, a more detailed resistance model is developed here
for the 1-node mug of coffee in which the exposed rim is treated like a heat
transfer fin.
The main point of this chapter has already been made. Therefore, the next
section, and the detailed appendices, provides detailed numerical and analytical
applications, but those can be skipped if that level of detail is not of interest. The
final results are worth reviewing, however, if all the details are not of interest.
252 7 Heat Transfer Fins (and Handles)

Fig. 7.10 Schematic of a D W

partially filled mug of coffee

L
H H0
V
coffee
mug

Engineering Problem Statement

Coffee (Fig. 7.10) at 1 atm, 95  C is quickly poured into a room temperature (23  C)
ceramic mug with mass of mmug ¼ 0.335 kg, inner diameter D ¼ 0.083 m, inner
height H ¼ 0.098 m, and nominal wall thickness w ¼ 0.0040 m. The mug of coffee
is placed on a table in a still room. Estimate the time it takes for the water to cool to
70  C (above which it is too hot) and to 50  C (defined as “tepid”). Consider all
ranges of coffee amounts indicated by the distance between the top of the mug and
the coffee surface (L).

7.3.1 Pure Lumped Capacity Result

The computer programs written to solve the lumped capacity model can handle this
scenario by merely changing a single input parameter, the volume of coffee. The
thermal resistance between the system boundary (the mug outer surface plus the
free surface of the coffee) and ambient air (Rs1) is defined, using a heat transfer
coefficient (h) applied to the entire surface, as:

1
R¼
hAsurface

The capacitance is the sum of the coffee and mug. Table 7.1 compares the full, half,
and quarter mug cases for the pure lumped model, in which an effective heat
transfer coefficient has ¼ 15 W/m2/K is applied to the entire exposed surface area.
The exposed area increases (and thermal resistance decreases) as the volume of
coffee is reduced from a full cup to a half to a quarter due to the inner surface of the
rim. The capacitance (mug plus coffee) decreases. Therefore, the time constant is
reduced due to both a smaller thermal mass and an enhanced surface heat transfer.
The initial temperature (after the coffee and mug come to thermal equilibrium)
decreases as well due to a higher percentage of capacitance being attributed to the
initially cold mug. In fact, for a quarter mug, the coffee temperature after pouring is
below 70  C as soon as the mug and coffee thermally equilibrate.
The pure lumped model assumes that the entire surface of the mug is at the bulk
temperature of the coffee. Is that a good model? For the full cup, it seems a
reasonable first approximation, since the external resistance was shown to dominate
7.3 Half a Mug of Coffee 253

Table 7.1 Effect of coffee volume on cooling time: pure lumped capacity analysis with
h ¼ 15 W/m2/K applied to entire exposed surface
Quarter
Quantity Symbol Unit Full mug Half a mug mug
Coffee volume V L 0.500 0.250 0.125
Total exposed surface area Asurface cm2 452.2 573.3 633.9

Resistance R C/W 1.47 1.16 1.05
Capacitance C J/ C 2,392 1,342 817
Time constant τ ¼ RC min 58.6 25.9 14.3

Initial temperature Τ0 C 83.3 76.6 66.6
Time to cool to 70  C ttoo hot min 14.6 3.4 NA
Time to cool from 70  C topportunity min 32.5 14.4 6.85
(or initial temperature) to 50  C

the side channel. Furthermore, probing a hot mug of coffee with your fingertips
confirms it to be uniformly hot. This situation would not exist for a Styrofoam mug,
in which case the conductive resistance across the wall could not be considered a
secondary effect. But for half a mug, the very top rim feels a little cooler than the
side. For a quarter cup, the top of the rim feels like it might not even be warm. So
something very different is happening in the mug above the level of the coffee
compared to what is below it. Heat flows directly across the mug wall where it is
exposed to coffee. Above that, heat must be flowing upward into the portion of the
rim exposed to air, and then transferring heat to the air on both sides. The rim is
acting like a fin.
For the purposes of the following series of analyses, the mug is divided into two
parts. The “rim” of the mug is defined to be that portion of the mug above the level
of coffee. The “mug” is what is left below. The heat transfer physics that takes place
in the rim portion is very different than the conduction across the mug portion. Heat
is transferred to the air by the rim, but in order to do so, heat must first conduct into
the rim, parallel to the wall of the rim. It is intuitive that there is a vertical
temperature gradient in the rim.

7.3.2 Modified Lumped Capacity Model

The lumped capacity model developed in Chap. 4 accounts for the rim effect in an
approximate way. In that model, the surface area in Ras (between the mug outer
surface and air) includes the inner surface exposed to air. Also, the conductive
resistance between the inner mug wall and outer wall (a portion of which includes
the inner wall above the level of the coffee) uses the geometric average of these two
areas. Conceptually, this model is reasonable when the rim effects are small, that is,
when the mug is nearly full.
Figure 7.11 shows a modified 1-node RC network designed for partially full mugs.
The top and bottom channels for heat flow are the same as before. The side channel,
254 7 Heat Transfer Fins (and Handles)

Fig. 7.11 Extended 1-node

thermal resistance network of
a half a mug of coffee top h, rim
rim

k, rim

T cm ma
T∞

mug
Ceff

bottom

however, branches into two parallel channels. Heat enters the mug wall along the
submerged surface and is split into a channel directly across the mug wall and a
channel that flows vertically through the mug itself into the rim, and from the rim to
ambient. The rim, while not necessarily designed this way, acts like a heat transfer fin.
The effective capacitance of the system is modified to account for the fact that
the rim is not fully involved. The total capacitance in the model is the sum of the
coffee capacitance, the mug capacitance below
the level

of the coffee,
 and a
fraction, f, of the capacitance of the rim: Ceff ¼ Ccoffee þ Cmug  Crim þ ðf Crim Þ
The value of “f” is determined from a subsequent analysis of the rim branch,
where two different models are developed, depending on the thermal length of
the rim.
The equivalent resistance of the entire system is given by a parallel network with
three branches:

1
R¼ 1
Rtop þ Rside þ Rbottom
1 1

The equivalent resistance of the side channel is given by:

1
Rside ¼ Rcm þ 1
Rma þ Rk, rkm þR
1
h, rim

The resistance between the coffee and the non-rim portion of the mug consists of a
convective resistance between the coffee and the mug inner wall in series with a
conductive resistance between the mug inner surface and center of the wall of the mug:

1 ðw=2Þ
Rcm ¼ þ
hcoffee πDðH 0  LÞ kmug π ðD þ w=2ÞðH 0  LÞ
7.3 Half a Mug of Coffee 255

In this expression, the height of the area is taken to be the outside mug height
(H0) minus the rim height (L), which seems “wrong” but is used to account for edge
effects on the floor. It is instructive to express this resistance and those to follow in
terms of the convective resistance and a factor that has the ratio of conductive to
convective resistance, a form of Biot number:
 
1 hcoffee ðw=2ÞD
Rcm ¼ 1þ
hcoffee πDðH 0  LÞ kmug ðD þ w=2Þ

Defining a Biot number to account for the inner portion, Bcoffee ¼ hcoffee(w/2)/kmug:
 
1 Bcoffee
Rcm ¼ 1þ
hcoffee πDðH 0  LÞ ð1 þ w=2DÞ

The geometric factor in the denominator of the Biot number arises from the
different area associated with the convective resistance and the conductive resis-
tance. Defining a modified Biot number (B0 ) that incorporates the geometric factor:

0 Bcoffee
Bcoffee ¼
ð1 þ w=2DÞ

The resistance between the coffee and the mug is thus:

0
1 þ Bcoffee
Rcm ¼
hcoffee πDðH 0  LÞ

Similarly, the resistance between the non-rim portion of the mug and ambient
consists of a conductive resistance between the mug wall center and the outer
surface in series with a convective/radiative resistance between the outer surface
and ambient:

ðw=2Þ 1
Rma ¼ þ
kmug π ðD þ 3w=2ÞðH 0  LÞ hma π ðD þ 2wÞðH0  LÞ

In terms of the Biot number for the outer portion, Bair ¼ hair(w/2)/kmug:

 
1 Bair ð1 þ 2w=DÞ
Rma ¼ 1þ
hair π ðD þ 2wÞðH 0  LÞ ð1 þ 3 w=2D Þ

Introducing a modified Biot number:

0 Bair ð1 þ 2w=DÞ
Bair ¼
ð1 þ 3w=2DÞ
256 7 Heat Transfer Fins (and Handles)

The resistance between the mug and ambient air is:

0
1 þ Bair
Rma ¼
hair π ðD þ 2wÞðH 0  LÞ

7.3.3 Rim Details

Since the mug and rim capacitances have been lumped into the coffee, the initial
mug heating process cannot be observed in this model. In effect, the mug and the
rim are both considered to be in a quasi-steady state equilibrium. They are changing
with time, but not independently from the coffee temperature. The emphasis in this
model is on estimating the effective thermal resistances associated with the rim,
which involves the following complex web of individual conductive and convective
resistances in series and parallel channels.

7.3.3.1 Rim Convective Resistance, Rh,rim

Heat leaves the rim by convection in parallel with thermal radiation all along its
surface (inner, top edge, and outer) as shown in Fig. 7.12. In series with that, lateral
(horizontal) conduction can be accounted for by introducing resistance from the
centerline of the mug wall to the inner or outer surface. The rim branch convective
resistance is related to system parameters as:

Fig. 7.12 Schematic showing the use of a corrected length to account for heat transfer from the
top surface of the rim by attributing it to the sides
7.3 Half a Mug of Coffee 257

1
Rh, rim ¼ 1
RInnerSurface þ ROuterSurface þ RTopSurface
1 1

There are several subtle differences between the inner and outer surfaces:

• The convective coefficient can be different due to the different type of natural
convection taking place.
• The air temperature inside the mug is warm (and humid), and not at ambient
temperature as implied by the resistance network, in which all the rim surface
effects are combined into a single resistance.
• The radiative coefficient can be different since the inner wall “sees” a hot coffee
surface looking down, the mug inner wall looking across and ambient temperature
looking up.

In this model, all of these effects could be accounted for, however, by suitable
choice of the inner convective/radiative coefficient (and/or a modified resistance
network).

7.3.3.2 Corrected Length

Most of the action is anticipated to take place at the inner and outer wall surfaces.
Rather than merely neglecting heat leaving the top, it can be accounted for by
introducing a corrected length to the rim, sketched in Fig. 7.12. The top surface of
the mug is imagined to be cut down its center and bent upward on both sides. The
new top “surface” is insulated (no heat transfer leaves it), indicated with hatch
marks. The corrected length, Lc, to be used for estimating convective heat loss from
the rim, is merely the sum of the actual length, L, and the added length, ΔL:
Lc ¼ L þ ΔL. For this transient analysis, the added length is of a massless material
in order to preserve the correct thermal capacitance.
The added length, ΔL, adds precisely the area of the top surface to the inner and
outer wall:
   
Area Added Area Added
ATop ¼ þ
to Inner Wall to Outer Wall

Expressed in terms of the geometry of the mug:

h i
π ðD þ 2wÞ2  D2 ¼ ΔLπD þ ΔLπ ðD þ 2wÞ
258 7 Heat Transfer Fins (and Handles)

Solving for the added length:

h i
ðD þ 2wÞ2  D2
ΔL ¼
2ðD þ wÞ

For thin walls (2w  D), the analysis yields a simpler result from which the
controlling physics is clearer:
   
Area Added Area Added
ATop ¼ þ
to Inner Wall to Outer Wall

Expressed in terms of system geometry:

πDw ¼ ðπDΔLÞ þ ðπDΔLÞ

Solving for ΔL:

w
ΔLThinWall ¼
2
It is much easier to see the controlling physics in this thin wall limit than for the
thick wall.
The thin-wall approximation and equal convection coefficients (hma) on the
inside and outside surfaces will be used going forward. Lateral conduction effects
could be included, giving rise to a form of Biot number as a correction factor to the
convective resistance term.
The convective resistance for a thin-walled rim is given by:

1 1
Rh, rim ¼ ¼
hma ARimSurface hma 2πDLc

The “2” arises in the denominator because there are two surfaces, inner and outer.

7.3.3.3 Rim Conductive Resistance

The rim branch conductive resistance is related to system parameters as:

ðAverage Conduction DistanceÞ ðL=2Þ

Rk, rim ¼ ¼ π h i
ðConductivityÞðConduction AreaÞ kmug ðD þ 2wÞ2  D2
4
The numerator is the distance to the centroid of the area (L/2), which represents the
average distance heat must conduct from the base of the rim. Again, the corrected
length should not be used here. The denominator is the thermal conductivity times
the area of the base of the rim.
7.3 Half a Mug of Coffee 259

For thin walls, this expression simplifies to:

ðAverage Conduction DistanceÞ ðL=2Þ

Rk, rim, THINWALL ¼ ¼
ðConductivityÞðConduction AreaÞ kmug πDw

The determination of whether a given rim has thermally short or long behavior can
be made by comparing the magnitude of the conductive thermal resistances
between the base of the rim and the bulk of the rim (Rk,rim) with the convective/
radiative resistance between the surface of the rim and the air (Rh,rim). For thermally
short fins, the conductive resistance is less than the convective, and the average
temperature of the rim is closer to the mug than to ambient. The thermal length is
defined by equating the convective and conductive resistances:

ðL =2Þ 1
¼
kmug πDw hma 2πDL

Solving for the thermal length:

sffiffiffiffiffiffiffiffiffiffiffiffi
kmug w
L ¼
hma

For the test case:

L ¼ 0:017 m ¼ 1:7 cm

If the actual length of the rim is shorter than the thermal length, then the rim is
thermally short, otherwise, it is thermally long. The appropriate thermal resistances
to use are summarized in the following table:

Rim resistance Thermally short rim: L < L* Thermally long rim: L > L*
Rk,rim ðL=2Þ pffiffiffiffiffiffiffiffiffiffiffiffiffi
1 ffi
kmug πDw 2πD kmug whma

Rh,rim 1
hma 2πDLc
pffiffiffiffiffiffiffiffiffiffiffiffiffi
1 ffi
2πD kmug whma

7.3.3.4 Effective Rim Capacitance

Since not all of the capacitance is involved, the total lumped capacitance can be
reduced appropriately to reflect that the rim temperature is between the mug
temperature and ambient. For the capacitance for thermally short cases, the frac-
tion, “f” of the capacitive action of the rim is defined (by instinct) to be the
temperature ratio:

T rim  T 1 1
f short ¼ ¼ 
T mug  T 1 R k , rim
þ1
Rh, rim

For thermally long fins, the capacitance fraction “f” is defined by the ratio of half
the thermal length divided by the actual length:
260 7 Heat Transfer Fins (and Handles)

L
f long ¼
2L
In summary, the method outlined above for the “1-node Fin” is a versatile technique
that can be used to approximate heat transfer rates and temperatures from any
so-called extended surface. The mathematics is algebra based, and often yields
simple closed-form relationships between geometric properties and thermal
properties (k and h). The tricky part is in identifying resistances, and relating
them to system parameters. This general method can be useful as a design tool to
help assess performance of engineered heat transfer devices.

7.3.4 Results of Extended 1-Node Model

The coffee/mug problem is revisited as a 1-node transient solution but with the rim
treated as a fin, using the resistance network of Fig. 7.11. In coding this problem, the
thin-wall approximations are used, as is justified in this case (hw/k ~ 0.05  1). The
thin-wall approximations yield simple closed-form expressions that more clearly
highlight the underlying physics in the model. A main spreadsheet was used to
perform the calculations, and a separate spreadsheet was used to produce results for
the full range of coffee volume. Table 7.2 lists the input parameters used. The
coffee volume was changed to generate a table of results. Table 7.3 lists all derived
quantities, including individual resistances that make up the three thermal channels
(not all of which are shown explicitly in Fig. 7.11), the three channel equivalent
resistances, and the final results for effective capacitance, equivalent resistance, and
the time constant. Note that there is an “if-then-else” statement for the fin resistance
that uses either the short fin or long fin model as needed. The long fin model is used
for this set of input parameter values because the actual rim length (L ¼ 6.06 cm) is
longer than the thermal length of the rim (L* ¼ 1.7 cm)
The behavior of the rim as a fin is demonstrated in Fig. 7.13 which shows the
thermal resistance of the fin (the sum of the conductive and convective resistances in
series) and the fin efficiency as a function of coffee volume. When the mug is nearly
full, there is a high thermal resistance due to the low exposed surface area (the top edge
is always present, even when filled to capacity). As the coffee volume is reduced, the
thermal resistance drops rapidly as more surface area is exposed, and the conduction
resistance into the rim is lower than the convective resistance (the short fin model).
Below approximately 0.45 L volume, the rim resistance remains constant because the
conductive resistance exceeds the convective, and the long fin model applies. The fin
efficiency drops continuously as coffee volume is reduced. The efficiency is used to
calculate the effective capacitance of the rim, considered next.
Figure 7.14 shows the thermal capacitance of the coffee, the mug, and the
calculated initial temperature of the coffee/mug combination after the mug heating
phase (during which time “thermal mixing” of the mug and coffee occur). The
coffee capacitance is proportional to coffee volume, while the effective mug
capacitance depends on the portion of the rim that is involved as a fin. This effective
7.3 Half a Mug of Coffee 261

Table 7.2 Input parameters used for partially filled mug 1-node analysis
Input quantities
D 0.08255 m Mug inner diameter
H 0.098 m Mug inner height
H0 0.106 m Mug outer height
V 2.00E04 m3 Coffee volume
w 0.003975 m Mug thickness
mmug 0.335 kg Mug mass
kmug 1 W/m/K Mug thermal conductivity
rhomug 1,700 kg/m3 Mug density
cmug 800 J/kg/K Coffee density
rhocoffee 1,000 kg/m3 Mug-specific heat
ccoffee 4,200 J/kg/K Coffee-specific heat
hair/outside 13.9 W/m2/K Outside convection coefficient
hair/top 20 W/m2/K Convection coefficient between free surface and
ambient
hcoffee/mug 400 W/m2/K Convection coefficient between coffee and mug
hcoffee/surface 400 W/m2/K Convection coefficient between coffee and free
surface
hfloor 10 W/m2/K Effective coefficient of table

Tcoffee, init 95 C Initial coffee temperature

Tmug,initial 23 C Initial mug temperature

Tambient 23 C Ambient temperature

mug capacity is simply the total capacitance of the rim multiplied by the fin
efficiency, plus the capacitance of the mug below the coffee level.
The initial coffee/mug temperature after the mug heating phase remains high as
the coffee volume is decreased and decreases markedly below approximately 0.2 L
as the effective capacitance of the mug becomes comparable to that of the coffee.
The thermal resistances of the three main channels are shown in Fig. 7.15. The
top and bottom resistances are independent of the coffee volume because the top
and bottom surface areas do not change. On the other hand, the side channel
equivalent resistance decreases sharply with coffee volume, the primary cause of
which is the increase in surface area across the sides. The parallel channel of heat
into the rim reduces the equivalent side resistance somewhat, but since the rim
resistance is always greater than approximately 16  C/W (Fig. 7.13), the rim heat
transfer does not have a dominant effect. Overall, the net equivalent resistance has a
mild dependency on coffee volume.
Finally, the response time of the partially filled mug with the rim treated as a heat
transfer fin, the time needed for the coffee to cool to drinkable temperature (70  C)
and the time to cool to a tepid temperature (50  C) are shown in Fig. 7.16.
262 7 Heat Transfer Fins (and Handles)

Table 7.3 Derived quantities and final results for partially filled mug 1-node model
Derived quantities
H-L 0.0374 m Depth of coffee
L 0.0606 m Rim length
Crim 89.1 J/ C Total capacitance of rim
Crim,effective 6.4 Effective capacitance of rim
Cmug 178.9 J/ C Capacitance of mug (below rim)
Ccoffee 840 J/ C Capacitance of coffee
Atop,bottom 0.0054 m2 Top and bottom surface area
Aoutside 0.0106 m2 Outside surface area below coffee level
Amiddleside 0.0102 m2 Average surface area below coffee level
Ainside 0.0097 m2 Inside surface area below coffee level

Tinitial 82.0 C Initial temperature (after mug heating)
rim conduction area 0.0011 m2 Base area of rim (as fin)
rim surface area 0.0340 m2 Convection area of rim (as fin)
L* 0.0169 m Fin thermal length
fin efficiency 0.0722 Fin efficiency
Individual resistances

Rtop,innerconv 0.47 C/W Between coffee and free surface

Rtop,outerconv 9.34 C/W Between free surface and ambient

Rbottom,innerconv 0.47 C/W Between coffee and bottom surface

Rbottom, wall 0.74 C/W Across bottom wall

Rbottom,floor 18.68 C/W Between bottom and table

Rcoffee, innersurface 0.26 C/W Between coffee and inner side surface

Rsidewall 0.39 C/W Across side wall (below coffee level)

Rside,outerconv 6.77 C/W Between ambient and outer side wall
(below coffee)

Rcm 0.45 C/W Coffee to mid plane of side wall

Rma 6.97 C/W Mid plane of side wall to ambient

Rk,rim SHORT 28.06 C/W Conduction into rim

Rh,rim SHORT 2.11 C/W Convection from rim to ambient

Rh,rim ¼ Rk,rim_LONG 8.20 C/W Long model resistances

Rfin 16.40 C/W Long fin model resistance
Channel resistances

Rtop 9.81 C/W Equivalent top channel resistance

Rbottom 19.89 C/W Equivalent bottom channel resistance

Rside 5.34 C/W Equivalent side channel resistance
Results
Ceff 1,025.3 J/ C Effective total capacitance

Requiv 2.95 C/W Effective resistance
RC 3,021.5 s Time constant
RC 50.4 min Time constant
7.3 Half a Mug of Coffee 263

70 1

Thermal Resistance of Rim (as a fin, oC/W)

0.9
60
0.8

Efficiency of Rim as a Fin

50 0.7
0.6
40
0.5
30
0.4

20 Rim Resistance 0.3

0.2
10
Rim Efficiency 0.1
0 0
0 0.1 0.2 0.3 0.4 0.5 0.6

Coffee Volume (liters)

Fig. 7.13 Thermal resistance of the rim (acting as a fin) and the efficiency of the rim as a function
of coffee volume

3000 100
90
2500
80
Initial Temperature
Total Capacitance (J/oC)

Initial Temperature (oC)

70
2000
60
1500 50
40
1000
30
20
500
10
0 0
0 0.1 0.2 0.3 0.4 0.5 0.6
Coffee Volume (liters)

Fig. 7.14 Thermal capacitance of coffee and mug and initial temperature (after mug heating
phase) as a function of coffee volume. 0.525 L constitutes a mug filled to capacity

The response and therefore cooling times increase with coffee volume. The
increased capacitance has a stronger effect than the decrease in thermal resistance,
since the capacitance increases with the volume, while the thermal resistance
decreases with the surface area.
264 7 Heat Transfer Fins (and Handles)

25

Channel Thermal Resistance (oC/W) 20

Bottom Channel

15

Top Channel
10

5 Side Channel

Equivalent Resistance
0
0 0.1 0.2 0.3 0.4 0.5 0.6
Coffee Volume (liters)

Fig. 7.15 Equivalent resistance of the three main thermal channels. The side channel includes the
rim as a fin in series with the channel across the wall

90

80

70 )
iv
C equ
e (R
60 Tim
ponse
Time (min)

50 Res
°C
50
to
40
C ool
e to
30 Tim

20
°C
o 70
10 to C ool t
Time
0
0 0.1 0.2 0.3 0.4 0.5 0.6
Coffee Volume (liters)

Fig. 7.16 Thermal response time for 1-node partially filled mug, and time to cool to drinking and
tepid temperatures
7.4 3-Node Fin Analysis 265

7.4 3-Node Fin Analysis

Improved accuracy of the numerical method is obtained by discretizing space more

finely. To demonstrate the basic method, the fin is broken into three equal sized
elements, and nodes are placed in their centers (Fig. 7.17, top schematic). This grid
is defined by first defining the nodal volumes and then placing the nodes at their
centroids. Thermal resistances are shown between nodes and ambient as needed
(bottom schematic). Each nodal element has a thermal capacitance associated with
it, but since only the steady-state solution is sought, there is no change in energy
stored, and the capacitances are not shown. A transient analysis using techniques
developed previously would include appropriate capacitances for each node.
An alternative approach to a spatial grid would be to first place three unknown
nodes, and then define the volumes that are associated with them, as in Fig. 7.18.
In this case, a node is placed at the end, and that nodal element has half the volume
of the other two. A good practice problem is to develop this approach and compare
results. Conceptually, the finite volume approach is considered to be better, espe-
cially for transient problems, but the two methods yield comparable results. Care
must always be taken when defining the resistances and capacitances to be consis-
tent with whichever approach is taken.
There are three unknown temperatures, and three nodal equations (steady-state
energy balances) are available. If all thermal resistances are treated as constant

L/3
Size of
L/3 L/3
Nodal
qfin
T0 Elements
T1 T2 T3
Distance
between
L/6 L/3 L/3 Nodes
Fig. 7.17 Grid for 3-node
network using the “define qfin R1∞ R2∞ R3∞ T∞
volumes, then place nodes” T0
1 2 3
method (top) and resistance
network (bottom) R01 R12 R23

Distance
L/3 L/3 L/3
Between
Nodes
0 1 2 3

Size of
Fig. 7.18 3-Node grid using
Nodal
the “place nodes, then define L/6 L/3 L/3 L/6
volumes” method Elements
266 7 Heat Transfer Fins (and Handles)

(as is done here), these three equations constitute a closed set of linear algebraic
equations that can be solved using a variety of methods (i.e., linear algebra).
However, the objective here is to develop a numerical approach to solving
steady-state problems, so this approach is not developed. Rather, the three nodal
equations will be written directly into a form suitable for the Gauss–Seidel iteration
method, a Method of Successive Substitution method, described in Appendix 3.
In general, as derived in Chap. 1, for a node “i” in thermal contact with neighbors
“j”, a nodal balance rearranged for node “i” in terms of its neighbors (some of which
are also unknown) is:

000 X T ðj k, kþ1Þ
Q_ i þ
Rij
ðkþ1Þ j
Ti ¼ X 1

j
Rij

In this general recursion equation, a superscript “k” is used to represent iteration

number, and “k + 1” refers to the next iteration. For unknown neighbors, whether
“k” or “k + 1” is used depends on the order in which the calculations are conducted.
Superscript “k” represents a guess for an unknown temperature, while “k + 1”
represents an improved guess. A general demonstration of a spreadsheet implemen-
tation of Gauss–Seidel is shown in Fig. 7.19. Input (and derived, not shown)
parameters can be organized in a block of cells at the top. The iteration table placed
below that (or in another worksheet) has a heading for iteration number and the
three unknown temperatures. An initial guess to each unknown is made in the first
row (“k” ¼ 0). The three nodal recursion equations are entered into the “k” ¼ 1 row.
The arrows indicate which cells are referred to when the equations are inserted. The
formulas also refer to the parameter block. The equations can be entered in any
order, but the use of “k” or “k + 1” must be consistent with that order or a circular
error will arise. Once these recursion relations are entered, the formulas can be
copied down for as many rows are needed for convergence of the solution.
Convergence is guaranteed if all the thermal resistances are constants. If there are
variables, the method may fail, but often (usually for heat transfer problems) the
method works just fine.

Input
Parameters

"k" T1 T2 T3

0 guess guess guess

Fig. 7.19 Demonstration of
a spreadsheet implementation 1 formula formula formula
of the Gauss–Seidel method T
7.4 3-Node Fin Analysis 267

For this 3-node problem, the nodal equations will be implemented in the order
1, 2, 3. For the problem at hand, there are no heat sources. The base of the fin (fixed
temperature T0) receives a heat source, so a nodal balance there will yield the
required value of the heat source (qfin) once T1 is solved for. The equation for node
1 is:

ðk Þ
T
ðkþ1Þ
T0
R01 þ R212 þ RT11
1

T1 ¼ 1
R01 þ R112 þ R11
1

Notice that the superscripts on the right hand side of the first equation used are all
“k”, representing a “guess” for that temperature. There is only one of them in this
case, but a more complicated network would have more. Nodal equation 2 is:

ðkþ1Þ ðk Þ
T1 T
ðkþ1Þ R12 þ R323 þ RT21
1

T2 ¼ 1
R12 þ R123 þ R21
1

Here, an improved guess (k + 1) for T1 is used in the recursion relation for T2 since it
is available. Nodal equation 3 is:

ðkþ1Þ
T2
ðkþ1Þ R23 þ RT31
1

T3 ¼ 1
R23 þ R31
1

The improved guess for T2 is used here. A spreadsheet implementation is shown in

Fig. 7.20 for the original problem statement. The top block contains direct input
parameters, the same as for the 1-node case, and the user of this program can simply
insert different values for other problems that fall in the rectangular fin geometry.
The derived parameters include primarily the resistance formulas, as follows.
The three conductive resistances are:

L=6 L=3
R01 ¼ and R12 ¼ R23 ¼
kA kA

Notice that the distance from the wall to node 1 is half that between node 1 and 2, or
between 2 and 3. The convective resistances are:

x11 1 x31 1
R11 ¼ R21 ¼ þ and R31 ¼ þ
kA11 hPðL=3Þ kA31 h½PðL=3Þ þ A
The surface area associated with node 3 contains the sides, and the end of the fin.
The first term represents the lateral conduction across the thickness of the fin, and is
neglected here, but it could be easily included, approximated in a similar manner to
the conduction across a cylinder or sphere in the 1-node transient analysis.
268 7 Heat Transfer Fins (and Handles)

Fig. 7.20 Spreadsheet

implementation of 3-node
model showing 10 iterations
and final converged solution
after 100 iterations

The Gauss–Seidel iteration table contains a first guess (set to be equal to the fluid
temperature, but anything could be placed here), and the second row contains the
recursion formulas. Ten iterations are shown, and then the 99th and 100th iterations
are shown as a check for convergence. In this case, convergence to within seven
significant figures is demonstrated. Between the 10th and 99th iteration, there are
7.4 3-Node Fin Analysis 269

100
90
80
70
Temperature (oC)

60
50
40
30
20
10
0
0 0.02 0.04 0.06 0.08 0.1
Distance from Wall (m)

Fig. 7.21 Temperature vs. distance from the wall for the exact solution (solid line) and the 3-node
model ( filled circles)

Table 7.4 Comparison of the numerical models and the exact solution
Model Heat transfer rate (W) Percent error
1-Node 79.2 27.7
3-Node 102.4 6.5
Exact 109.6 –

changes to the fourth significant figure. In a computer program, a while loop test for
convergence is usually a better practice than a “for loop,” but spreadsheets really
only allow the effect of a “for loop.” When a test for convergence is made, it is
important to test all nodes for convergence, not just one. It is common that a given
node may not change for a particular iteration, but some of its neighbors will. At the
next iteration, that given node will change, and a different one may not.
A comparison of the results of the 3-node model with the exact solution (using a
corrected length) is made in Fig. 7.21 which shows the temperature and Table 7.4,
which shows the heat transfer rate which is simply q3‐node ¼ ðT 0  T 1 Þ=R01 . The
temperature predicted by the 3-node model is somewhat below the exact solution,
but the agreement is really remarkable. The error in the heat transfer rate is 6.5 %,
considered quite a good agreement for such a coarse grid. A more general “n-node”
model would be straightforward to implement if improved accuracy were needed.
Be ever mindful, however, that the underlying uncertainty in the inherent
assumptions of the problem statement (especially the convection coefficient)
completely dwarf those made here. The comparison of the numerical model with
the “exact” model does not mean the “exact” model captures the real problem
accurately. For example, the convection coefficient is highly uncertain, and may
270 7 Heat Transfer Fins (and Handles)

vary with position. The fluid temperature might not really be uniformly ambient,
depending on the location of other fins. And so on. . .

Appendix 1. Constant-Area Fins: Differential Solution

Figure 7.2 shows an extended surface (fin) that has a base area A and perimeter
p. The shape of the surface remains constant in the x direction, perpendicular to the
base. The total length of the surface is L, and the material has a thermal conductiv-
ity k. The fin is exposed to a fluid at temperature T1 with an effective heat transfer
coefficient, h. The fin is initially at the fluid temperature, and the base temperature is
suddenly changed and maintained at a temperature T0.

Model Development
A detailed derivation of the governing PDE for this case is developed for a general
transient fin case, including lateral conduction. A cross-section of a differential
slice of thickness dx at an arbitrary distance x placed at the center of the element is
shown in Fig. 7.22. Heat enters the left face (at x – dx/2) by conduction, and splits
into two channels, one leaving the surface into the fluid, and the other by conduction
(at x + dx/2) into the adjacent slice.
A verbally stated energy balance applied to the slice, consistent with the
assumed directions shown by the arrows of Fig. 7.22, is:
     
Rate of Change of Rate of Conduction Rate of Conduction
¼ 
Energy Stored in slice  into left face  out of right face
Rate of Convection

out of sides

Fig. 7.22 Energy flows in a differential element in an extended surface. The right sketch shows
the element in a resistance network, which allows conductive resistance between the center and the
surface exposed to the fluid
Appendix 1. Constant-Area Fins: Differential Solution 271

Expressed mathematically:
 

∂T ∂T ∂T

ρcAdx ¼ kA  kA  hpdx T surface  T 1
∂t ∂x xdx=2 ∂x xþdx=2

In the transient and conduction terms, the temperature, T, represents the average
temperature of the cross-section. In the convection term, the temperature is the
surface temperature.
The conduction at x + Δx (where Δx ¼ – dx/2 or + dx/2) can be related to that at
the center (x) using a Taylor expansion:

∂T ∂T ∂ ∂T
kA ¼ kA þ kA Δx þ   
∂x xþΔx ∂x x ∂x ∂x x

The net conduction into the differential element is therefore:



 
∂T ∂T ∂T ∂ ∂T dx
kA  kA ¼ kA þ kA
∂x xdx=2 ∂x xþdx=2 ∂x x ∂x ∂x 2


  
∂T ∂ ∂T dx
 kA þ kA
∂x x ∂x ∂x 2

∂T ∂T ∂ ∂T
kA  kA ¼ kA dx
∂x xdx=2 ∂x xþdx=2 ∂x ∂x x

The surface temperature can be related to the average temperature using a voltage
divider branch between the center of the fin and its surface:

T  T 1 T surface  T 1
¼
Rk þ Rh Rh
The resistances are in the lateral direction (y and z). Solving for the surface
temperature:
 
Rh
 T  T1
T surface  T 1 ¼ T  T1 ¼
R h þ Rk 1 þ Rk = Rh

The conductive resistance is modeled as:

ðA=pÞ=2
Rk ¼
kðpΔx=2Þ

The numerator, (A/p)/2, is half of the characteristic half thickness of the section,
which is the average conduction distance from the surface to the center. The area in
272 7 Heat Transfer Fins (and Handles)

the denominator is the average between the area at the center (zero) and at the
surface. The convective resistance is straightforward:

1
Rh ¼
hpΔx
The ratio is:

Rk hðA=pÞ
¼ B
Rh k
The governing partial differential equation (energy balance) is therefore:

∂T ∂ ∂T hp

ρcA ¼ kA  T  T1
∂t ∂x ∂x x 1 þ B

The factor “B”, a Biot number, accounts for temperature variation in the lateral
direction, an effect that is usually neglected (and justifiably so) for fins.
For a rectangular fin, with w  t:

A tw t t
¼ ¼

p 2ðt þ wÞ 2ð1 þ t=wÞ 2

Introducing a dimensionless temperature that has a value of zero at the base, and
approaches unity as it approaches ambient:

T  T0
θ¼
T1  T0
Introducing a dimensionless time and distance with normalization constants to be
determined shortly:

t x
^t ¼ x^ ¼
τ Lt

Assuming constant thermal conductivity and area and putting it together:

2
ρcAðT 1  T 0 Þ ∂θ kAðT 1  T 0 Þ ∂ θ hp
¼  ðT 0 þ θ ðT 1  T 0 Þ  T 1 Þ
τ ∂^t L2t ∂^
x2 1 þ B

Simplifying and rearranging:

2
ρcL2t ∂θ ∂ θ hpL2t
¼ þ ð1  θ Þ
kτ ∂^t ∂^
x 2 ð1 þ BÞkA
Appendix 1. Constant-Area Fins: Differential Solution 273

The time constant and characteristic dimension can be defined by setting the
coefficients to unity. For the convection term:
sffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
ð1 þ BÞkA
Lt ¼
hp

For the transient term:

ρcL2t
τ¼
k
The governing partial differential equation becomes:

2
∂θ ∂ θ
¼ θþ1
∂^t ∂^
x2
Since the governing equation is first order with respect to time, a single initial
condition must be specified, for problems that are in fact transient. For example, in
the problem statement, the fin is initially at the fluid temperature (T1), and the base
temperature is suddenly changed and maintained at T0. Therefore, the initial
condition (in dimensions) is:

T ðx; 0Þ ¼ T 1

In dimensionless variables:

θð^x ; 0Þ ¼ 1

The governing equation is second order with respect to x, and therefore two
conditions are required. One condition is that the base temperature is specified:

T ð0; tÞ ¼ T 0

In dimensionless variables:

θð0; ^t Þ ¼ 0

The second condition on x is trickier. The precise statement of it is that the rate of
heat conduction to the end (at x ¼ L) must equal the rate of convection away from
the end:

∂T 

kA  ¼ hA T ðL;tÞ  T 1
∂x x¼L
274 7 Heat Transfer Fins (and Handles)

In terms of dimensionless variables, with a dimensionless fin length L^ L=Lt :


kðT 1  T 0 Þ ∂θ 
¼ hðT 1  T 0 Þθ
e ^
Lt ∂^x x^ ¼L^ L t ;t

  
∂θ  hLt
¼ θðL;
x x^ ¼L^
^ ^t Þ
∂^ k

This type of boundary condition, a convection boundary condition, involves a

gradient of the variable and the variable itself. It dramatically increases the algebra
required to solve the problem and results in a complicated solution.
There is an alternative approach to the second boundary condition which will be
developed here, namely the corrected length method. In it, as sketched in Fig. 7.3,
the heat transfer out the end of the fin is considered by making the fin slightly longer
than the actual fin, with an insulated end.
The length of the fin used is called a “corrected” length, Lc, expressed as the
original length plus the added length:

Lc ¼ L þ ΔL

The area added onto the sides by the added length (the perimeter times the added
length ΔL) is the same as the end of the fin. That is:

pΔL ¼ A

The corrected length is therefore:

A
Lc ¼ L þ
p
The boundary condition on the end of the extended fin is determined by an energy
balance at the end, in which heat conducted to the end from the left is set to the heat
convected away, which is set to zero:

∂T 
kA ¼0
∂x x¼Lc

Introducing dimensionless variables and cancelling factors results in a simpler

boundary condition:

∂θ 
¼0
x x^ ¼Lc =Lt
∂^
Appendix 1. Constant-Area Fins: Differential Solution 275

To restate the result of the model development, the governing equation and initial
and boundary conditions are:
2
∂θ ∂ θ
Governing Equation : ¼ θþ1
∂^t ∂^x2
Initial Condition : θð^x ; 0Þ ¼ 1
Boundary Condition at base : θð0; ^t Þ ¼ 0

∂θ 
Boundary Condition at end : ¼0
∂^x x^ ¼Lc =Lt

The governing equation is a linear partial differential equation with three terms
(transient, conduction, convection) and no parameters (they are “buried” in the
dimensionless variables). The second boundary condition does contain a parameter
(its location, namely the dimensionless thermal length of the fin). The use of
dimensionless variables has reduced the number of parameters in the system from
seven (A, p, L, k, h, T0, T1) to one.
As is often the case in heat transfer applications, it is not so much the tempera-
ture that is sought, but the heat transfer rate. For a fin, the rate at which heat transfer
enters the base of the fin is:

∂T 
qfin ¼ kA 
∂x x¼0

In dimensionless variables:

kAðT 1  T 0 Þ ∂θ 
qfin ¼ 
Lt x x¼0
∂^

Introducing the thermal length:

 rffiffiffiffiffiffiffiffiffiffiffiffi
∂θ  hpkA
qfin ¼ ðT 0  T 1 Þ
x x¼0 1 þ B
∂^

The derivative of dimensionless temperature with respect to dimensionless distance

is all that is required to determine the heat transfer rate.

Model Implementation: Analytical Solution

The general solution can be expressed as the sum of a transient and steady-state
solutions:

θ ¼ θtransient ð^x ; ^t Þ þ θss ðx^ Þ

276 7 Heat Transfer Fins (and Handles)

The transient solution involves all three terms and requires a lot of work to develop
a complete solution that satisfies the initial and boundary conditions. The full
solution is developed here, mainly to demonstrate the basic method of solving
linear partial differential equations. The steady-state solution is much easier to
obtain, and if the problem can be effectively modeled as being at steady-state, it can
be used without bothering with the transient case.
In practice, however, the transient response of a heat fin may be important, for
example, for electronic equipment that is cycled on and off. During the initial start-
up period, the rate of heat transfer into a fin can start high, as heat is going to change
the energy stored, and gradually decay to its steady-state response and dissipate
heat to ambient. A good design might take advantage of that behavior.

Transient Solution

The governing equation for the transient solution is:

2
∂θtransient ∂ θtransient
¼  θtransient
∂^t ∂^x2
Note that the term “1” in the governing equation will be grouped into the steady
state solution. Using the separation of variables technique, the solution is assumed
to be expressible as two separable functions: one depending on position, and the
other on time:

θtransient ¼ Tð^t ÞXðx^ Þ

Insertion into the governing equation:

 2 
dT d X
X ¼T  X
d^t x2
d^

Dividing through by XT:

 
1 dT 1 d2 X
¼ 1 ¼ λ2
T d^t x2
X d^

In this expression, the term on the left is a function of time only, the term in the
middle is a function of position only, and it is recognized that that can only be
possible if they are both equal to at most a constant, called a separation constant,
and given the symbol λ, whose value will be determined by application of the initial
and boundary conditions. The negative sign avoids complex numbers and squaring
makes for a cleaner final solution (not obvious why at this point).
The time-dependent equation is:

dT
þ λ2 T ¼ 0
d^t
Appendix 1. Constant-Area Fins: Differential Solution 277

The general solution is:

T ¼ c0 eλ ^t
2

The space-dependent equation is:

d2 X
þ λ2 X ¼ 0
x2
d^
The general solution is:

X ¼ c1 sin λ^
x þ c2 cos λ^
x

The full general solution, in terms of undetermined coefficients A ¼ c0c1 and

B ¼ c0c2 is:

θtransient ¼ eλ ^t ðA sin λ^

2
x þ B cos λ^
xÞ

The values of A, B, and λ are determined by application of the initial and boundary
conditions. First, applying the boundary condition at the base of the fin, θð0; ^t Þ ¼ 0

0 ¼ eλ ^t ðA sin λ0 þ B cos λ0Þ ¼ Beλ ^t

2 2

The only way this condition can be satisfied is if the coefficient B is set to zero. The
general solution, after application of boundary condition at the wall, has been
reduced to:
2^
θtransient ¼ Aeλ t sin λ^
x

Application of the second boundary condition:


∂θ 
¼ 0 ¼ Aλeλ t cos λL^
2^

∂^ 
x x^ ¼L^ ¼Lc =Lt

Where a dimensionless length (the thermal length of the fin) has been defined as:

Lc
L^ ¼
Lt
The only way this boundary condition can be met (without setting A or λ to zero,
which would make satisfying the subsequent initial condition impossible) is to set
the separation constant to specific values, namely:

cos λL^ ¼ 0
278 7 Heat Transfer Fins (and Handles)

This constraint is true for the following infinite number of discrete values of the
separation constant, namely:

π 3π 5π ð2n þ 1Þπ
λL^ ¼ , , , ..., , ...
2 2 2 2
These “quantized” values of l are called eigenvalues.
The solution can now be expressed as the sum of all possible solutions (prior to
satisfying the initial condition) for these discrete values of λ, with a different
coefficient (A) possible for each term:

π^x 3π x^
θtransient ¼ A0 eð2L^ Þ t sin þ A1 eð2L^ Þ t sin
2 3π 2 ^
π ^
þ 
2L^ 2L^
ð2πþ1Þ 2 ð2n þ 1Þπ^ x
þ An eð 2L^ Þ t sin
^
þ 
2L^
Expressed in terms of an infinite series:

X
1

 2πþ1 ^t ð2n þ 1Þπ x^
θtransient ¼ An e 2L^ 2 sin
n¼0 2L^

Notice that the transient solution decays to zero as time goes to infinity. Last, but
not least, the initial condition must be satisfied:

θð^x ; 0Þ ¼ 1
X1
ð2n þ 1Þπ^
x
1¼ An sin
L^
n¼1

To determine the coefficients An, the final step in this tortured affair, multiply both
sides by a sin term with an integer “m” as a frequency, and integrate over the range
of x:
2 3
ZL^ X
1 ZL^
ð2m þ 1Þπ^
x 6 ð2m þ 1Þπ^
x ð2n þ 1Þπ^
x 7
sin x¼
d^ 4 An sin sin x5
d^
^
2L ^
2L ^
2L
n¼1
x^ ¼0 x^ ¼0

The integral on the right is zero (sketch two sinusoids of different frequencies, and
look at their areas) when m is not equal to n. That is, for every value of n, all but one
of the terms in the infinite series are identically zero. Therefore:

ZL^ ZL^  
ð2n þ 1Þπ^
x x 2
ð2n þ 1Þπ^
sin x¼
d^ An sin d^
x
2L^ 2L^
x^ ¼0 x^ ¼0
Appendix 1. Constant-Area Fins: Differential Solution 279

Performing the integration on the left hand side:

ZL^  ^
ð2n þ 1Þπ^
x 2L^ 
 x L
ð2n þ 1Þπ^
sin x¼
d^  cos
2L^ ð2n þ 1Þπ  2L^ 
0
x^ ¼0
 
2L^ ð2n þ 1Þπ 2L^
¼ cos 1 ¼
ð2n þ 1Þπ 2 ð2n þ 1Þπ

The right hand side, with the help of a trigonometric identity for sin2x ¼
(1  cos x)/2, evaluates to a single number (noting that the second term involves a
cosine integrated over a full cycle, and its integral is thus zero):

ZL^   ZL^  
x 2
ð2n þ 1Þπ^ 1 2ð2n þ 1Þπ x^ An L^
An sin x ¼ An
d^ 1  cos dx ¼
2L^ 2 2L^ 2
x^ ¼0 x^ ¼0

The coefficient is therefore determined:

4
An ¼
ð2n þ 1Þπ

The complete transient solution is thereby attained and expressed as an infinite

series:

X
1
4 ð2nþ1Þπ 2 ð2n þ 1Þπ^
x
eð 2L^ Þ t sin
^
θtransient ¼
ð2n þ 1 Þπ 2 ^
L
n¼0

Expressing this result for the first few terms:

4 1 ð π^ Þ2 ^t πx 1 ð 3π^ Þ2 ^t 3πx 1 ð 5π^ Þ2 ^t 5πx

θtransient ¼ e 2 L sin þ e 2 L sin þ e 2 L sin þ 
π 1 2L^ 3 2L^ 5 2L^
Notice how each term is multiplied by an exponential factor with a negative
exponent that grows in absolute value with the number of the term. Except for
very early time, where t  1, the first term dominates the solution.
To determine the heat transfer rate associated with the transient solution, the
derivative with respect to x is taken and evaluated at x ¼ 0:

∂θtransient  X1 ð2nþ1Þπ 2
2eð 2L^ Þ t ¼ 2eð2L^ Þ t þ 2eð2L^ Þ t þ 2eð2L^ Þ t þ   
^ π 2^ 3π 2 ^ 5π 2 ^

 ¼
∂^ x x^ ¼0 n¼0

The heat transfer rate associated with the transient solution is therefore:
rffiffiffiffiffiffiffiffiffiffiffiffi
hpkA X1 ð2nþ1Þπ 2
2eð 2L^ Þ t
^
qfin, transient ¼ ðT 0  T 1 Þ
1þB n¼0
280 7 Heat Transfer Fins (and Handles)

Steady-State Solution

The steady-state solution is much easier, especially with the corrected length
boundary condition. The governing equation (a linear ordinary differential equation
with constant coefficients) and boundary conditions are:

d2 θss
 θss ¼ 1
d^x2

∂θ 
¼0
x ^
∂^ x^ ¼L

The general solution comprises a homogeneous and particular solution:

θss ¼ θss, H þ θss, P

The homogeneous solution is that with the right hand side set to zero, and the
solution is:

θss, H ¼ c1 e^x þ c2 ex^

The particular solution is:

θss, P ¼ K

Substitution of these expressions into the governing equation yields K ¼ 1, and the
general solution is therefore:

θss ¼ c1 e^x þ c2 ex^ þ 1

Application of the boundary condition at the base (at x ¼ 0):

0 ¼ c1 þ c2 þ 1

Application of the second boundary condition (at x ¼ Lc):

^ ^
0 ¼ c1 eL þ c2 eL

Combining them:

^ ^
0 ¼ ðc2 þ 1ÞeL þ c2 eL

Solving for c2:

^
eL
c2 ¼
eL^ þ eL^
Appendix 1. Constant-Area Fins: Differential Solution 281

And
 
^ ^ ^
eL^ e L  e L þ e L eL
^
c1 ¼ 1¼ ¼
eL^ þ eL^ eL^ þ eL^ eL^ þ eL^
The steady-state solution is therefore:
! !
eðL ^x Þ þ eðL ^x Þ
^ ^ ^ ^
eL e^x þ eL ex^
θss ¼ 1  ¼1
eL^ þ eL^ eL^ þ eL^

This equation can be expressed in a hyperbolic cosine form:

ey þ ey
coshðyÞ ¼
2
ey  ey
sinhðyÞ ¼
2
sinhðyÞ ey  ey
tanhðyÞ ¼ ¼
cosh
ðyÞ ey þ ey
cosh L^  x^
θss ¼ 1 

cosh L^

Returning to dimensions:
 
T ss  T 0 cosh LcLx
¼1  
t

T1  T0 cosh Lc
Lt

The derivative evaluated at x ¼ 0 is:



∂θss 
^
e L  e L
^
sinh L^

 ¼ c1 þ c2 ¼ ^ ¼
 ¼ tanh L^
∂^
x x^ ¼0 e þe
L  ^
L cosh L^

The steady-state heat transfer rate, entering the definition of L^ , is therefore:

 rffiffiffiffiffiffiffiffiffiffiffiffi
Lc hpkA
qfin, ss ¼ tanh ðT 0  T 1 Þ
Lt 1þB

The full solution is therefore mercifully upon us:

" # "
#
X
1
4 ð2nþ1Þπ 2
ð 2L^ Þ ^t ð2n þ 1Þπ^
x cosh L^  x^
θ¼ e sin þ 1

n¼0
ð2n þ 1Þπ 2L^ cosh L^
282 7 Heat Transfer Fins (and Handles)

where the first bracketed term is the transient solution, and the second is the steady-
state solution. The heat transfer rate is:
rffiffiffiffiffiffiffiffiffiffiffiffi "   X #
1
hpkA Lc ð2nþ1Þπ 2
ð 2L^ Þ ^t
qfin ¼ ðT 0  T 1 Þ tanh þ 2e
1þB Lt n¼0

hðA=pÞ
where a transverse Biot number is defined as B ¼ , the corrected length is
k ffi
sffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
ð1 þ BÞkA
Lc ¼ L þ A=p , and the thermal length is Lt ¼ .
hp
Personal comments: I consider this really elegant mathematical solution to the
transient rim problem to be “Good Mathematics” and “Bad Engineering.” Engi-
neering is about addressing a human need. To make this value judgment about
whether the method is good or bad engineering requires me to revisit, and perhaps
modify, the problem statement, an iteration of the modeling process. If the objective
is to merely find the heat transfer rate, then the 1-node solution is really simple
mathematically, and yields the same physics as the PDE solution.
The big problem I have with the PDE solution is not that it requires a sophisti-
cated mathematical excursion. If that’s what it takes, then so be it. My problem is
that the solution ties me to some tight restrictions. The base wall temperature must
be fixed in time. The initial condition must be that the fin is entirely at ambient, etc.
The 1-node model gives all that. An extension to an n-node solution requires some
more work, but that numerical method can be applied to much more general
situations quite easily, once the technique is mastered. For example, suppose a fin
is designed to dissipate heat from a motor that turns on and off intermittently, but
has a clear maximum time it is on. A fin might be designed that takes advantage of
high initial heat rates (where heat goes in part to changing the energy stored, and the
rest transferred to the fluid).
In short, the 1-node model teaches the essential physics and yields key design
parameters. The multi-node numerical method yields more detail and is arguably
more quantitatively accurate than the 1-node model. On the other hand, there is so
much uncertainty in convection coefficient values that accuracy can be misleading.
Heat transfer modeling is indeed an art.

Appendix 2. Variable Area Fins: Finite Difference Numerical

Solution

Figure 7.23 shows a circumferential fin attached to a pipe, an example of an

extended surface (fin) that has a base area Abase and base perimeter pbase and a
cross-sectional area A(x) and perimeter p(x) that vary in the distance perpendicular
to the base (x, radial position). The total length of the fin is L, and the material has a
thermal conductivity k. The fin is exposed to a fluid at temperature T1 with an
Appendix 2. Variable Area Fins: Finite Difference Numerical Solution 283

Fig. 7.23 Circumferential tfin

fin of length L, thickness
t place on a pipe of outer
diameter D. The left sketch is
an end view of the pipe; the L
right view is a cross-section
from a side view of the pipe
Dtube

Ñ
X

effective heat transfer coefficient, h. The fin is initially at the fluid temperature, and
the base temperature is suddenly changed and maintained at a temperature T0.

Model Development
The governing differential equation was derived in the previous appendix by
application of the energy balance to a differential element:

∂T ∂ ∂T hp

ρcA ¼ kA  T  T1
∂t ∂x ∂x x 1 þ B

With an initial condition and two boundary conditions (at the base and end of the
fin). The difference between this case and the constant-area is that the area cannot
be taken out of the conduction derivative term. Assuming constant thermal conduc-
tivity, using the chain rule for the conduction term and rearranging:

∂T ∂ T
2
1 ∂A ∂T hp

¼α 2 þα  T  T1
∂t ∂x A ∂x ∂x ρcAð1 þ BÞ

In a finite difference explicit numerical solution, the time derivative is expressed as

a forward finite difference approximation, and the spatial derivatives as central
differences. The boundary conditions are expressed as forward for x ¼ 0, and
backward for x ¼ L second-order finite differences. The details are not developed
here.
284 7 Heat Transfer Fins (and Handles)

Appendix 3. Iterative Solutions of Nonlinear Algebraic Equations

Nonlinear algebraic equations arise in many problems of engineering interest. They

cannot be solved analytically, in general. Rather, some form of iterative procedure
is required. In an iterative procedure, a guess to the solution is first made. Then a
recursion formula based on the governing equations is used to make another guess.
Then another guess is made, then another and another. . . Once the initial guess is
made, the algorithm takes over for all subsequent guesses (called an iteration). If the
method is successful, successive iterations eventually “converge” to the exact
solution. It never reaches it, but a CONVERGENCE CRITERION can be defined
that means “Close Enough!” The method is said to DIVERGE if the absolute value
grows to infinity. Some systems exhibit CHAOS.
Some general points. . .

• Many different algorithms have been “invented.” Heat transfer problems are
generally solvable using the Method of Successive Substitution. It is easy to
implement and usually works. It can easily be extended to systems of equations.
Newton’s method is another, and will be introduced for a single nonlinear
algebraic equation and outlined for systems of nonlinear equations. It is much
harder to implement with systems (with more than one unknown) and is occa-
sionally necessary in heat transfer problems.
• There is no guarantee that a given method will work, that is, find the desired
solution to the problem.
• If an iterative procedure converges, it does so to a correct solution of the system
of equations. However, that solution may make no physical sense (i.e., a
negative absolute temperature).
• If an implementation diverges, it might converge for another initial guess.

Single Nonlinear Algebraic Equations

Solution of a single nonlinear algebraic equation is the natural place to start. The
methods can be extended to systems of equations (with more than one unknown).
The quadratic equation will be used to demonstrate how to implement the methods
and to compare the response on a system that has an analytical solution. In addition,
if the goal is not to solve the equation, but to study CHAOS, iterative techniques on
the quadratic equation are a great place to start.
Quadratic equation:

ax2 þ bx þ c ¼ 0

The quadratic equation is about the simplest nonlinear algebraic equation there
is. As a nonlinear equation, it has the rare distinction of having an exact solution. It
has two solutions, its two roots x1 and x2:
Appendix 3. Iterative Solutions of Nonlinear Algebraic Equations 285

pffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi pffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
b þ b2  4ac b  b2  4ac
x1 ¼ x2 ¼
2a 2a
Keep in mind that in what follows, the objective is to learn new techniques that can
be implemented to solve more difficult nonlinear equations, not to solve the
quadratic equation. . .

Method of Successive Substitution: x ¼ g(x)

The method of successive substitution is easy to implement, and it works often. It
can be used to solve for many systems of linear or nonlinear equations as well.
STEP 1: Rearrange equation:
The governing nonlinear equation is rearranged into the recursion form:

xkþ1 ¼ gðxk Þ

where the subscript “k” refers to the iteration number. There are an infinite number
of ways this rearrangement can be done. For example, the quadratic equation
written in its conventional form is:

ax2 þ bx þ c ¼ 0

Three equivalent forms of the quadratic equation (with subscripts showing the
iteration number) are:

1 

Form 1 : xkþ1 ¼ c þ ax2k isolating the 2nd term
brffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
1
Form 2 : xkþ1 ¼ ½c þ bxk  ðisolating the 1st termÞ
a
Form 3 : xkþ1 ¼ ax2k þ ðb þ 1Þxk þ c ðadding x to both sidesÞ

STEP 2: Make an initial guess, x0, and ITERATE:

Examples:
A spreadsheet was used to investigate the how the method works. Three cases
are shown in Figs. 7.24, 7.25, 7.26, 7.27, 7.28:
Case 7.1 shows a situation in which both roots are found, one of them by two
forms. A plot of the “trajectory” is shown.
Case 7.2 shows one form converging, one form “crashing” immediately
(because the square root of a negative number was asked for), and one form
diverging.
Case 7.3 shows CHAOS. The iteration trajectory neither converges nor diverges.
Sensitivity to initial guess is shown by plotting two trajectories with nearly identical
initial guesses.
286 7 Heat Transfer Fins (and Handles)

1.6

1.4

1.2

1
xk+1=g(xk)

Form3
0.8 Form2
Form1
0.6

0.4

0.2

0
0 2 4 6 8 10 12 14 16 18
Iteration Number

Fig. 7.24 Iteration trajectory for Case 7.1

10

6
xk+1=g(xk)

4 Form3
Form1

0
0 2 4 6 8 10 12 14 16 18

-2
Iteration Number

Fig. 7.25 Iteration trajectory for Case 7.2

Appendix 3. Iterative Solutions of Nonlinear Algebraic Equations 287

2.5
2.0
1.5
1.0
0.5
X

0.0
0 5 10 15 20
-0.5
-1.0
-1.5
x0 = 1.000000
-2.0
x0 = 1.000001
-2.5
Iteration Number

Fig. 7.26 First 20 iterations for Case 7.3

2.5

2.0

1.5

1.0

0.5

0.0
X

0 10 20 30 40 50
-0.5

-1.0

-1.5

-2.0 x0 = 1.000000
x0 = 1.000001
-2.5
Iteration Number

Fig. 7.27 First 50 iterations for Case 7.3

288 7 Heat Transfer Fins (and Handles)

2.5

2.0

1.5

1.0

0.5

0.0
X

-0.5 0 50 100 150 200

-1.0

-1.5

-2.0 x0 = 1.000000
x0 = 1.000001
-2.5
Iteration Number

Fig. 7.28 First 200 iterations for Case 7.3

Method of Successive Substitution [x = g(x)] Applied to Quadratic Equation

a= -1
b= 1.5

ax2 + bx + c = 0
c= -0.2
Exact Solution
x1= 0.14792
x2= 1.35208

ITERATION TABLE RECURSION RELATIONS:

Iteration Form1 Form2 Form3
0 1 1 1 FORM 1:
1 0.8 1.140175 1.3
2 0.56 1.228928 1.36 − ax2 − c
3 0.3424 1.281948 1.3504 x=
4 0.211492 1.312601 1.35242 b
5 0.163153 1.330001 1.35201
6 0.151079 1.339777 1.352094
7 0.14855 1.345238 1.352077 FORM 2:
8 0.148045 1.348279 1.35208
9 0.147945 1.34997 1.35208
− bx − c
10 0.147925 1.350909 1.35208
x=
11
12
0.147921
0.14792
1.35143 1.35208
1.351719 1.35208
a
13 0.14792 1.35188 1.35208
14 0.14792 1.351969 1.35208
15 0.14792 1.352018 1.35208 FORM 3:
16 0.14792 1.352046 1.35208
x = ax3 + (b + 1) x + c

Case 7.1 Three forms converge to different solutions

Appendix 3. Iterative Solutions of Nonlinear Algebraic Equations 289

Method of Successive Substitution [x = g(x)] Applied to Quadratic Equation

a= 1
b= 1.5

ax2 + bx + c = 0
c= -0.2
Exact Solution
x1= 0.123212
x2= -1.623212

ITERATION TABLE RECURSION RELATIONS:

Iteration Form1 Form2 Form3
0 1 1 1 FORM 1:
1 -0.533333 #NUM! 3.3
2 -0.056296 #NUM! 18.94 − ax2 − c
3 0.13122 #NUM! 405.8736 x=
4 0.121854 #NUM! 165747.9 b
5 0.123434 #NUM! 2.75E+10
6 0.123176 #NUM! 7.55E+20
7 0.123218 #NUM! 5.7E+41 FORM 2:
8 0.123211 #NUM! 3.25E+83
9 0.123213 #NUM! 1.1E+167
− bx − c
10 0.123212 #NUM! #NUM!
x=
11 0.123212 #NUM! #NUM! a
12 0.123212 #NUM! #NUM!
13 0.123212 #NUM! #NUM!
14 0.123212 #NUM! #NUM!
15 0.123212 #NUM! #NUM! FORM 3:
16 0.123212 #NUM! #NUM!
x = ax2 + (b + 1) x + c

Case 7.2 Two forms diverge

290 7 Heat Transfer Fins (and Handles)

a= -1.0 Exact Solution

b= -1.0 x1= -1.9662881
c= 1.9 x2= 0.966288
epsilon= 0.000001
Iteration Form1 Form1
0 1.000000 1.000001
− ax 2 − c
1 0.900000 0.899998 x=
2 1.090000 1.090004 b
3 0.711900 0.711892
4 1.393198 1.393210
5 -0.041002 -0.041033
6 1.898319 1.898316
7 -1.703614 -1.703605
8 -1.002302 -1.002269
9 0.895390 0.895456
10 1.098277 1.098158
11 0.693789 0.694049
12 1.418657 1.418296
13 -0.112589 -0.111563
14 1.887324 1.887554
15 -1.661991 -1.662859
16 -0.862214 -0.865100
17 1.156587 1.151602
18 0.562308 0.573813
19 1.583810 1.570739
20 -0.608455 -0.567221

Case 7.3 CHAOS

Newton’s Method: f(x) ¼ 0

Newton’s method is an iterative method that uses the rate of change of the function
to help make an “intelligent” choice for the next iteration. The recursion relation
can be derived with the help of a graphical analysis. The following sketch shows a
general function f(x) in the vicinity of its solution, where it equals 0.

f(x)

f(x0)

f(x1)
X
X2 X1 X0
Appendix 3. Iterative Solutions of Nonlinear Algebraic Equations 291

The first “guess”

 is xo and the function value is f(xo) at that point. The slope of
df 
the function:  is drawn at that point. The next iteration value chosen, x1, is taken
dx x0
to be where that slope intersects the x-axis.

df  f ðx0 Þ  0
¼
dxx0 x0  x1

Solving this equation for x1:

f ðx 0 Þ
x1 ¼ x0  
df 
dx x0

In general, the recursion relation is:

f ðx k Þ
xkþ1 ¼ xk  
df 
dx xk

To implement the method, an initial “guess” is made, and the recursion relation is
used in an iterative fashion.

Newton’s Method for Systems of Nonlinear Algebraic Equations

In a resistance network, if the thermal resistances are expressed as functions of
temperature (which are unknown), for example, with radiation coefficients, then the
system of algebraic equations that results from discretizing space will be nonlinear. In
most heat transfer applications, the relatively easy to implement Method of Successive
Substitutions (i.e., Gauss–Seidel method applied to 3-node fin) works well. When it
fails, Newton’s method (which is harder to implement, but more likely to succeed) can
be implemented. The system of n equations in n unknown temperatures can be written
in the form (one for each node of a space-discretized network):

f 1 ðT 1, T 2 , . . . , T n Þ ¼ 0
f 2 ðT 1, T 2 , . . . , T n Þ ¼ 0

f n ðT 1, T 2 , . . . , T n Þ ¼ 0

The method involves making an initial guess for all unknown temperatures, and
then making an improved guess based on how the function is changing. The initial
guess is given a superscript “k” and the improved guess the superscript “k + 1”. The
total derivative of each nodal equation consists of the contribution from each
unknown variable:
292 7 Heat Transfer Fins (and Handles)

ðkÞ   ðkÞ   ðkÞ

∂f 1  ðkþ1Þ ðk Þ ∂f 1  ðkþ1Þ ðkÞ ∂f 1 
ðkþ1Þ 
df 1 ¼ ∂T 1
 T 1  T 1 þ ∂T 2
 T 2  T 2 þ    þ ∂T n
 Tn  T ðnkÞ
ðkþ1Þ ðkÞ ðkÞ
¼ f1  f
0  f1
ðkÞ  1  ðkÞ   ðkÞ
∂f 2  ðkþ1Þ ðk Þ ∂f 2  ðkþ1Þ ðkÞ ∂f 2 
ðkþ1Þ 
df 2 ¼ ∂T 1  T1  T 1 þ ∂T 2
 T 2  T 2 þ    þ ∂T n 
Tn  T ðnkÞ
ðkþ1Þ ðkÞ ðkÞ
¼ f2  f2
0  f2
 ðkÞ   ðkÞ   ðkÞ
∂f n  ðkþ1Þ ðk Þ ∂f n  ðkþ1Þ ðkÞ ∂f n 
ðkþ1Þ 
df n ¼ ∂T 1
 T 1  T 1 þ ∂T 2  T 2  T 2 þ    þ ∂T n 
Tn  T ðnkÞ
¼ f ðnkþ1Þ  f ðnkÞ
0  f ðnkÞ

The right hand side represents an approximation to the total change. The zero
reflects that the function evaluated at the next iteration is supposed to be zero, but
it should be closer to zero than the previous iteration. Since all the functions and
their derivatives (possibly by numerical approximation) can be evaluated at the “k”
iteration, this represents a system of linear algebraic equations that can be solved
for the temperatures at the next iteration.

Workshop 7.1: Triangular Fins

Compare the steady-state performance of triangular fins with rectangular fins with
the same base thickness (t) and the same total mass of material. Assume that the
depth of the fins (not shown) are large compared to the thickness. Think about it
before analyzing it.

L
k
t
Rectangular Fin

T0 h
T¥
2L
k
t Triangular Fin

h T¥
T0

Level 1: Develop a 1-node analysis for the triangular fin (place the node at the
centroid), and develop an expression for the ratio of the heat flow from the
rectangular fin to that of the corresponding triangular fin. Is there a key dimension-
less parameter to plot against?
Level 2: Develop a few-node model for both in spreadsheets, and compare the
results with the 1-node model.
Level 3: Develop an “n” node model for both using a program code, and
compare with the 1-node model.
Workshop 7.2: Circumferential Rectangular Fin 293

Workshop 7.2: Circumferential Rectangular Fin

Analyze the steady-state performance of a fin constructed of a rectangular plate

placed on the outer surface of a pipe, a common design. The fin is exposed to an
ambient fluid (T1) with an effective heat transfer coefficient h. The spacing between
fins (Δx) is shown but not used in this workshop in which a single fin is isolated.

tfin

Dtube
Wfin

ΔX

• Conduct a 1-node analysis for a thermally short model.

• Develop a 1-node thermally long model.
• Develop a few-node model (identify and take advantage of symmetry).
Steady-State Conduction
8

After the initial mug heating phase when hot coffee is poured into a ceramic mug,
the coffee and mug gradually cool together. During this time, the rate of heat
transfer from the coffee to the inner wall of the mug is approximately equal to
the rate of heat transfer from the outer wall to ambient. That near balance defines a
quasi-steady state situation for the mug itself, and is why the lumping of its capacity
with the coffee is reasonable. For problems that are steady-state in nature (or quasi-
steady state), a transient model can be developed and then run to equilibrium to
obtain the steady-state solution. However, it is possible to solve directly for the
steady-state solution, which involves somewhat different mathematics to execute.
Methods for doing so were initiated in the previous chapter on fins, and elaborated
on in this chapter. Furthermore, the mathematics involved is similar to that required
to execute a step forward using an implicit transient. The method is first developed
and applied to the mug of coffee, and then to a classic 2D problem. Extension to
3D is straightforward.
In general, heat transfer problems are transient and three dimensional. That is,
the temperature varies with time and position. Many heat transfer applications are
inherently constant in time, or are best modeled as such. The objective of this
chapter is to develop a numerical method for problems that are inherently steady-
state. The method is developed in a simple 2D (x,y) space, and the extension to 3D
is straightforward, conceptually.
In classic heat transfer texts, conduction shape factors (S) are introduced for
conduction in purely 2D geometries of homogeneous materials. The heat transfer
rate per unit depth (q/P, in W/m, where P is the depth, or the dimension into the
page) between two surfaces separated by a homogeneous material of thermal
conductivity k and with temperatures maintained at T1 and T2 can be expressed as:

q=P ¼ kSðT 1  T 2 Þ

The conduction shape factor is a measure of the conductive thermal resistance

between two surfaces with a solid between them and depends only on the geometry

# Springer International Publishing Switzerland 2015 295

G. Sidebotham, Heat Transfer Modeling: An Inductive Approach,
DOI 10.1007/978-3-319-14514-3_8
296 8 Steady-State Conduction

of the 2D object. Note that the conductive thermal resistance can be expressed in
terms of the shape factor, namely,

1
Rk ¼
kS
Tables of conduction shape factors are available in classic heat transfer texts and on
the Internet. A conduction shape factor workshop is included at the end of this chapter.
The treatment of the rim as a fin with a 1-node analysis in the previous chapter
allowed for a good prediction of the overall behavior of the cooling mug of coffee.
The focus of this chapter will be on inherently 2D detail that the 1-node model
cannot reveal. Focusing on the mug wall, there is a discontinuity in the type of fluid
at the free surface of the coffee. There is hot coffee on the inside surface up to the
coffee level, and warm, humid air above that. However, the temperature cannot be
discontinuous in either a solid or fluid.
Outside of the thermal boundary layer, the temperature of the fluid adjacent to
the mug surface is different than the temperature of the mug surface temperature.
Since the convection coefficient between coffee and surface is much higher than
that between air and the surface, the temperature of the mug inner surface exposed
to coffee is close to the coffee temperature, while a big drop in temperature occurs
between the outer surface and air. To make an extreme case, in this chapter one
hypothetical scenario is analyzed in which the convection on both air and coffee
contact surfaces are so large that the surface temperature is fixed by the fluid
temperature. A discontinuity in surface temperature is imposed, even though that
would be impossible to achieve experimentally. In reality, the temperature gradient
can be very steep, but never discontinuous. Mathematically, a discontinuous surface
temperature is frequently prescribed in the literature.

8.1 Graphical Method

The following “warm-up” exercise outlines a procedure for solving for the tempera-
ture field and heat transfer rates in a 2D steady-state conduction situation, where the
temperature is fixed along key surfaces. The application is modeled after the mug of
coffee problem. There are two important ways this model differs from the actual
experiment. First, the boundary conditions are of a fixed temperature type, not
convection/radiation. That is, the mug surfaces are fixed at the fluid temperatures.
Second, these “fluid” temperatures are fixed at values of 0 and 100  C. This method
can be used to estimate the conductive resistance between two surfaces that are NOT
maintained at fixed temperature, to be used in a 1D approximation to a 2D problem.
The general steps of the graphical method are:

• Draw geometry to scale. It MUST be to scale.

• Sketch isotherms.
• Sketch heat flux lines.
8.1 Graphical Method 297

The following warm-up exercise breaks these steps into a detailed procedure.

Warm-up
1. On a piece of paper, draw two vertical lines across the entire page (Fig. 8.1
shows steps 1–6). These lines represent the walls of a solid material, with a
dimension W (¼πD) into the page.
2. Draw a thicker line on the left line from somewhere in the middle to the bottom.
The surface represented by the thick line is the “coffee” and has been
maintained at 100  C for a long time. The other surfaces (both walls) have
been maintained at 0  C.
3. Near the bottom of the page, place a dot where you think the temperature would
be 50  C. Place another dot about a centimeter higher up.
4. Place a dot near the coffee surface where you think it would be 50  C. Place
another dot about a centimeter away.
5. Alternate between the two ends until the entire curve is traced out. If a dot is
placed where you wished you hadn’t, just leave it, and place other dots where
you think they should be.
6. Define the 50  C isotherm by adding more dots in between others.
7. Near the bottom of the page, place a dot where you think the temperature would
be 25 and 75  C (Fig. 8.2 shows steps 7–9). Add two more dots, about a
centimeter away.
8. Near the coffee surface, place dots at 25 and 75  C. Add two more dots, about a
centimeter away.
9. Add more dots to define the 25 and 75  C isotherms.
10. Draw a horizontal line near the bottom of the page from the hot wall to the cold
wall (Fig. 8.3 shows steps 10–12). The line just drawn is a heat flux line. As
additional heat flux lines are drawn, observe the following three rules: each line
starts at the 100  C isotherm and ends at the 0  C isotherm, each “pseudo box”
(called an element) created should be nominally square (Fig. 8.4), and each line
must be perpendicular to any isotherm it crosses.
11. Draw a heat flux line around the coffee surface, keeping each element approxi-
mately square.
12. Return to the bottom and add another heat flux line. Alternate between top and
bottom until the web is complete. It is possible that the final heat flux lane will
not have square elements.

Heat Transfer Analysis of Graphical Sketch

Consider any mesh element created, for example in Fig. 8.5, with thermal
resistances between the isothermal lines. Two of its sides are isotherms, and the
two other lines are heat flux lines. Heat does not cross a heat flux line and are drawn
as insulated surfaces (hatched lines). There is solid material between them, so heat
flows from the hot side to the cold side by conduction. The rate of heat transfer is:

ΔT element
qelement ¼
R
298 8 Steady-State Conduction

Fig. 8.1 Placement of 50  C isotherm. Steps 1–6 of the warm-up exercise. The two vertical lines
represent the surfaces of a solid in between. The temperature of the right face is maintained at 0  C.
The left face is maintained at 0  C on the top, and the bottom (the thicker line) is maintained at
100  C. The dots on the left sketch are best guesses as to where the temperature is 50  C.
Additional dots are added to the sketch on the right in order to define the 50  C isotherm

Fig. 8.2 Placement of 25 and 75  C isotherms. Steps 7–9 of the warm-up exercise
8.1 Graphical Method 299

Fig. 8.3 Placement of heat flux lines. Steps 10–12 of the warm-up exercise

Fig. 8.4 Demonstration of “skewed” and “squarish” mesh elements, which can take some
practice to achieve

Fig. 8.5 “Squarish” element

with a conductive resistance
between adjacent isotherms.
The dimension “P” is the
depth into the page
300 8 Steady-State Conduction

Here ΔTelement is the temperature difference between the two isotherms (25  C in
this example, which is the overall temperature difference divided by the number of
temperature increments). Expressing the conductive resistances in terms of the
dimensions of the cell, where Δx is the average distance between the isotherms,
Δy is the average distance between the heat flow lines, and P is the depth of the wall
(into the page) so that ΔyP is the area of an isothermal surface:

ΔT element
qelement ¼

Δx
kΔyP

If the intent to keep each cell square was successfully followed, then the distance
between isotherms (Δx) is nominally the same as the distance between heat flux
lines (Δy), and they cancel each other. Therefore:

qelement ¼ kPΔT element

At steady-state, the rate at which heat enters an element equals the rate it leaves.
Adjacent elements along the same heat flow path (called a heat flow “lane”)
experience the same rate of heat flux. That is:

qelement ¼ qlane

The temperature drop across an element is simply related to the over-all temper-
ature drop by the number of temperature increments, Nincrements:

N increment ΔT element ¼ ΔT overall

ΔT overall
qlane ¼ kP
N increments
Each lane acts in parallel with the others in transferring heat from the hot surface to
cold. There is nothing to distinguish one lane from another, so each lane contributes
the same rate of heat transfer. That may seem surprising. However, where isotherms
are close together (steep gradients), the heat flux lines will also be close together, so
the same rate of heat transfer occurs in a narrow lane. The total heat transfer rate is
simply the sum of all the lanes created in the drawing, or the number of lanes, Mlane,
which can be a fraction for the last lane drawn and the web closes on itself. The total
heat transfer rate is therefore:


PMlanes ðT H  T C Þ
qtotal ¼ k ðT H  T C Þ  kSðT H  T C Þ ¼
N increments RHC

The shape factor is therefore:


PMlanes
S¼
N increments
8.1 Graphical Method 301

Fig. 8.6 Graphical analysis

of side wall of the mug. The
ratio of the wall thickness to
the length of the air column
on the inside (left surface
above the coffee level) is to
scale (0.3  2.2) with the
problem that will be solved
numerically

The conductive thermal resistance between the two surfaces is:

N increments
RHC ¼
kPMlanes
The values for M and N will depend on the specific drawing. For the drawing of
Fig. 8.3, there are four temperature increments (N ¼ 4), and a little more than six
heat flow lanes (M ¼ 6.2, say).
Figure 8.6 shows another sketch of a similar problem. It is a side view, to scale,
but considered infinite in extent below the coffee level, and this problem will be
solved with a numerical method in the next section. The isotherms and heat flux
lines were constructed according to the procedure in the warm-up exercise. There
are four temperature increments (Nincrements) and 23 heat flow lanes generated
(Mlanes). Therefore, the heat transfer rate is:

PMlanes
qtotal ¼ k ΔT overall
N increments
The shape factor is:

PMlanes Pð23Þ
S¼ ¼ ¼ 5:75P
N increments 4
302 8 Steady-State Conduction

The heat transfer rate for the coffee mug
problem (P¼ πD, k ¼ 1.0 W/m/K, when
PMlanes
TH ¼ 70  C and TC ¼ 23  C) is: qtotal ¼ k ðT H  T C Þ
N increments



W ð0:25 mÞð23Þ
qtotal ¼ 1:0  ð70  23Þ C qtotal ¼ 67:6 W
m C ð 4Þ

The method reveals a region of high activity near the coffee surface, where heat
enters the mug wall from the left (coffee side) and exits to the right (to the air just
above the coffee surface). Above a distance of a couple wall widths from the coffee
surface, the temperature is nearly ambient, and below a couple wall widths, the heat
transfer is 1D (straight isotherms and heat flow lanes). The graphical method quickly
provides an excellent feel for this problem, and it can be amazingly accurate.
The solution from the up and coming finite element solution is 69.84 W when the
mug surface temperatures are artificially set equal to the adjacent fluid temperatures.
This “graphical method” of estimating heat transfer rates in a 2D solid can be a
valuable-modeling tool in helping to understand the temperature field in a compli-
cated shape. It can often be enough to help an Engineer make a design decision. If a
more detailed analysis is considered necessary, the graphical method can help
pinpoint regions of high action (where a fine grid would help) and regions of low
action (where a coarse grid is sufficient).
The method can also be used to turn an inherently 2D problem into a 1D
approximation when two surfaces can be identified. Heat flows from one surface
to the other, without branching that gives rise to more complicated 2D mathematics.
To reiterate, a caveat on the method is that the drawing on which isotherms and
then heat flow lanes are drawn must be to scale. Try a quick analysis on another
drawing with a different spacing between the walls. If the walls are farther apart, the
elements will be bigger, making fewer lanes, increasing the thermal resistance.

8.2 Numerical Method for 2D Steady-State Conduction

The finite element method provides a more detailed analysis than the graphical
method. In it, the physical space is broken into discrete elements, each represented
by a node. Thermal resistances between nodes and boundaries are defined. Each
node has a capacitance associated with it, but can be omitted from drawings with
the understanding that there is no change in energy stored in a steady-state problem.
The elements can be any shape, with care being made to account for all thermal
resistance. Finer grids near regions of high gradients can be considered.
The method will be developed with a specific problem statement as a base case
defined to focus on the wall of the mug of coffee problem, and several scenarios will
be run to explore the response. In the following problem statement, the side wall of
the mug is treated as a rectangular prism (making a thin-wall approximation), with
one side exposed to ambient air, and the other side exposed to hot coffee at a fixed
temperature (the quasi-steady approximation) below the free surface and ambient
8.2 Numerical Method for 2D Steady-State Conduction 303

above it. The top is exposed to ambient, and the bottom is exposed to a floor with an
effective heat transfer coefficient (Ub). The height of the wall (2.2 cm) is much
shorter than the wall of the mug investigated in earlier chapters in order to focus on
the thermal response in the vicinity of the coffee surface. The treatment of the
bottom surface is included for the purpose of introducing a variety of 2D responses.

Engineering Problem Statement

The outer surface and top of a 0.3 cm thick (w), 2.2 cm tall (H), 25 cm deep (P) solid
wall of a material with a thermal conductivity k ¼ 1.0 W/m/K is exposed to a
fluid (air) at a temperature of TC ¼ 23  C with a convection coefficient
hC ¼ 13.9 W/m2/ C. The other surface (inner surface) is exposed to air at TC with
hC from the top to the free surface (a distance L ¼ 0.6 cm below the top), and
to coffee at a temperature TH ¼ 70  C with a convection coefficient of
hH ¼ 420 W/m2/ C from the free surface to the bottom. The bottom surface is
exposed to ambient temperature through an effective heat transfer coefficient Ub.
Setting Ub to a “low value” like 1(10)6 W/m2/ C simulates an adiabatic bottom.
Setting it to a “large value” like 1(10)6 W/m2/ C simulates placement on a highly
conductive floor (like copper). Using a finite element numerical method on a coarse
numerical grid, calculate the steady-state temperature distribution in the wall, and
the rate of heat transfer between the hot (@ TH) and cold (@ TC) surfaces (with a
temperature difference ΔToverall).
Run the model for other hypothetical cases:

• Fixed boundaries: set the two convection coefficients to sufficiently high values
that the surface temperatures are approximately equal to the fluid temperatures.
Compare the graphical method with the numerical one for this case. Define a
“Shape Factor” and overall thermal resistance for this geometry defined by:

TH  TC
qtotal ¼ kSΔT overall ¼
RHC

• Styrofoam: set the thermal conductivity of the mug to that of Styrofoam

(k ¼ 0.033 W/m/K).
• Fan: set the cold side heat transfer coefficient to higher values (say, up to
50 W/m2/ C) to simulate a fan blowing on the mug.
• Conductive floor: set the bottom heat transfer coefficient (Ub) to a value of
333 W/m2/ C to simulate the base of the mug being fixed at ambient temperature
by placement on a highly conductive floor.

8.2.1 Finite Element Numerical Analysis

A finite element approach to modeling this problem with a relatively coarse grid
will be developed, and implemented in an Excel Workbook. One version contains a
MACRO that allows easier implementation of different scenarios.
304 8 Steady-State Conduction

Fig. 8.7 2D grid used to

model the quasi-steady state
temperature distribution in
the wall of the mug

8.2.1.1 Grid
The wall is broken up into the rectangular grid shown in Fig. 8.7. The first step is to
break the region into elements. In this case, three “columns” are chosen, three
“rows” are chosen for above the coffee level and eight “rows” below the coffee
level. These two choices define the size of the elements, Δx ¼ w/3 (¼1 mm for
the base case) and Δy ¼ L/3 (¼2 mm for the base case). The location of the
free surface coincides with the boundary between two elements. The total height
of the wall is 11*Δy (¼22 mm). Nodes represented by open circles are placed in
the center of each element. Filled circles show known fluid temperatures. The
nodal temperatures represent average temperatures in a finite region. Surface
temperatures can be determined from voltage divider calculations once the nodal
temperatures are determined.
8.2 Numerical Method for 2D Steady-State Conduction 305

It would be conceptually straightforward, and require more detail to implement,

to consider the thick wall case, where the surface area of the horizontal channels
varied with x. Either a constant grid spacing (Δx ¼ constant) or a constant elemen-
tal volume scheme could be used.

8.2.1.2 General Nodal Equation

Figure 8.8 shows a random element, i, with neighbors “j” in contact through thermal
resistances Rij. A transient energy balance for this node, including a thermal energy
production term is:

dT i X Tj  Ti
Ci ¼ qgen, i þ
dt j
Rij

Rearranging and recognizing that the temperature Ti is common to each term in the
summation:

dT i X Tj X 1
Ci ¼ qgen, i þ  Ti
dt j
Rij j
Rij

Fig. 8.8 General nodal RC

network for node “i”
306 8 Steady-State Conduction

“Solving” for Ti:

dT i X Tj
Ci þ qgen, i þ
dt j
Rij
Ti ¼ X 1

j
Rij

At steady-state, the transient term is identically zero, and the steady-state tempera-
ture of node i in terms and its neighbors is:

X T j, SS
qgen, i þ
j
Rij
T i, SS ¼ X 1

j
Rij

If the physical space is broken up into a total of N elements, each represented by a

node, there are a total of N equations of this type. The equations are algebraic, so a
system of N algebraic equations in N unknown temperatures is represented.
The system is linear if the thermal energy production and thermal resistances are
constant. If they vary with temperature (i.e., thermal conductivity depends on
temperature), the system is nonlinear.
For interior nodes, that is, those whose four faces are shared with another
element can be written in the general form:

T North T East T South T West

qgen, i þ þ þ þ
RNorth REast RSouth RWest
T i, SS ¼
1 1 1 1
þ þ þ
RNorth REast RSouth RWest
Here, the neighbors are given N, S, E, W designations. The thermal resistances are
all conductive resistances, and since the x and y spacing is constant, there are only
two variations:

Δy
RNorth ¼ RSouth ¼
kΔxP
Δx
REast ¼ RWest ¼
kΔyP

If the thin-wall approximation were relaxed, the East and West conductive
resistances would vary with radial position in the wall.
8.2 Numerical Method for 2D Steady-State Conduction 307

8.2.2 Boundary Resistances

Boundary elements, those that have at least one face exposed to a fluid, have a
similar form, but one (or two, for corners) of the temperatures is the fluid, and the
thermal resistances between the fluid and boundary node consist of a convective
resistance in series with a conductive resistance (over half the element).
Top boundary: For the nodal element on the top row in the middle (top of the
mug), the resistance channel to the north is modified:

TC T East T South T West

qgen, i þ þ þ þ
RTop REast RSouth RWest
T Top ¼
1 1 1 1
þ þ þ
RTop REast RSouth RWest

The thermal resistance for elements exposed to the top surface is a convective
resistance in series with a conductive resistance between the node and the surface:

=2
ðΔyÞ
1
RTop ¼ þ
kΔxP hC ΔxP
Rearranging:


1 hC Δy Btop þ 1
RTop ¼ þ1 ¼
hC ΔxP 2k hC ΔxP

This form reveals a Biot number for the element, Btop ¼ hCΔy/2/k, whose value
yields insight into the temperature of the surface. For the base case, in SI units,

ð13:9Þð0:002Þ
Btop ¼ ¼ 0:014
ð2Þð1Þ

The Biot number for this node being much less than 1 means that the temperature
in the element will be effectively uniform. In other words, the surface temperature
of the nodal element will be closer to the nodal temperature than to the fluid
temperature. Yet another way to say it, it is easier for heat to flow by conduction
inside the element than it is for heat to leave the surface into the fluid.
Right boundary: Elements that are exposed to the fluid on the right (ambient air)
experience a different resistance channel to the east:

T North TC T South T West

qgen, i þ þ þ þ
RNorth RRight RSouth RWest
T Right ¼
1 1 1 1
þ þ þ
RNorth RRight RSouth RWest
308 8 Steady-State Conduction

The resistance between the element and the fluid is a conductive resistance over half
the element in series with a convective resistance:

=2
ðΔxÞ
1
RRight ¼ þ
kΔyP hC ΔyP

Rearranging:


1 hC Δx Bright þ 1
RRight ¼ þ1 ¼
hC ΔyP 2k hC ΔyP

The Biot number here is different from the top one because the distance from the
center of the node to the surface is the horizontal distance, whereas it was the
vertical distance for the top boundary. For the base case:

ð13:9Þð0:001Þ
BRight ¼ ¼ 0:007
ð 2Þ ð 1Þ

This value is half that for the top boundary (because the horizontal grid spacing Δx
is half that of the vertical spacing Δy).
Bottom boundary: Elements that are exposed to the “fluid” on the bottom
(simulated floor) experience a different resistance channel to the south:

T North T East TC T West

qgen, i þ þ þ þ
RNorth REast RBottom RWest
T Bottom ¼
1 1 1 1
þ þ þ
RNorth REast RBottom RWest
The resistance between the element and the fluid is a conductive resistance over half
the element in series with a convective resistance:

=2
ðΔyÞ
1
RBottom ¼ þ
kΔxP Ub ΔxP
Rearranging:


1 UB Δy Bright þ 1
RBottom ¼ þ1 ¼
U B ΔxP 2k U B ΔxP

For the base case, with a simulated adiabatic bottom (low UB):

ð1E  6Þð0:002Þ
BBottom ¼ ¼ 1E  9
ð 2Þ ð 1Þ
8.2 Numerical Method for 2D Steady-State Conduction 309

Left boundary, below coffee level: Elements that are exposed to the fluid on the left
below the coffee line (coffee) experience a different resistance channel to the west:

T North T East T South TH

qgen, i þ þ þ þ
RNorth REast RSouth RLeft, H
T Left, H ¼
1 1 1 1
þ þ þ
RNorth REast RSouth RLeft, H

The resistance between the element and the fluid is a conductive resistance over half
the element in series with a convective resistance:

=2
ðΔxÞ
1
RLeft, H ¼ þ
kΔyP hH ΔyP

Rearranging:


1 hH Δx Bleft, H þ 1
RLeft, H ¼ þ1 ¼
hH ΔyP 2k hH ΔyP

For the base case:

ð420Þð0:001Þ
BLeft, H ¼ ¼ 0:21
ð 2Þ ð 1Þ

Due to the much higher convection coefficient between the coffee and surface, this
value is larger than the air-based Biot numbers. It is still less than 1, however.
Left boundary, above coffee level: Elements that are exposed to the fluid on the
left above the coffee level (humid air inside rim) experience a different resistance
channel to the west:

T North T East T South TC

qgen, i þ þ þ þ
RNorth REast RSouth RLeft, C
T Left, C ¼
1 1 1 1
þ þ þ
RNorth REast RSouth RLeft, C

The resistance between the element and the fluid is a conductive resistance over half
the element in series with a convective resistance:

=2
ðΔxÞ
1
RLeft, C ¼ þ
kΔyP hC ΔyP

Rearranging:


1 hC Δx Bleft, C þ 1
RLeft, C ¼ þ1 ¼
hC ΔyP 2k hC ΔyP
310 8 Steady-State Conduction

For the base case:

ð13:9Þð0:001Þ
BLeft, C ¼ ¼ 0:007
ð 2Þ ð 1Þ

This Biot number is the same as for the right boundary.

Corner Nodes
The four corner nodes experience two modifications from the general NESW form,
but all resistances have been defined already. The energy balance equations are, for
the top left:

TC T East T South TC
qgen, i þ þ þ þ
RTop REast RSouth RLeft, C
T TopLeft ¼
1 1 1 1
þ þ þ
RTop REast RSouth RLeft, C

For the top right:

TC TC T South T West
qgen, i þ þ þ þ
RTop RRight RSouth RWest
T TopRight ¼
1 1 1 1
þ þ þ
RTop RRight RSouth RWest

For the bottom right:

T North TC TC T West
qgen, i þ þ þ þ
RNorth RRight RBottom RWest
T BottomRight ¼
1 1 1 1
þ þ þ
RNorth RRight RBottom RWest

For the bottom left:

T North T East TC TH
qgen, i þ þ þ þ
RNorth REast RBottom RLeft, H
T BottomLeft ¼
1 1 1 1
þ þ þ
RNorth REast RBottom RLeft, H

Surface Temperatures
Once the nodal temperatures are determined, the surface temperatures can be
inferred from them using a voltage divider analysis. In general, for a boundary
element (at temperature Ti) in thermal contact with a fluid (at temperature T1), the
voltage divider equation can be expressed as:

ðT i  T 1 Þ ðT S, i  T 1 Þ
qi1 ¼ ¼
Rk þ Rh Rh
8.2 Numerical Method for 2D Steady-State Conduction 311

In this expression, TS,i is the surface temperature of node i, and the conductive (Rk)
and convective (Rh) thermal resistances are defined appropriately for the element in
question. Rearranging for the surface temperature:


Rh
T S, i ¼ T 1 þ ðT i  T 1 Þ
Rk þ Rh

Expressed as the ratio of thermal resistances:

1
T S, i ¼ T 1 þ ÞðT i  T 1 Þ
=Rh þ 1
Rk

The ratio of the conductive to convective resistance is a Biot number. For example,
for the top middle node:

ðΔy=2Þ
Rk hC Δy
¼ kΔxP ¼ ¼ BTop
Rh 1 2k
hC ΔxP
Corner Surface Temperatures
The corner surface temperatures are a little different, and can be handled by an
energy balance on a resistance network defined at the corner of the corner elements.
Figure 8.9 shows the bottom left corner element. The corner is in thermal contact
with the surface “node” to the north (a distance Δy/2 away, across a conductive area
PΔx/4), to the east with the surface “node” (a distance Δx/2 away, across a

Fig. 8.9 Bottom left corner

subelement used to calculate
the surface temperature at the
corner
312 8 Steady-State Conduction

conductive area PΔy/4), to the south with the cold fluid (across a convective area
PΔx/4), and to the east with the hot fluid (across a convective area PΔy/4).
The bottom left corner surface temperature, paying close attention to how the
resistances are defined, is related to its four neighbors by:

T S, above T S, right TC TH
ðΔy=2Þ
þ ðΔx=2Þ
þ 1
þ 1
kðΔx=4ÞP kðΔy=4ÞP UB ðΔx=4ÞP hH ðΔy=4ÞP
T BottomLeftSurface ¼
1 1 1 1
ðΔy=2Þ
þ ðΔx=2Þ
þ 1
þ 1
kðΔx=4ÞP kðΔy=4ÞP U B ðΔx=4ÞP hH ðΔy=4ÞP

For the top left corner surface:

TC T S, right T S, below TC
1
þ ðΔx=2Þ
þ ðΔy=2Þ
þ 1
hC ðΔx=4ÞP kðΔy=4ÞP kðΔx=4ÞP hC ðΔy=4ÞP
T TopLeftSurface ¼
1 1 1 1
1
þ ðΔx=2Þ
þ ðΔy=2Þ
þ 1
hC ðΔx=4ÞP kðΔy=4ÞP kðΔx=4ÞP hC ðΔy=4ÞP

For the top right corner:

TC TC T S, below T S, left
1
þ 1
þ ðΔy=2Þ
þ ðΔx=2Þ
hC ðΔx=4ÞP hC ðΔy=4ÞP kðΔx=4ÞP kðΔy=4ÞP
T TopRightSurface ¼
1 1 1 1
1
þ 1
þ ðΔy=2Þ
þ ðΔx=2Þ
hC ðΔx=4ÞP hC ðΔy=4ÞP kðΔx=4ÞP kðΔy=4ÞP

For the bottom right corner surface:

T S, above TC TC T S, left
ðΔy=2Þ
þ 1
þ 1
þ ðΔx=2Þ
kðΔx=4ÞP hC ðΔy=4ÞP UB ðΔx=4ÞP kðΔy=4ÞP
T BottomRightSurface ¼
1 1 1 1
ðΔy=2Þ
þ 1
þ 1
þ ðΔx=2Þ
kðΔx=4ÞP hC ðΔy=4ÞP U B ðΔx=4ÞP kðΔy=4ÞP

Once all the temperatures are known, the total heat transfer rate from the coffee to
the side wall can be calculated by adding the contributions at each node exposed to
coffee. A good debugging tool is to also calculate the total heat transfer rate out of
all other faces, which should be equal and opposite at steady-state. If not, there is a
bug somewhere.
All the pieces are now in place, and it is time to put them all together.

Model Implementation
An iterative procedure can be applied to a system of N algebraic equations in N
unknowns by guessing a solution, followed by repeated application of a recursion
8.2 Numerical Method for 2D Steady-State Conduction 313

relation formed from the governing equations. The nodal equations developed in
the previous section define a particular set of recursion relations. This form, where
for node i, Ti is expressed in terms of its neighbors, that is, the Method of
Successive Substitution (MOSS). Depending on the system, this iteration procedure
will result in either convergence to a correct mathematical solution (which might be
an unphysical solution), or divergence (chaos or numerical instability), which
means the particular method used has failed. It is possible that the same method
would work with a different set of initial guesses, or that the recursion relations can
be expressed differently.
A strongly diagonal linear system of equations is one in which all the diagonal
elements of each row are greater than or equal to the sum of the absolute values of
off-diagonal elements. For strongly diagonal linear systems, the MOSS will con-
verge to the correct solution. Heat conduction problems with constant thermal
energy production and resistances are strongly diagonal and therefore the MOSS
will work. However, if the thermal energy production and thermal resistances are
variable (for example, with radiation coefficients), making it a nonlinear system,
the method may work, and is generally worth trying. If not, the recursion relations
for nodes that have a high degree of nonlinearity (radiation coefficients are likely
culprits) can be written in another form. If that fails, then other iterative methods,
particularly Newton’s method, are possible. However, setting up a transient prob-
lem and allowing it to reach equilibrium is a legitimate approach if the MOSS fails
after a little tweaking.
A word of caution is in order. Implementation of the MOSS appears to give a
“transient solution” if guessed temperatures are plotted against iteration number.
It is not a transient solution, however, especially if the response time of nodal
elements varies significantly. If a transient solution is desired, set up and solve it as
a transient problem.
Superscripts “k” and “k + 1” are used to identify iteration numbers. In the Jacobi
form, the general equation for node i with superscripts is:

ðkÞ
X T ðj,kSS
Þ
qgen, i þ ðkÞ
ðkþ1Þ j Rij
T i, SS ¼ X 1
ðk Þ
j Rij

Superscripts have to be assigned to all factors on the right hand side, which
indicates thermal energy production and resistances are evaluated at the local
“guessed” temperature, at iteration “k”. The Gauss–Seidel method could alterna-
tively be used in which each nodal temperature is updated to “k + 1” as soon as
available. The order in which the calculations are performed is important in this
method. Systems converge with fewer iterations, generally, but the computational
savings is not dramatic, so the Jacobi method is implemented here.
314 8 Steady-State Conduction

Fig. 8.10 Screenshot of the “Params” worksheet of the excel workbook used to solve the quasi-
steady state mug side wall problem

This problem was solved using a macro-enabled Excel workbook, which is

described in detail here, followed by a presentation of results from running the
program for different scenarios. Figure 8.10 shows a screenshot of the workbook
with the worksheet entitled “Params” accessed. The other worksheets will be
described from left to right. The input parameters (grouped into geometry, thermal
properties, and fluid temperatures) on the left can all be changed by the user. The
derived parameters incorporate the equations developed for the model (grouped into
interior resistances, boundary Biot numbers and resistances, and numerical grid
sizes). The workbook can be protected so only the shaded cells can be changed,
preventing inadvertent changes to the program.
The next sheet, named “k”, shown in Fig. 8.11, shows the “guessed” temperatures.
The worksheet is shown at the very first iteration (k ¼ 0) in which the initial guess is
made. The x position is along the top row, and the y position is in the first column
(in millimeters). The cells around the perimeter represent the fluid temperatures. The
hot fluid (coffee) is shaded in red; the cold fluid (ambient air) is shaded in blue. Inside
those, the cells shaded in gray are the surface temperatures calculated from the guessed
temperatures and the adjacent fluids according the model. Finally, the 33 nodal
temperatures are shown in the heavy boxed region in bold. At this iteration (k ¼ 0),
all the temperatures are set to 50.00, somewhat arbitrarily.
To the right of the temperatures are two system checks. The “Test for Conver-
gence” references the next sheet, to be discussed shortly, and this cell finds the
maximum value of any change that takes place between iteration k and k + 1.
At this point, the maximum change is 0.0869. Underneath, an if-then-else statement
8.2 Numerical Method for 2D Steady-State Conduction 315

x (mm)= 0 0.5 1.5 2.5 3 Test for Convergence

y(mm) 23 23 23 23 23 23 23 MAXIMUM FRACTIONAL CHANGE
6 23 49.4 49.6 49.6 49.6 49.4 23 0.086934023
5 23 49.8 50 .00 50.00 50 .00 49.8 23 Not Converged
3 23 49.8 50 .00 50.00 50 .00 49.8 23 Author:
Run MACRO shortcut key
1 23 49.8 50 .00 50.00 50 .00 49.8 23 CTRL-i
to execute an iteration
-1 70 53.5 50 .00 50.00 50 .00 49.8 23

-3 70 53.5 50 .00 50.00 50 .00 49.8 23 Overall Heat Flux Test

-5 70 53.5 50 .00 50.00 50 .00 49.8 23 HEAT FLUX IN (Watts)= 28.96289
-7 70 53.5 50 .00 50.00 50 .00 49.8 23 HEAT FLUX OUT (Watts)= -3.01073
-9 70 53.5 50 .00 50.00 50 .00 49.8 23 FRACTIONAL DIFFERENCE= 0.896049
-11 70 53.5 50 .00 50.00 50 .00 49.8 23

-13 70 53.5 50 .00 50.00 50 .00 49.8 23

-15 70 53.5 50 .00 50.00 50 .00 49.8 23

-16 70 53.5 50.0 50.0 50.0 49.8 23
70 23 23 23 23 23 23

Fig. 8.11 Worksheet showing the “k” iteration number. The cells are arranged in a 2D grid like
the resistance network

compares that change with the user-defined convergence criterion. If the change is
greater than the criterion, the cell returns “Not Converged” as shown here. Other-
wise, it returns “Converged.” The comment by the author will be discussed in the
next sheet.
Below that comment is a test for overall heat balance. The total heat flux from
the hot fluid into the wall is calculated by adding up all the individual heat fluxes.
The total heat flux out is the heat flux from the cold fluid into the wall (with the
bottom considered part of the cold fluid). This value is negative, meaning that heat
is actually leaving the wall. The sum of the two is the net heat flux into the wall,
which when properly balanced, should equal zero. This overall balance is more of a
check on the overall implementation of the model. A system can converge to a set
of temperatures, but if there are errors in incorporating the model, an imbalance in
the heat flux may be revealed, without necessarily pinpointing precisely where it is.
The next worksheet, “k + 1”, shown in Fig. 8.12, is where the recursion relations
are enacted. The structure of the temperature table is the same as for worksheet “k”
except that the recursion relations are implemented in all 33 nodal cells, which refer
to the temperatures “guessed” in worksheet “k”. At the point shown, after one
iteration, the boundary cells have changed. The table on the right calculates the
difference between k and k + 1 for each node. Notice that the temperatures in the
middle of the wall have not changed (because all the neighbors were at the same
temperature), but the difference between “k” and “k + 1” iterations is 1.42(10)16,
an indication of the precision of the particular computer being used. At the bottom,
the maximum from each column is determined, and the maximum of these is what
is determined in worksheet “k”.
316 8 Steady-State Conduction

x (mm)= 0 0.5 1.5 2.5 3

y(mm) 23 23 23 23 23 23 23
6 23 23 FRACTIONAL CHANGES
5 23 49.561 49.918 49.561 23 8.78E-03 1.64E-03 8.78E-03
3 23 49.7538 50 49.7538 23 4.92E-03 1.42E-16 4.92E-03

1 23 49.7538 50 49.7538 23 4.92E-03 1.42E-16 4.92E-03

-1 70 53.7584 50 49.7538 23 7.52E-02 1.42E-16 4.92E-03
-3 70 53.7584 50 49.7538 23 7.52E-02 1.42E-16 4.92E-03
-5 70 53.7584 50 49.7538 23 7.52E-02 1.42E-16 4.92E-03

-7 70 53.7584 50 49.7538 23 7.52E-02 1.42E-16 4.92E-03

-9 70 53.7584 50 49.7538 23 7.52E-02 1.42E-16 4.92E-03
-11 70 53.7584 50 49.7538 23 7.52E-02 1.42E-16 4.92E-03
-13 70 53.7584 50 49.7538 23 7.52E-02 1.42E-16 4.92E-03
-15 70 54.3467 50 49.7051 23 8.69E-02 1.20E-10 5.90E-03
- 16 70 23
70 23 23 23 23 23 23
MAXIMUM 8.69E-02 1.64E-03 8.78E-03

Fig. 8.12 Worksheet showing the “k + 1” iteration

Execution of another iteration involves copying the 33 nodal temperatures

in worksheet “k + 1” and pasting their values into worksheet “k”. To emphasize,
worksheet “k” has numerical values that represent a guess to the solution,
while worksheet “k + 1” has formulas that look to those values, based on the
recursion relation. When another iteration is to be made, the values at “k + 1” are
used to update. In excel, the cells in “k + 1” are copied, and then “paste special as
values.” A MACRO program was written to do all these steps at once by entering
the shortcut key “CTRL-i”. Pressing and holding CTRL-i while in the worksheet
“k” will cause repeated executions, as fast as the computer will allow, and the state
of the convergence can be monitored. The program does not count how many
iterations are made (although you could modify it). For the problem at hand, it is the
final solution that is sought, not the route taken to get there.
Figure 8.13 shows the worksheet “k” after some several hundred (not counted)
iterations required to reach convergence to the user-defined criterion. Figure 8.14
shows a contour plot of just the 33 nodal elements, while Fig. 8.15 shows a contour
plot that includes the surface temperatures and the fluid temperatures. The problem
with this last plot is related to a limitation of making contour plots with the
version of Excel used which restricts the plot to uniformly spaced x and y series.
If the x and y spacing are not uniform, there is a distortion in the plot. This distortion
is not readily apparent in the base case, but it will be in some of the other cases
considered.
Away from the coffee-free surface, the converged solution confirms what was
learned from the transient analysis, namely, after the mug heating phase, there is a
relatively small drop in temperature from the hot coffee to the mug inner wall, and
then across the inner wall. A large temperature drop occurs between the outer wall
and the ambient air. Also, there is little variation in the vertical direction from the
floor to about 3 mm below the coffee surface. The heat transfer is effectively 1D
8.2 Numerical Method for 2D Steady-State Conduction 317

x (mm)= 0 0.5 1.5 2.5 3 Test for Convergence

y(mm) 23 23 23 23 23 23 23 MAXIMUM FRACTIONAL CHANGE
6 23 52.6 52.8 52.9 52.8 52.6 23 9.97343E-07

5 23 53.0 53.20 53.33 53.18 53.0 23 Converged

3 23 54.9 55.16 55.27 55.10 54.9 23 Author:
Run MACRO shortcut key
1 23 58.2 58.44 58.39 58.07 57.8 23 CTRL-i
to execute an iteration
-1 70 64.9 63.88 62.57 61.72 61.5 23
-3 70 66.7 66.03 64.93 64.10 63.8 23 Overall Heat Flux Test
-5 70 67.6 67.08 66.19 65.45 65.2 23 HEAT FLUX IN (Watts)= 4.181607

-7 70 68.0 67.63 66.87 66.18 65.9 23 HEAT FLUX OUT (Watts)= -4.18125

-9 70 68.3 67.91 67.22 66.57 66.3 23 FRACTIONAL DIFFERENCE= 8.58E-05

-11 70 68.4 68.06 67.41 66.78 66.5 23

-13 70 68.5 68.14 67.50 66.88 66.6 23
-15 70 68.5 68.17 67.54 66.93 66.6 23
-16 70 68.5 68.2 67.5 66.9 66.6 23
70 23 23 23 23 23 23

Fig. 8.13 Worksheet “k” after convergence to the final result

Fig. 8.14 Contour plot of converged solution showing just the 33 nodal elements

across the mug over most of the surface. This 1D behavior was inherently assumed
in the lumped model, and also was revealed by the graphical analysis (where
isotherms were vertical and heat flux lanes horizontal).
On the other hand, the 2D analysis reveals details near the coffee-free surface.
Figure 8.16 shows the boundary heat fluxes along the inner and outer walls (the top
and bottom not shown). Along the inner surface, the heat flux from the coffee to the
mug increases dramatically as the coffee surface is approached from below, and
then suddenly switches to a heat flux from the mug to the air immediately above the
surface. This effect was predicted by the graphical solution which shows heat flux
line spacing decreasing as the coffee surface is approached and the flux line turning
318 8 Steady-State Conduction

Fig. 8.15 Converged solution of the 33 nodal elements plus the surface nodes plus the fluid nodes

6
4
2
Vertical Position (mm)

0
−0.4 −0.2 −2 0 0.2 0.4 0.6 0.8 1 1.2
Inner Surface
−4
Outer Surface
−6
−8
−10
−12
−14
−16
Surface Heat Flux Into Wall (Watts)

Fig. 8.16 Surface heat fluxes into the wall (x axis) as a function of vertical position. The total heat
transfer rate between the submerged surface and that exposed to air (inside and out) is 4.18 W

back on itself. Heat that enters near the coffee-free surface must transfer across the
mug wall (as everywhere below it) but also must transfer vertically into the rim,
resulting in a high heat flux near the surface. Also, the heat loss from the top corners
is higher than from lower nodes due to the heat transferred vertically to the top
surface (in parallel with the horizontal heat flux). The heat flux on the inner and
8.2 Numerical Method for 2D Steady-State Conduction 319

outer surfaces of the rim are essentially equal and closer to the mug temperature
(at the coffee surface) than ambient, indicating that the rim behaves like a thermally
short heat transfer fin.

8.2.3 Case Studies

Each case study from the original problem statement is introduced, followed by its
converged solution, then the contour plot and heat flux plot.

8.2.3.1 Fixed Boundary Temperatures

Figure 8.17 shows the converged results and Fig. 8.18 the contour plot and surface
heat flux when the convection coefficients are artificially set to sufficiently high
values to force the surface temperatures equal to the adjacent fluid temperatures.
This case corresponds to that of the graphical method.
There are steep gradients in temperature across the wall, as expected since the
overall temperature drop across the wall is forced to be that way. Also, the gradients
in the vertical direction are quite steep in the region just above the coffee level. The
rim acts more like a “handle” than a “fin” because three sides are forced to ambient
temperature. This case is considered mainly as a direct comparison to the graphical
method example, and the heat flux of both methods agrees very well.

y(mm) x (mm)= 0 0.5 1.5 2.5 3 Test for Convergence

23 23 23 23 23 23 23 MAXIMUM FRACTIONAL CHANGE
6 23 23.0 23.0 23.0 23.0 23.0 23 4.37626E-07

5 23 23.0 23.06 23.11 23.05 23.0 23 Converged

3 23 23.0 23.49 23.76 23.34 23.0 23 Author:
Run MACRO shortcut key
1 23 23.0 26.76 27.18 24.66 23.0 23 CTRL-i
to execute an iteration
-1 70 70.0 58.39 42.32 29.18 23.0 23
-3 70 70.0 61.66 45.73 30.50 23.0 23 Overall Heat Flux Test
-5 70 70.0 62.08 46.37 30.78 23.0 23 HEAT FLUX IN (Watts)= 69.83995

-7 70 70.0 62.14 46.48 30.83 23.0 23 HEAT FLUX OUT (Watts)= - 69.8399

-9 70 70.0 62.15 46.50 30.84 23.0 23 FRACTIONAL DIFFERENCE= 5.12E-07

-11 70 70.0 62.16 46.50 30.84 23.0 23

-13 70 70.0 62.16 46.50 30.84 23.0 23
-15 70 70.0 62.16 46.50 30.84 23.0 23
-16 70 70.0 62.2 46.5 30.8 23.0 23
70 23 23 23 23 23 23

Fig. 8.17 Converged results for the case of fixed surface temperatures (achieved by setting the
convection coefficients to sufficiently high values)
320 8 Steady-State Conduction

70

Temperature (oC)
60-70
60
50-60
50 40-50
30-40
40 20-30
30
20 Coffee
Inner Surface
0.5
-16

1.5 Horizontal
-15
-13
-11

2.5 Position (mm)

-9
-7
-5

Outer Surface
-3
-1
1

Ambient
3
5
6

6
4 Inner Surfaace
2 Outer Surface
0
Vertical Position (mm)

−10 −5 0 5 10 15
−2
−4
−6
−8
−10
−12
−14
−16
Surface Heat Flux Into Wall (Watts)

Fig. 8.18 Contour and heat flux plot for the case of fixed boundary temperatures. The x-scale was
increased from 1.2 to 15 compared to the base case. The total heat transfer rate between the
submerged surface and ambient air is 69.84 W

8.2.3.2 Styrofoam Mug

Figure 8.19 shows the converged results and Fig. 8.20 the contour plot and surface heat
flux when the thermal conductivity is changed from a value of 1.0 W/m/K (typical of
“ceramic”) to a value of 0.033, typical of Styrofoam. There is a large temperature drop
across the mug wall, and the surface heat flux is much lower than the base case
(ceramic mug). The rim acts like a thermally long fin in this case, with the temperature
at the top much closer to ambient than to the base of the rim (at the coffee surface).
y(mm) x (mm)= 0 0.5 1.5 2.5 3 Test for Convergence
23 23 23 23 23 23 23 MAXIMUM FRACTIONAL CHANGE
6 23 24.0 24.2 24.4 24.2 24.0 23 7.91353E-07
5 23 24.5 . . . 24.4 23 Converged
3 23 26.7 . .
. 26.4 23 Author:
Run MACRO shortcut key
1 23 33.0
.
. 
. 30.9 23 CTRL-i
to execute an iteration
-1 70 68.7 . . . 38.7 23
-3 70 69.2 . . . 42.0 23 Overall Heat Flux Test
-5 70 69.3 . . . 43.0 23 HEAT FLUX IN (Watts)= 1.352793
-7 70 69.3 . . . 43.3 23 HEAT FLUX OUT (Watts)= -1.3528
-9 70 69.3 . .
 .
43.4 23 FRACTIONAL DIFFERENCE= -5.4E-06
-11 70 69.3 . .
. 43.5 23
-13 70 69.3 . .
 . 43.5 23
-15 70 69.3 . .
 . 43.5 23
-16 70 69.3 65.0 56.4 47.8 43.5 23
70 23 23 23 23 23 23

Fig. 8.19 Converged solution for a Styrofoam mug

Temperature (oC) 60-70

70 50-60
60 40-50
50 30-40
20-30
40
30
-15 -13 20 0.5
-11 -9
-7 -5 1.5
Vertica -3 -1 Horizontal
l Positio 1 2.5
n (mm) 3 Position (mm)
5

0
-0.4 -0.2 0 0.2 0.4 0.6 0.8 1 1.2
Vertical Position (mm)

-2
Inner Surface
-4
Outer Surface
-6

-8

-10

-12

-14

-16
Surface Heat Flux Into Wall (Watts)

Fig. 8.20 Contour plot and surface heat flux for a Styrofoam mug. The scale on the surface flux is
that same as the base case, revealing the lower heat transfer rate across the mug wall
322 8 Steady-State Conduction

8.2.3.3 Copper Floor

The base case assumed an adiabatic floor, that is, it prevented heat from flowing
through the bottom surface by using a sufficiently low value of the effective heat
transfer coefficient to the floor (UB). In reality, heat will flow from the mug to the
floor. To simulate the other extreme, namely, a highly conductive floor capable of
providing an effective means of pulling heat from the base of the wall, the floor heat
transfer coefficient (Ub) is set to 333 W/m2/ C (which considers the conduction
across a mug thickness). With this value of UB, the converged solution is shown in
Fig. 8.21 and the contour plot and heat flux plots in Fig. 8.22. It is apparent that the
floor effects increase the heat flux (by more than a factor of 2). However, the effects
are confined to a few millimeters, so for the much taller mug wall than analyzed
here, the majority of the wall will experience 1D heat transfer effects (except near
the rim base and the floor).

y(mm) x (mm)= 0 0.5 1.5 2.5 3 Test for Convergence

23 23 23 23 23 23 23 MAXIMUM FRACTIONAL CHANGE
6 23 52.4 52.6 52.7 52.6 52.4 23 9.88294E-07
5 23 52.8 . 
. . 52.8 23 Converged
3 23 54.7 . . . 54.7 23 Author:
Run MACRO shortcut key
1 23 58.0  .  . .
57.6 23 CTRL-i
to execute an iteration
-1 70 64.8 
. . . 61.2 23
-3 70 66.4 . . 
. 63.4 23 Overall Heat Flux Test
-5 70 67.1 . . . 64.4 23 HEAT FLUX IN (Watts)= 9.471576
-7 70 67.1 .
. . 64.4 23 HEAT FLUX OUT (Watts)= -9.47123
-9 70 66.6 . . 
.  63.5 23 FRACTIONAL DIFFERENCE= 3.67E-05
-11 70 65.3 . . . 61.3 23
-13 70 62.5 .  . . 57.0 23
-15 70 57.0 . . .
 49.1 23
-16 70 50.1 46.4 44.0 42.7 42.6 23
70 23 23 23 23 23 23

Fig. 8.21 Converged solution for the case of a highly conductive (i.e., copper) floor
8.2 Numerical Method for 2D Steady-State Conduction 323

60-70
70 50-60

Temperature (oC)
60 40-50
50 30-40
20-30
40
30
-15 -13 20 0.5
-11 -9
-7 1.5
-5 -3 Horizontal
-1 1 2.5
Vertica 3 Position (mm)
l Positio 5
n (mm)

0
-2 -1 0 1 2
Vertical Position (mm)

-2

-4
Inner Surface
-6
Outer Surface
-8

-10

-12

-14

-16
Surface Heat Flux Into Wall (Watts)

Fig. 8.22 Contour plot and heat flux plots when the floor is highly conductive (i.e., copper floor).
The x-scale was increased from 1.2 to 2 in order to observe the floor effects

8.2.3.4 Fan
The effectiveness of the heat transfer from the outer surface to the ambient can be
enhanced by blowing across it (Fig. 8.23). This effect can be simulated by increas-
ing the value of the convection coefficient on the outside. Changing it to a value of
50 W/m2/ C (a factor of 3.5 greater than the base case) results in the contour and
heat flux plots of Fig. 8.24. The heat flux increases by a factor of 2.4, and there is a
greater drop in temperature across the mug wall. Recall that the outer convective
resistance dominates the total resistance from coffee to ambient in the lumped
model, and therefore lowering that resistance has a large effect. Also, the rim as a
fin is thermally longer than the base case, as the temperature at the top of the rim is
closer to ambient than the rim base.
y(mm) x (mm)= 0 . . . 3 Test for Convergence
23 23 23 23 23 23 23 MAXIMUM FRACTIONAL CHANGE
6 23 36.1 36.5 36.7 36.4 36.1 23 9.53685E-07
5 23 36.8
.
.

. 36.8 23 Converged
3 23 40.0 .
. . 39.8 23 Author:
Run MACRO shortcut key
1 23 45.6 . . .

44.8 23 CTRL-i
to execute an iteration
-1 70 59.1 .  
. . 51.2 23
-3 70 62.5 .   . . 55.4 23 Overall Heat Flux Test
-5 70 64.0 . .  . 57.6 23 HEAT FLUX IN (Watts)= 10.52989
-7 70 64.8 . . . 58.8 23 HEAT FLUX OUT (Watts)= -10.5303
-9 70 65.2 . . .
 59.4 23 FRACTIONAL DIFFERENCE= -3.6E-05
-11 70 65.4 . . . 59.7 23
-13 70 65.5 . .
. 59.8 23
-15 70 65.5 . . .
59.9 23
-16 70 65.5 64.6 62.7 60.8 59.9 23
70 23 23 23 23 23 23

Fig. 8.23 Converged solution for the case of a fan blowing across the side wall

60-70
70
Temperature (oC)

50-60
60 40-50
50 30-40
20-30
40
30
-15 -13 20 0.5
-11 -9
-7 -5 1.5
Vertica -3 Horizontal
l Positio -1 1 2.5
n (mm) 3 Position (mm)
5

0
-2 -1 -2 0 1 2 3
Vertical Position (mm)

-4

-6

-8 Inner Surface
Outer Surface
-10

-12

-14

-16
Surface Heat Flux Into Wall (Watts)

Fig. 8.24 Contour and surface heat flux plots for the case of an enhanced convection on the outer
surface of the mug (i.e., a fan blowing on it). Again, the x-scale is changed compared to the
base case
8.3 Classic 2D Problem Statement 325

It is intriguing that the natural response of an impatient coffee drinker is to blow

on the surface of coffee to cool it down, and that evolution has not progressed to the
point that she blows on the side.

8.3 Classic 2D Problem Statement

The bottom surface (called south) of a long w  H m2 rectangular structural

member (Fig. 8.25, with length L into the page, width w ¼ 4.2 m, height H ¼ 1.3 m)
is maintained at temperature TH (¼100  C). The other three surfaces (north, east,
and west) are maintained at a temperature TC ¼ 0  C. The member has a thermal
conductivity k ¼ 20 W/m/K (typical of steel). A uniform internal generation of
thermal energy is possible (i.e., conversion of electrical energy into thermal
energy). For the case of no internal heat generation, determine the steady-state
temperature distribution within the member and heat transfer rate (per unit length)
from the bottom surface to the three other sides (qHC/L) and the conduction shape
factor. Determine the internal heat generation rate (in W/m3) that will yield a
maximum temperature of 200  C.

8.3.1 Numerical Analysis

As with transient problems, the region of interest can be divided into discrete
elements. Each element is represented by a node. Rather than think of each
element as having a uniform temperature, the nodal temperature represents an
average temperature of the element. That way, boundary temperatures between
nodes can be approximated through voltage divider relations, giving more infor-
mation. The accuracy of the model improves as the region is divided up into
smaller and smaller elemental sizes, at the cost of initial setup work, and
computational effort.

Fig. 8.25 Cross-section (left) and oblique projection (right) of 2D rectangular member
326 8 Steady-State Conduction

Fig. 8.26 Thermal resistance network for a 1  1 model of the classic 2D problem. The surface
temperatures specified are for the classic 2D problem

8.3.2 1  1 Model

A lumped capacity thermal resistance network is shown in Fig. 8.26 for a more general
case, where the temperatures of all four faces can be set to different values and there is
internal heat generation generated. The thermal capacitance (C ¼ ρc*(wHL)) of the
system is shown, along with the internal heat generation (qgen being the volumetric
heat addition, that is, heat added per unit volume). The nodal temperature is given a
subscript 1,1, in anticipation of generalizing to a finer grid, and represents the average
temperature in the member. The node is in thermal contact with the four sides through
conductive resistances (north, east, south, and west). The three sides can be at different
temperatures, but for the classic problem are set to be the same temperature. It is
relatively easy to extend the problem further, by specifying convective boundary
conditions, a good workshop problem.
An energy balance applied to the single node is:

dT 1, 1 T N  T 1, 1 T E  T 1, 1 T S  T 1, 1 T W  T 1, 1
C ¼ qgen wHL þ þ þ þ ¼ 0 @ SS
dt RN RE RS RW

Recognizing that the four conductive resistance terms have the nodal temperature in
common, this expression can be written (now at steady-state) as:
8.3 Classic 2D Problem Statement 327



TN TE TS TW 1 1 1 1
0 ¼ qgen wHL þ þ þ þ  T 1, 1 þ þ þ
RN RE RS RW RN RE R S RW

Solving for the nodal temperature:


TN TE TS TW
qgen wHL þ þ þ þ
RN RE RS RW
T 1, 1 ¼

1 1 1 1
þ þ þ
RN R E RS R W

Alternatively, expressing the heat generation rate as a function of all temperatures:



1 1 1 1 TN TE TS TW
T 1, 1 þ þ þ  þ þ þ
R N R E RS R W RN RE R S RW
qgen ¼
wHL
The four thermal resistances are conductive, and expressed as the distance between
the nodes divided by the thermal conductivity and area. The north and south
resistances are the same (and renamed “v” for vertical). Similarly for the east and
west (named “h” for horizontal):

ðH=2Þ
RN ¼ RS ¼ Rvert ¼
kwL
ðw=2Þ
RE ¼ RW ¼ Rhoriz ¼
kHL


TN TE TS TW
qgen wHL þ þ þ þ
R Rh Rv Rh
T 1, 1 ¼
v 
2 2
þ
Rv Rh


H
qgen wHL þ ðT N þ T S þ H=w ðT E þ T W ÞÞ
2kwL
T 1, 1 ¼
2ð1 þ H=w Þ

qgen H 2 T N þ T S þ ðH=wÞðT E þ T W Þ
T 1, 1 ¼ þ
4kð1 þ H=wÞ 2ð1 þ H=wÞ

The rate of heat transfer from the bottom (south) surface is:

T s  T 1, 1
q¼
Rs
This channel splits into three parallel channels to the other three surfaces. In terms
of a conduction shape factor (which does not apply if there is heat generation):
328 8 Steady-State Conduction

Example 1 Square member (H ¼ w), qgen ¼ 0

qgen H 2 1
T 1, 1 ¼ þ ðT N þ T S þ T E þ T W Þ
8k 4
For the classic problem stated (no heat generation, three sides fixed at equal
temperatures), the temperature of the node is the geometric average of its four
neighbors:

1
T 1, 1 ¼ 0 þ ð0 þ 100 þ 0 þ 0Þ ¼ 25  C
4
The heat transfer rate is:

T s  T 1, 1 100  25
q¼ ¼ ¼ 150kL ¼ 1:5kLΔT overall
Rs H=2
kwL
The last expression defines the shape factor to be 1.5, by inserting the overall
temperature drop between surface 1 (the floor at 100  C) and surface 2 (the other
three sides at 0  C).

Example 2 For the problem statement (H ¼ 1.3 m, W ¼ 4.2 m, qgen ¼ 0)

ð0 þ 100 þ ð1 þ 1:3=4:2Þð0 þ 0ÞÞ

T 1, 1 ¼ 0 þ ¼ 38:2  C
2ð1 þ 1:3=4:2Þ

The heat transfer rate is:

T s  T 1, 1 2wkL ðT s  T 1, 1 Þ 2ð4:2 mÞkL ð100  38:2Þ

q¼ ¼ ΔT overall ¼ ΔT overall
Rs H ΔT overall 1:3 ð100  0Þ
¼ 3:99kLΔT overall

The shape factor for this 4.2 by 1.3 shape is 3.99.

Example 3 For the problem statement (H ¼ 1.3 m, W ¼ 4.2 m, and Tmax ¼ 200  C)


4kð1 þ H=wÞ T N þ T S þ H=wðT E þ T W Þ
qgen ¼ T 1, 1 
H2 2ð1 þ H=wÞ

qgen ¼ 10, 031 W=m3

A good check on the execution is to see whether the net heat flux out of all four
sides equals the total heat generation rate.
8.3 Classic 2D Problem Statement 329

Improved 1-Node Model

The problem statement for this classic case is symmetric about the center vertical
plane, and an improved 1-node model can be made by considering that plane to be
an insulated surface, and placing the node at the center of one half of the rectangle.


qgen wHL TN TE TS
þ þ þ
2 Ra Rb Ra
T0 ¼

2 1
þ
R a Rb

Note that these resistances differ from the previous model (distances and area) and
are given different subscripts (a and b) to distinguish. A good practice exercise is to
compare the results of this model with the previous, and with the more detailed
models that follow. It is emphasized that the 1-node model yields closed-form
solutions, while the detailed models require iterative calculations.

8.3.3 Few-Node Model

It is hopefully obvious that breaking the region into a finer grid will yield more
detail. A “few-node” model is defined here as one which is readily solved in a
spreadsheet, where each node can be listed as its own column. A fun classroom
exercise (Workshop 8.3) that clearly demonstrates the nature of the iterative
calculation is to assemble the class into a grid pattern, and treat each person as a
node. Figure 8.27 shows the right half broken into four equal sized nodes, with the
centerline as an insulated boundary, and each node labeled as both a person, and in a
Cartesian (xy) indexing scheme (discussed further in the next section). Horizontal
resistances are not shown, as heat does not flow across the centerline. Also, the heat
generation sources are shown, but each node receives ¼ of the total heat generated
in the right half (which is half of the total heat generated). It is a common error to
330 8 Steady-State Conduction

Fig. 8.27 A 4-node model.

Using symmetry, the right
side of the member is
considered, broken into four
equal sized elements. The left
surface is an insulated surface

forget that symmetry has been used. For example, the total heat transfer from the
south to the three other sides equals twice the heat leaving the south side of the right
half, an easy calculation, but one which might slip by with being done.
A spreadsheet implementation of this 4-node model is shown in Fig. 8.28. In the
derived parameters block, the horizontal and vertical resistances are those between
nodes. The resistances between the nodes and their associated boundaries are half
since the distance is half, and accounted for when the formulas are entered into the
iteration table, which uses the Gauss–Seidel method. The table to the right of the
iteration table shows the absolute value of the change between iterations. Only
10 iterations are shown, but 100 iterations are conducted to produce the final results
(but as shown here, 10 iterations are sufficient for convergence to 5 decimal places
in temperature change).
The converged temperature results reported in Fig. 8.29 are oriented in the
original grid pattern, and both sides of the rectangle are shown. The heat balances
show that the heat entering the two bottom nodes is equal and opposite to the heat
leaving the top and right faces, a good check for consistency. The net heat flux in is
8 orders of magnitude lower than the heat flux from the bottom surface, and is
considered to be a demonstration of computer precision.
The same spreadsheet is used for the case of a non-zero internal heat generation,
specifically one that yields a maximum temperature of 200  C, and is shown in
Fig. 8.30. The heat generation (qgen) is an input parameter to this problem, but its
value is not specified beforehand. Rather, an iterative solution (on top of the Gauss–
Seidel iteration) is needed. A guess to the heat generation is made, and the
Gauss–Seidel iteration is executed until it converges. Then another guess for
qgen is made. In this case, the “goal seek” feature of MS-Excel was used rather
than a brute force entry of qgen values. The final result is that a heat generation rate
of 12,555.32 W/m3 yields a maximum temperature of 200  C.
The converged temperature results are reported in Fig. 8.31. The heat balances
show that heat leaves all surfaces, and that the total rate of heat loss equals the total
internal heat generation rate.
8.3 Classic 2D Problem Statement 331

INPUT PARAMETERS
k 20 W/m/K thermal conductivity
W 4.2 m full width
H 1.3 m Height
Tsouth 100 oC Base temperature
Teast 0 oC side temperature
Tnorth 0 oC top temperature
qgen 0 W/m^3 volumetric heat generation rate

DERIVED PARAMETERS
Dx 1.05 m horizontal distance between nodes
Dy 0.65 m vertical distance between nodes
Rh 0.080769 horizontal resistance between nodes
Rv 0.030952 vertical distance between nodes
qgen,node 0 heat generation per node

CARTESIAN GRID INDEX

1, 2 2, 2
1, 1 2, 1
ITERATION TABLE (GAUSS-SEIDEL METHOD) CHANGE FROM 'k' to 'k+1'
k T11 T21 T12 T22 T11 T21 T12 T22
0 0.000 0.000 0.000 0.000 - - - -
1 59.115 53.656 17.473 14.544 59.11528 53.65600 17.47308 14.54385
2 70.358 58.199 22.443 16.098 11.24229 4.54305 4.97035 1.55381
3 72.341 58.757 23.206 16.302 1.98371 0.55764 0.76234 0.20478
4 72.630 58.833 23.314 16.331 0.28849 0.07599 0.10847 0.02833
5 72.670 58.843 23.329 16.335 0.04067 0.01058 0.01523 0.00396
6 72.676 58.845 23.332 16.335 0.00570 0.00148 0.00213 0.00055
7 72.677 58.845 23.332 16.335 0.00080 0.00021 0.00030 0.00008
8 72.677 58.845 23.332 16.335 0.00011 0.00003 0.00004 0.00001
9 72.677 58.845 23.332 16.335 0.00002 0.00000 0.00001 0.00000
10 72.677 58.845 23.332 16.335 0.00000 0.00000 0.00000 0.00000

Fig. 8.28 Spreadsheet implementation of 4-node 2D classic problem, with heat generation set
to zero

8.3.4 n  m Model

An “n  m” model (or n  m  p for a 3D space) is defined to be one in which the

number of nodes is specified as a numerical input parameter that can be readily
changed in a computer program. The few-node model is considered to be one where
a specific breakdown of space is decided on beforehand. The few-node model can
be readily solved in a spreadsheet or a computer program. The n  m model cannot
readily be solved with a spreadsheet. This section demonstrates the method and
considers a case study of the computation requirements needed to solve these
problems as a function of the number of nodes.
332 8 Steady-State Conduction

Fig. 8.29 Results from the TEMPERATURE RESULTS (after  iterations)

spreadsheet solution for a 0 0 0 0
4-node model. Temperature
results are shown on both 0 .

.

.
.
0
sides of the centerline, but
calculations were conducted
on the right side, with the 0  . . .  . 0
centerline as an insulated
surface. The heat balances 100 100 100 100
(in Watts) are conducted as a HEAT BALANCES (Right half)
check of the results
INTO BOTTOM 4424.7
INTO RIGHT -1861.6
INTO TOP -2563.1
TOTAL IN 4.3E-05
TOTAL GEN 0.0E+00

SHAPE FACTOR
4.42

INPUT PARAMETERS
k 20 W/m/K thermal conductivity
W 4.2 m full width
H 1.3 m Height
Tsouth 100 oC Base temperature
Teast 0 oC side temperature
Tnorth 0 oC top temperature
qgen 12555.32 W/m^3 volumetric heat generation rate

DERIVED PARAMETERS
Dx 1.05 m horizontal distance between nodes
Dy 0.65 m vertical distance between nodes
Rh 0.080769 horizontal resistance between nodes
Rv 0.030952 vertical distance between nodes
qgen,node 8569.007 heat generation per node

CARTESIAN GRID INDEX

1, 2 2, 2
1, 1 2, 1
ITERATION TABLE (GAUSS-SEIDEL METHOD) CHANGE FROM 'k' to 'k+1'
k T11 T21 T12 T22 T11 T21 T12 T22
0 0.000 0.000 0.000 0.000 - - - -
1 137.511 124.812 119.041 104.987 137.51136 124.81222 119.04119 104.98749
2 186.835 154.667 145.512 114.627 49.32335 29.85526 26.47083 9.63920
3 198.041 158.025 149.916 115.843 11.20588 3.35775 4.40403 1.21587
4 199.723 158.474 150.551 116.009 1.68206 0.44834 0.63490 0.16668
5 199.961 158.536 150.640 116.032 0.23845 0.06219 0.08936 0.02324
6 199.995 158.544 150.653 116.036 0.03346 0.00869 0.01252 0.00325
7 199.999 158.546 150.655 116.036 0.00469 0.00122 0.00175 0.00045
8 200.000 158.546 150.655 116.036 0.00066 0.00017 0.00025 0.00006
9 200.000 158.546 150.655 116.036 0.00009 0.00002 0.00003 0.00001
10 200.000 158.546 150.655 116.036 0.00001 0.00000 0.00000 0.00000

Fig. 8.30 Spreadsheet implementation of 4-node 2D classic problem, with heat generation set to
a value that yields a maximum nodal temperature of 200  C. The “goal seek” feature of MS-Excel
was used for this purpose
8.3 Classic 2D Problem Statement 333

Fig. 8.31 Results from the TEMPERATURE RESULTS (after  iterations)

spreadsheet solution for the
4 node model with internal 0 0 0 0
heat generation set to a value
that yields a maximum 0 . . . . 0
temperature of 200  C.
Temperature results are
shown on both sides of the 0  . . .  . 0
centerline, but calculations
were conducted on the right 100 100 100 100
side, with the centerline as an HEAT BALANCES (Right half)
insulated surface. The heat
INTO BOTTOM -10244.5
balances (in Watts) are
conducted as a check of the INTO RIGHT -6799.2
results. The shape factor is not INTO TOP -17232.4
applicable to a case where TOTAL IN -3.4E+04
internal heat generation TOTAL GEN 3.4E+04
occurs
SHAPE FACTOR
Not Applicable

Indexing Confusion
There are two common ways to index a 2D variable, and it is easy to introduce bugs
in codes by confusing them, as demonstrated in Fig. 8.32. In matrix indexing (the
natural mode in MATLAB), the grid is considered to be row/column, that is, the
first number is the row, the second the column. Cell 1,1 is in the top left corner and
cell n,m is in the bottom right of a grid. In Cartesian indexing (x/y), the first number
is the x position and the second is the y position. Cell 1,1 is at the origin of a
Cartesian grid (bottom left corner) and cell n,m is at the top right. For the general
node i,j, the relationship between the immediate neighbors is different, as shown in
the isolated i,j nodes with their neighbors.

MATRIX (row by column) n X m INDEXING CARTESIAN (x by y) n X m INDEXING

1,1 1,2 1,3 1,4 … 1,m 1,m 2,m 3,m 4,m … n,m
2,1 2,2 2,3 2,4 … 2,m … … … … … …
3,1 3,2 3,3 3,4 … 3,m 1,3 2,3 3,3 3,4 … n,3
… … … … … … 1,2 2,2 3,2 4,2 … n,2
n,1 n,2 n,3 n,4 … n,m 1,1 2,1 3,1 4,1 … n,1

NODE i, j and Neighbors NODE i, j and Neighbors

i-1, j i, j+1
i, j-1 i, j i, j+1 i-1, j i, j i+1, j
i+1, j i, j-1

Fig. 8.32 Matrix and Cartesian indexing scheme

334 8 Steady-State Conduction

The nodal equation for node i,j in the form suitable for the Jacobi iteration
scheme (with superscripts “k” or “k + 1” for iteration number) is:

ðkÞ ðkÞ ðkÞ ðkÞ

ðk Þ T i1, j T i, jþ1 T iþ1, j T i, j1
qgen, i, j þ þ þ þ
ðkþ1Þ Rnorth, i, j Reast, i, j Rsouth, i, j Rwest, i, j
T i, j ¼ for MATRIX indexing
1 1 1 1
þ þ þ
Rnorth, i, j Reast, i, j Rsouth, i, j Rwest, i, j
ðk Þ ðk Þ ðk Þ ðk Þ
ðk Þ T i, jþ1 T iþ1, j T i, j1 T i1, j
qgen, i, j þ þ þ þ
ðkþ1Þ Rnorth, i, j Reast, i, j Rsouth, i, j Rwest, i, j
T i, j ¼ for CARTESIAN indexing
1 1 1 1
þ þ þ
Rnorth, i, j Reast, i, j Rsouth, i, j Rwest, i, j

The heat generation rate is given a superscript “k” to indicate that it may depend on
temperature. It is also possible for some or all of the thermal resistances to be
temperature dependent, and these therefore would be recalculated at each iteration,
but the superscripts are left off for brevity. Notice that 16 elementary operations
(+, , *, /) are needed to execute one of these expressions. It is possible to reduce
that number for the case of equal spacings (when Reast ¼ Rwest, for example).
A pseudo-code (MATLAB based) for the n  m problem that solves for the heat
generation rate required for the maximum temperature to be 200  C using the
Gauss–Seidel iteration with Cartesian indexing is presented in the Appendix to
this chapter. The Gauss–Seidel iteration method will yield faster convergence than
the Jacobi method and requires careful consideration of the order in which the
computer code calculates the next iteration values. Results from modifications of
this program are presented in this section. Figure 8.33 shows the temperature profile
for the base case of a 4.2 m wide  1.3 m high rectangle, with the bottom surface set
to 100  C and the three sides set to 0  C. The centerline (at x ¼ 0) is an insulated

1.2
10 10
10
1 20 20
20
30
y position (m)

30
0.8
40 40
10

0.6 50 50
20
30

60 40
60 50
0.4 70 60
70
70
0.2 80
80
1002000
345

90 90 90 78060
0 0
0 0.5 1 1.5 2
x position (m)

Fig. 8.33 Contour plot of the right half of the 4.2 m  1.3 m rectangle (using a 100  100 grid)
8.3 Classic 2D Problem Statement 335

1.E+10

1.E+09 y = 44.861x1.9327

1.E+08

1.E+07
y = 34.418x1.9158
1.E+06
Count

1.E+05
y = 5.1732x0.928
1.E+04

1.E+03 y = 3.969x0.9111
1.E+02

1.E+01

1.E+00
1 10 100 1000 10000
Number of Nodes

Fig. 8.34 Computational requirements for the 4.2  1.3 rectangle. The convergence criterion
(defined as the maximum change of any node from one iteration to the next) is set to be 1(10)6  C.
The solid symbols are for the Jacobi method and the open symbols are for the Gauss–Seidel method

surface. This figure shows what a well-constructed graphical analysis would show.
The heat transfer is effectively 1D over the left half of the space.
The computational requirements for this case are shown in Fig. 8.34, on a log–
log scale for a 2  2 up to a 100  100 grid. The number of iterations required for
the solution to converge to the defined convergence criterion and the total number
of elementary operations required are shown for both the Jacobi (filled symbols)
and Gauss–Seidel (open symbols) as a function of number of nodes (n ¼ m for all
cases). These values are determined by adding simple counter statements (i.e.,
iter ¼ iter + 1) at appropriate places in the code (not shown in the Appendix code
for brevity). Power law fits to each curve are shown. The results show that the
number of iterations required to converge increases almost linearly (to the 0.9
power approximately) with the number of nodes. That is, the more nodes, the
more iterations are required for the mathematical system to converge to a solution.
Then, since one nodal equation is needed for each node, the number of elementary
operations is proportional to the total number of nodes. The elementary operations
are not exactly a power of 1 higher than the iterations because the centerline nodes
have fewer elementary operations required than interior nodes, and the percentage
of these changes with the number of nodes. The main point is that the computational
requirements increase dramatically with the number of nodes taken, and is a
fundamental challenge of computational methods in general. A good understanding
and experience aimed at addressing specific questions using a minimum number of
nodes is an important and valuable asset.
336 8 Steady-State Conduction

10 43
9 42.8
8 42.6

Average Temperature (oC)

7 42.4
Shape Factor

6 42.2
5 42
4 41.8
3 41.6
2
Average Temperature 41.4
1 41.2
0 41
1 10 100 1000 10000
Number of Nodes

Fig. 8.35 Calculated shape factor as a function of number of nodes for the base case (semi-logx axes)

Grid Insensitivity
A formal way to determine the minimum number of nodes needed to attain a certain
degree of accuracy is to demonstrate grid insensitivity. The general concept is that
the results of a numerical calculation should be independent of the size of the grid,
which is a numerical parameter, not a physical one. That can only be obtained if the
calculation is done at least twice with two grid sizes. If the results agree sufficiently,
then the coarser grid is shown to be sufficient.
Demonstration of grid insensitivity requires that something specific be compared
at different grid sizes. There is, however, some art required to appropriately define a
specific outcome, such as a temperature at a specific location, or an average
temperature, or a heat transfer rate (which is generally a good global measure).
For example, Fig. 8.35 shows two different measures for this case study; the
predicted shape factor (heat transfer rate from the bottom surface to the other
three surfaces) and the average temperature as a function of the number of nodes
(using semi-logx axes). This plot suggests that the shape factor does NOT show grid
insensitivity. As the number of nodes increases, the heat transfer rate continues to
increase, with no sign of leveling off. However, the temperature (which is plotted in
a narrow 2  C temperature band) shows a leveling off to a value of 41.6  C. The
1-node model actually gives a better agreement than the 2  2 model.

8.3.5 Comment on Exact Solution

The exact solution to this problem involves solving a boundary-value type partial
differential equation, resulting in an infinite series solution for the temperature field.
8.3 Classic 2D Problem Statement 337

Evaluation of the heat transfer rate between the bottom and side surfaces yields a
non-converging power series. That is, the heat transfer rate is infinite, due to the
discontinuity in temperature that occurs at the corner. The mathematical statement
of the classic problem is a physical impossibility, however elegant.
h i
X
1 ð1Þnþ1 þ 1 cos ðnπ  1Þ
q¼
n¼0
n tanhðnπw=H Þ

Even though the magnitude of successive terms decreases with n (becoming 1/n for
large n), this series fails the formal testing for convergence. The culprit for this
behavior is the physically unattainable discontinuous temperature jump at the
corner, which results in locally infinite heat transfer rates. This result explains
why the numerically predicted shape factor increased with the number of nodes,
as smaller and smaller nodes were placed in the corner.

8.3.6 Results with Internal Heat Generation

The same system was solved with an internal heat generation rate (qgen), specifically,
that value that yields a maximum temperature of 200  C. The code (reported in the
Appendix) was written with a Regula-Falsi scheme in which a high and low value of
heat generation was initially selected (assisted by 1-node results). Use of the low value
(0 initially) results in a maximum temperature below the target, and use of the high
value (twice the 1-node prediction was sufficient) results in a maximum temperature
above the target. An iteration loop was implemented in which the next “guess” for heat
generation was taken to be the average of the high and low values (a Golden Ratio
algorithm would be a little faster, but requires more explanation). If the maximum
temperature attained at this average value exceeded the target, then the high value is
replaced by this new value, and vice versa if the maximum temperature is too low. A
Gauss–Seidel iteration loop is used for each “guessed” value of qgen.
The code for this problem has three nested loops. The outer loop simply runs the
algorithm for increasing number of horizontal nodes (n, with vertical nodes, m, set
equal to n). The next loop executes the Regula-Falsi method in which a value of
heat generation is “guessed.” The inner loop executes the Gauss–Seidel iteration
scheme with this selected value.
Temperature contour results for a 100  100 grid are shown in Fig. 8.36. There
are strong temperature gradients toward the three isothermal surfaces as heat
generated in the interior must conduct to these surfaces. The peak temperature is
located on the centerline, and somewhat closer to the hotter bottom surface.
The effect of grid size is shown in Fig. 8.37 in which the internal heat generation
rate required to achieve a maximum temperature of 200  C is plotted against the
number of nodes. Grid insensitivity is demonstrated above approximately
100 nodes (a 10  10 grid). However, even the 2  2 model gives a predicted heat
generation rate that is within 10 % of the grid insensitive value. The 1-node model
338 8 Steady-State Conduction

20 20
1.2 40 40 20
60 60 40
80 80
100 100 80 60
1 120 120 100
140 140 12
0
160
y position (m)

0.8 160
180

80

20
140
18

40
100
0

60
0.6
200

120
160
0.4
180 180
0.2 160 0

80
160 14

20 0
100

4 60
140 140
120 120 120
0
0 0.5 1 1.5 2
x position (m)

Fig. 8.36 Temperature contour for the case with internal heat generation that yields a maximum
temperature of 200  C. A 100  100 grid was used for this plot

16000

14000

14050.20551
14049.45736
14045.98868
14048.36914

14048.84524
14042
14037
14026
14002
13990
13977
13893
Heat Generation Rate (W/mX
)

13717
13383
13249

12000
12555
to achieve Tmax = oC

10000
8915

8000

6000

4000

2000

0
1 10 100 1000 10000
Number of Nodes

Fig. 8.37 Internal heat generation rate required to achieve a maximum temperature of 200  C as a
function of number of nodes. Values are shown to ten decimal places to demonstrate computer
precision issues

exhibits a large deviation, but places the system in the right neighborhood. Any
input parameter value can be changed to a vastly different problem, and the 1-node
model will find an excellent first approximation.
The numerical values are shown (to ten significant figures) in Fig. 8.37 for the
cases that appear from the plot to be equal. These values are not monotonically
Workshop 8.1. Graphical Method on Classic 2D Problem 339

increasing (which would suggest the result was still sensitive to the grid size), but
rather fluctuate about an average value. This behavior is attributed to perhaps both
computer precision, and that a specific convergence criterion was used (1(10)6) for
both the Regula-Falsi algorithm and the Gauss–Seidel algorithm. There is a curious
bump for the case of 25 nodes (5  5) which is similarly related to computational
issues, but in this case, it is the graininess of the spatial grid at fault. The basic shape
of this curve would be the same, with differences in minor detail if a different
convergence criterion were selected.

Workshop 8.1. Graphical Method on Classic 2D Problem

The graphical conduction method is a useful tool for helping visualize the flow of
heat and estimating conductive resistances in a complicated 2D solid under quasi-
steady state conditions (the temperature may be slowly changing with time).
It involves first specifying two surfaces (1 and 2) that are set to fixed temperatures,
then sketching isotherms and heat flow lines (analogous to streamlines in fluid flow)
in a prescribed manner. The rate of heat transfer (for a length L into the page)
between the two surfaces is obtained from the sketch. A conduction shape factor
(S) can be defined as: q12 ¼ kLSðT 1  T 2 Þ. Published exact solutions for specific
shapes can be used to compare results of this approximate method.
The base of a long 2  1 m2 rectangular structural member is maintained at
temperature T2 (¼100  C) and the other three sides are maintained at temperature
T1 (¼0  C). The member has a thermal conductivity k, and heat flows between
the base and the three sides at a rate q12. Use the graphical method to estimate the
conduction shape factor, S.

1m

2m

Graphical method
1. Draw object to scale (already done here).
2. Use symmetry to identify any insulated surfaces and “hatch” them (note:
insulated surfaces are also heat flow lines).
3. Carefully sketch the average temperature isotherm, creating two temperature
increments (note: isotherms are perpendicular to insulated surfaces).
340 8 Steady-State Conduction

4. Carefully sketch average isotherms to create four total temperature increments.

Decide if it’s worth going to 8, then 16, then 32. . ..
5. Carefully sketch heat flow lines, keeping all boxes created approximately square
(Note: heat flow lines are perpendicular to isotherms).
6. S ¼ M/N, where M is the number of heat flow lanes created, and N is the number
of temperature increments.

Workshop 8.2. Graphical and Thermal Resistance Practice

A steam pipe of outer diameter D ¼ 0.1 m is encased in a concrete (k ¼ 0.3 W/m/K)

square channel of dimension w ¼ 0.4 m. The steam pipe temperature is maintained
at T1 ¼ 120  C, while the outside of the concrete is maintained at T2 ¼ 20  C.
Estimate the rate of heat transfer through the concrete layer per unit length using
three methods: conduction shape factor (“exact” solution taken from the literature),
the graphical method, and by creating an approximate conductive resistance.

0.4 m

D = 0.1 m 0.4 m = w

RESULTS:
Method q12 /L(W/m) % Differrence
Exact Conduction Shape Factor:
2p q12 L NA
S= =
ln(1.08w/D) k12 (T2 − T1)

Graphical Method

Conductive Resistance

Bonus: If the outside were exposed to air at 20  C and a convective/radiative heat

transfer coefficient of 15 W/m2/K, would the heat transfer rate change substan-
tially? (Justify with a resistance model calculation.)
Workshop 8.4. Classroom Fun! 341

Workshop 8.3. Graphical Analysis Practice

Conduct a graphical analysis on each of these five 2D shapes. Record the value of
the shape factor next to each.


q Mlanes
Graphical Method : ¼k ðT 1  T 2 Þ ¼ kSðT 1  T 2 Þ
L N increments

200 x 50 200 x 50 200 x 100

Heavy line is T1, light line is T2, hatched lines are insulated

Workshop 8.4. Classroom Fun!

For a class of 24 students, arrange them into a 6  4 system, where each student is
assigned a node. The number of rows and columns can be chosen depending on the
class size (for example, a class of 25 could use a 5  5 grid). The method is applied
to a system where Δx ¼ Δy, so that the iteration formula results in the next guess for
each temperature to be equal to the geometric average of the four neighbors. For
students on the perimeter, one of the neighbors is fixed. For students at the corners,
two are fixed. Each student fills out a Worksheet, which requires verbally obtaining
the current iteration neighbor’s values (and a good ice-breaker, especially for shy
students).
342 8 Steady-State Conduction

0 C 0 C 0 C 0 C 0 C 0 C
0 C • • • • • • 0 C
0 C • • • • • • 0 C
0 C • • • • • • 0 C
0 C • • • • • • 0 C
100  C 100  C 100  C 100  C 100  C 100  C

Worksheet instructions: use integer values only.

1. For the first iteration, use engineering judgment to make a guess as to your
temperature based on your location. Enter in “MY GUESS” column, iteration
1 row.
2. Ask your four neighbors what their guesses are and enter them in the appropriate
column for iteration 1.
3. Your next guess (iteration 2) is the average of your neighbors. Iterate until
convergence (or you run out of time).
4. Enter your final results on a grid visible to the entire class.

Iteration MY GUESS North East South West

1
2
3
4
5
6
7
8
9
10

An example of the final result of this exercise is shown in Fig. 8.38.

Workshop 8.5. Storage of Energetic Materials—The Chemically

Reacting Pile

A pile of material that slowly reacts chemically (i.e., coal dust, fertilizer) is stored in
a long w  H (use 2  1 m2 as default) rectangular bin. The base is maintained at
ambient temperature T1 ¼ 25  C and the other three sides are exposed to ambient
air with an effective heat transfer coefficient of U ¼ 1.2 W/m2/K (across a thermal
channel consisting of a conductive wall resistance and a convective/radiative
resistance to ambient in series). The material has a thermal conductivity
k ¼ 0.6 W/m/K, and it generates heat internally at a constant volumetric rate (qgen
with units W/m3, so that the rate of heat generation is qgen times the volume). After
storage for a sufficiently long time for a steady-state to be established, it is found
Workshop 8.5. Storage of Energetic Materials—The Chemically Reacting Pile 343

Fig. 8.38 Photo of results at the end of a “Class Fun” workshop. Approximately 30 min was
allotted for this exercise. The results were tabulated before the system was fully converged as
shown by the asymmetry. Notice that each node is the first name of the student, making this a “few-
node” model

that the temperature at the center of the pile is T0 ¼ 40  C. Determine the heat
generation rate, qgen, and the maximum observed temperature and its location. The
material is known to spontaneous combust if it reaches a temperature
Tignition ¼ 60  C.
344 8 Steady-State Conduction

• Perform a 1-node analysis to obtain a closed-form solution in terms of system

parameters (w, H, k, qgen, h, T1, T0). Calculate the required heat generation rate
and the maximum safe size of the pile (for various aspect ratios, w/H).
• Conduct a 4-node analysis (see over for thermal resistance network) and use the
MOSS (Gauss–Seidel method being one variation) to obtain a second approxi-
mation. Using symmetry, the model considers only the right side of the pile, with
equal size nodal elements with the same aspect ratio (w/H) as the pile. Consider
the temperature at the center to be the average temperature of two left side nodal
elements. Set up a spreadsheet or program code with a block of input parameters,
then an iteration table that implements the MOSS with sufficient iterations for
convergence. Once the model is in place, use brute force, or “goal seek” or create
a “while loop” to find the heat generation rate.
• Write program code to conduct an analysis with a finer “n  m” grid. Determine
the maximum temperature in the pile and its location. Explore the effect of grid
size on heat generation rate and on the computational requirements.
• Investigate the behavior of this material for different sizes and aspect ratios (i.e.,
peak temperature vs. height and/or width). Determine the maximum size pile
that can be used without spontaneous combustion.

In general, for node i, with neighbors j, an energy balance is:

dT i X Tj  Ti
Ci ¼ qgen  V i þ ð¼ 0 @ Steady StateÞ
dt j
Rij
Workshop 8.6. Cold Plate 345

where Vi is the volume associated with the nodal element, and Rij is the thermal
resistance between the node “i” and its neighbor “j”. At steady-state, expanding the
summation (noting that Ti is common to all summation terms):

X Tj X 1
0 ¼ qgen  V i þ  Ti
j
Rij j
Rij

Note that either Celsius or Kelvin temperatures can be used (but consistently)
because in the original energy balance temperature appears as a temperature
difference everywhere. Rearranging for the nodal temperature:

X Tj
qgen  V i þ
j
Rij
Ti ¼ X 1

j
Rij

For example:


T1 T 2, 2 T 1, 1
qgen  ðH=2Þ  ðw=4Þ  L þ þ þ
Ra þ Rd 2Rb 2Ra
T 1, 2 ¼

1 1 1
þ þ
Ra þ Rd 2Rb 2Ra

Workshop 8.6. Cold Plate

In a refrigeration application, the lower left corner of an 11.2 cm by 34.4 cm block

of steel (k ¼ 73 W/m/K) is maintained at a temperature T0 ¼ 20  C (by an
evaporating refrigerant) and the top surface is exposed to a fluid at T1 ¼ 5  C
with a convection coefficient of 35 W/m2/K and exchanges radiation with walls at
Tw ¼ 25  C. The other three sides are insulated. For the two nodal networks shown
(one places the nodes, then defines the volumes, Fig. 8.39; the other defines
volumes, then places the nodes, Fig. 8.40), solve for the temperatures of all
nodes, and for the rates of heat transfer from the cold corner, to the fluid being
cooled (by convection), and to the walls (by radiation).
346 8 Steady-State Conduction

Walls @ Tw00 =25°C

Fluid @ T∞= 5°C with h = 35 W/m2/K

emissivity, ε = 0.85
f 5 6 7 8 Δy = 5.6 cm

e 1 2 3 4

a b c d

T° = −20°C
Δx = 8.6 cm

Fig. 8.39 Nodal network where the nodes are placed, then the volumes (not shown) are defined
around them

Walls @ Tw00 = 25°C

Fluid @ T∞= 5°C with h = 35 W/m2/K

emissivity, ε = 0.85

Δy = 5.6 cm
5 6 7 8

1 2 3 4

T° = −20°C
Δx = 8.6 cm

Fig. 8.40 Nodal network where the volumes are defined, then the nodes are placed in the center
Appendix. MATLAB Code for Classic 2D Problem 347

Appendix. MATLAB Code for Classic 2D Problem

%rectangle_GaussSeidel_TMAX_VARYn.m George Sidebotham 2012, revised 2014

%Program to calculate temperatures and heat flux from a rectangle whose base
%is fixed at one temperature, and the other 3 sides are fixed at another,
%with a constant heat generation throughout. This version finds the Qdot
%that yields a defined maximum temperature
clear all
clc
%INPUT PARAMS
k = 20; %W/m/K, thermal conductivity
width = 2.1; %m, halfwidth of rectangle
height = 1.3; %m, height of rectangle
length = 1; %m, depth "into the page"
Tsouth = 100; %oC, bottom surface temperature (base)
Tc = 0; %oC, temperature of other 3 sides
Teast = Tc; %East wall temp
Tnorth = Tc; %North wall temp
Tmax_set = 200; %oC, set maximum temperature to find
converge = 1e-6; %convergence criterion

%1-Node Model
Rnorth = height/2/(k*width*length);
Rsouth = Rnorth;
Reast = width/2/(k*height*length);
Qdot_1node = (-Tc/Rnorth - Tc/Reast - Tsouth/Rsouth + ...
Tmax_set*(1/Rnorth+1/Reast+1/Rsouth))/(height*width*length);

%NUMERICAL GRID AND DEFINED PARAMS

CASES = 100;
Qset(1:CASES) = Qdot_1node; %allocate final heat generation rate
%LOOP FOR NUMBER OF NODES
for n = 2:CASES; %number of elements in x direction (columns)
m = n; %number of elements in y direction (rows)
Qdot_low = 0; %heat generation yields Tmax < Tmax_set
Qdot_high = 2* Qdot_1node;%heat generation yields Tmax > Tmax_set
dx = width/n; %x grid size in x direction
dy = height/m; %y grid size in y direction
Rh = dx/k/dy/length; %oC/W, horizontal conductive resistance
Rv = dy/k/dx/length; %oC/W, vertical conductive resistance
T(1:n,1:m) = Tc; %"initialize" temperature to cold temperature
Tnext(1:n,1:m) = Tc; %"initialize" next guess
for i = 1:n %define x location vector
x(i) = dx*(i-1/2);
end
for j = 1:m %define y location vector
y(j) = dy*(j-1/2);
end
%LOOP FOR REGULA-FALSI (High/Low) METHOD OF FINDING HEAT GENERATION
while abs(Qdot_high - Qdot_low) > converge
Qdot = 1/2*(Qdot_high + Qdot_low); %Next guess for heat generation
qgen = Qdot*(dx*dy*length); %W, heat generation per cell
change = converge+eps; %needed to get into main iteration loop...
%LOOP FOR GAUSS-SEIDEL ITERATION LOOP
while change > converge
change = 0; %reset for this iteration
%Bottom left corner
348 8 Steady-State Conduction

Tnext(1,1) = (qgen + T(2,1)/Rh + T(1,2)/Rv +…

Tsouth*2/Rv)/(1/Rh + 3/Rv);
%Bottom row
for i = 2:n-1
Tnext(i,1) = (qgen + T(i+1,1)/Rh + Tnext(i -1,1)/Rh +…
T(i,2)/Rv + 2*Tsouth/Rv)/(2/Rh + 3/Rv);
end
%Bottom right corner
Tnext(n,1) = (qgen + T(n,2)/Rv + Teast*2/Rh + Tsouth*2/Rv + ...
Tnext(n-1,1)/Rh)/(3/Rh+3/Rv);
%Right wall
for j = 2:m-1
Tnext(n,j) = (qgen + T(n,j+1)/Rv + Teast*2/Rh +…
Tnext(n,j-1)/Rv + T(n-1,j)/Rh)/(3/Rh +2/Rv);
end
%Top right corner
Tnext(n,m) = (qgen + Tnorth*2/Rv + Teast*2/Rh +…
Tnext(n,m-1)/Rv + T(n-1,m)/Rh)/(3/Rv + 3/Rh);
%Top row
for i = 2:n-1
Tnext(i,m) = (qgen + Tnext(i+1,m)/Rh + T(i -1,m)/Rh +…
T(i,m-1)/Rv + 2*Tnorth/Rv)/(2/Rh + 3/Rv);
end
%Top left corner
Tnext(1,m) = (qgen + Tnext(2,m)/Rh + T(1,m -1)/Rv +…
Tnorth*2/Rv)/(1/Rh + 3/Rv);
%Left wall
for j = 2:m-1
Tnext(1,j) = (qgen + Tnext(1,j+1)/Rv + T(1,j-1)/Rv +…
T(2,j)/Rh)/(1/Rh +2/Rv);
end
%Interior cells
for i = 2:n-1
for j = 2:m-1
Tnext(i,j) = (qgen + T(i,j+1)/Rv + T(i+1,j)/Rh + ...
Tnext(i,j-1)/Rv + Tnext(i-1,j)/Rh)/(2/Rh + 2/Rv);
end
end
change = max(max(abs(T-Tnext))); %Find the maximum change between
iterations
T = Tnext; %update for next iteration
end
%RESET THE APPROPRIATE HIGH OR LOW VALUE
if max(max(T)) > Tmax_set
Qdot_high = Qdot;
else
Qdot_low = Qdot;
end
disp(sprintf('Qdot low, high n= %g %g %g',Qdot_low, Qdot_high, n))
end
%OUTPUT HEAT
Qset(n) = Qdot;
end
 Part IV
Open Systems: Convection and Heat
Exchangers
Nusselt Number Correlations
9

It can be argued that the exchange of energy between a solid and a fluid is important
in most, if not virtually all, practical situations for which heat transfer plays a role.
In the mug of coffee problem, the prediction of the cooling rate is most sensitive to
the values of the convection coefficient and radiation coefficient used to model the
heat transfer rate between the outer side wall and the surroundings. The side wall
channel has the lowest thermal resistance of the three parallel channels (because it
has the largest exposed area), but within this channel, the outer resistance is the
largest. The effective Biot number of the combined system (h00 L00 /kf) is low. The
initial mug heating period is most sensitive to the value of the convection coeffi-
cient chosen between the coffee and the mug inner wall.
The goal of this chapter is to determine the convective heat transfer coefficients
between the coffee and the mug and between the mug and ambient that have been
used in the various thermal models. The general approach to using Nusselt number
correlations is developed, and applied to the mug inner and outer walls. Rough
order of magnitude “Rules of Thumb” presented in Chap. 3 are thereby tested for
this case study. The next chapter explores more fundamentally the physics of the
thermal boundary layer.
Nusselt number correlations can be obtained from a variety of printed and online
sources, and only a few relevant examples are introduced here, as needed.

9.1 Flow Classification and Dimensionless Parameters

Convection flows can be classified as being external or internal. In external flows,

the fluid surrounds the object. For example, a sailor testing for the direction of the
wind wets her finger and holds it vertically. The fluid (air) surrounds the finger
making this scenario one of external flow. On the other hand, a refrigerant that
enters a condenser or evaporator does so inside a tube that enters a heat exchanger.
The refrigerant is completely surrounded by the inner wall of the tube making this

# Springer International Publishing Switzerland 2015 351

G. Sidebotham, Heat Transfer Modeling: An Inductive Approach,
DOI 10.1007/978-3-319-14514-3_9
352 9 Nusselt Number Correlations

scenario one of internal flow. Internal flows and heat exchangers will be addressed
in separate chapters.
There is also a distinction between local and average heat transfer coefficients.
For the sailor, the human nervous system can discern that the rate of heat loss on the
upwind side is greater than that on the downwind side, and thereby determine the
direction of the wind. Therefore, the local convection coefficient on the upwind side
is different than the downstream value. In fact, the local convection coefficient
varies continuously across the entire surface. However, as is common in engineer-
ing analysis, if the total rate of heat loss is of interest, an averaged heat transfer
coefficient would be used that yields the correct total heat transfer rate, but with no
distinction with position.
Finally, flows can be classified as being forced, natural, or mixed. In forced
convection, there is a discernible relative flow velocity between the solid object and
the fluid that surrounds it (for external flows). Therefore, if the sailor can discern the
wind direction by holding up a moist finger, then there is a significant wind speed,
and the finger experiences a forced convection. Even a gentle breeze can be
discerned this way. On the other hand, if there is a dead calm, the sailor will not
experience a spatial difference and will conclude that there is no wind. However,
“hot air rises.” The sailor’s finger is near body temperature and is warmer (and
therefore less dense) than the surrounding air, so this layer of air will rise like a
helium balloon. The word “convective” implies that there is fluid motion. In forced
convection, the fluid motion is imposed. In natural convection, the fluid motion is
induced by the heat transfer process itself.
Mixed convection flows are for those cases where there is a forced velocity, but
buoyancy effects are strong enough to influence the flow field as well. A legitimate
modeling strategy is to calculate both a forced and a natural convection coefficient,
and use the higher of the two. This approach would underestimate the convection
coefficient. Alternatively, a fourth-order polynomial fit can be made across a
defined transition range that matches the slope and value at locations before and
after the crossing point, where the correlations for natural and forced flow are equal.
Natural convective flows are driven by buoyancy and therefore require a gravi-
tational field. One of the many engineering challenges of space travel is to design
systems that are, for economic reasons, tested on the Earth where natural convective
cooling effects (of electronic components, for example) occur. However, these
systems are intended to be operational on a spacecraft which experiences micro-
gravity. Fire safety is of particular concern because the high gas phase temperatures
due to combustion cause intense upward flows on the Earth, but not on a spacecraft.
For engineering applications, convection coefficients for a variety of flow
geometries have been correlated in terms of the dimensionless Nusselt number,
defined as:
hx
Nux ¼
kf

where h is the convection coefficient, x is the appropriate linear dimension for the
problem in question, and kf is the thermal conductivity of the fluid. Special care
9.1 Flow Classification and Dimensionless Parameters 353

must be taken to ensure that the correct dimension for the published correlation is
used. For example, for a pipe of diameter D, length L, the dimension used differs
depending on the flow situation (D for an external cross-flow, L for a vertical pipe in
a still room). Also, the Nusselt number is not to be confused with the Biot number
(Bi ¼ hx/k, where k is that of the solid), which has an entirely different physical
interpretation, namely, the ratio of a conductive to a convective thermal resistance
associated with a solid. The Nusselt number will be shown in the next chapter to be
the ratio of the conductive to convective thermal resistance across a thermal
boundary layer.
Convection coefficients depend on the nature of the flow field (higher speeds
mean better heat transfer) and the nature of the fluid in question (better conducting
fluids with large thermal masses (ρc) convect better). Therefore, the Nusselt number
depends on two other dimensionless numbers, one characterizing the flow field, the
other characterizing the nature of the fluid.
For forced convection, the flow is characterized by the Reynolds number:

ρVx Vx
Rex ¼ ¼
μ ν
where V is the forced flow velocity, ρ is the fluid density, μ the dynamic viscosity,
ν ¼ μ/ρ is the kinematic viscosity. The other dimensionless number is a fluid
property, the Prandtl number, which is the ratio of the kinematic viscosity to the
thermal diffusivity of the fluid.
ν
Pr ¼
α

For natural convection, there is no forced velocity, but there is a flow. The
temperature differences in the fluid give rise to density differences which, in a
gravitational field, cause buoyant forces that induce fluid flows. The dimensionless
number that characterizes this flow is the Grashof number:

gβjðT w  T 1 Þjx3
Gr x ¼
ν2

where g is the acceleration due to gravity, Tw is the wall surface temperature, T1 is

the free stream fluid temperature, and β is the coefficient of expansion (how the
density changes with temperature) given by:


1 ∂ρ
β¼
ρ ∂T P

For an ideal gas, β ¼ 1/T, where T is the absolute temperature (R or K). This relation
can be derived using the ideal gas law, P ¼ ρRT. The absolute value of the
temperature difference between surface and fluid is required to prevent negative
values of an inherently positive quantity.
354 9 Nusselt Number Correlations

Fig. 9.1 Schematic of side

wall of coffee mug

Many (if not most) of the natural convection correlations involve yet another
combination dimensionless parameter, the Rayleigh number, defined by:

gβðT w  T 1 Þx3
Rax ¼ Gr x Pr ¼
να
A few forced convection scenarios could be imagined for the mug of coffee
problem. During the initial pouring, it could be argued that there is a forced
convection, and a characteristic velocity could be wildly stabbed at by estimating
the time it takes to pour the coffee and distance travelled (like the height of the
mug). Another variation would be to stir the coffee, in which case a velocity could
be defined based on the stirring rate and diameter of the rotation. Yet another would
have the impatient train commuter blowing on the top. Notice, however, that results
from previous chapters suggest that blowing on the sides would actually be more
effective. This result is counterintuitive as the vast majority of people would blow
on the top, not the sides.
In the scenario detailed here, the mug of coffee is simply placed on a table and
allowed to cool without intervention. This case is clearly one of natural convection.
There is no imposed flow velocity, but there is fluid motion induced by gravity.
Another classification that can be made is whether the flow is internal or
external. Internal flows are generally cases where fluid flows in some type of
duct, and would therefore almost always be a forced convection case. In the present
case, it would be easy to at first think that the inner convection environment is an
internal flow, but it really is not a forced flow in a duct. Imagine, rather, that the mug
has a very large diameter.
The schematic of Fig. 9.1 shows the side wall of a mug of coffee as a vertical
wall with two different fluids on either side. On the left, the coffee is hotter than the
wall, and a plunging flow in a thermal boundary layer next to the wall (indicated by
the dashed line) is shown. The fluid near the wall is slightly colder, and therefore
denser, than the bulk of the fluid far from the wall, and gravity induces a
9.2 Nusselt Number Example Calculation 355

downward flow. On the right (air) side, the wall is hotter than the free stream air
temperature, and the air in the thermal boundary layer is less dense than ambient air
far from the mug, and gravity induces an upward flow. Note the no-slip condition,
where the velocity of the fluid in contact with the solid is motionless.
So for modeling purposes, the coffee side (inner surface) can be classified as a
natural convection on a vertical heated plate of height H, and width πD. The air side
(outer surface) can be classified as a natural convection on a vertical cooled plate of
height H0 and width π(D + 2w).
A difficulty with this case is that the temperatures of the two fluids can be
considered to be known (at an instant in time), but the temperatures of the two solid
surfaces that they see are not known. The convection coefficients depend on the
temperatures of the solid surfaces for these natural convection cases. The way
around that circular argument is to develop an iterative procedure, a trial and
error approach, developed later in this chapter. The key step in this iterative
procedure is the calculation of the convection coefficients for an assumed value
of surface and fluid temperatures. The problem will now be recast in a way that
obviates, for now, the need for iteration.

9.2 Nusselt Number Example Calculation

Problem Statement
Calculate the convection coefficient between a vertical solid surface of height
H ¼ 0.098 m maintained at a temperature Tsurface and an adjacent stagnant fluid
maintained at temperature Tfluid. Consider a range of temperatures for air and liquid
water.

Analysis
A power law relationship (y ¼ axb) between the Nusselt number and Rayleigh
number has been shown to correlate a wide range of data for natural convection
situations:
 m
NuH ¼ C Gr H Prf

The constants C and m take on different values depending on the value of the
Rayleigh number (Ra ¼ GrHPrf). Table 9.1 lists published values for two ranges
that occur in the case study under investigation. These ranges span nine orders of
magnitude in the Rayleigh number, and thereby cover most practical problems of
interest. For cases where the Rayleigh number is below the lowest range can be
handled approximately by either treating the fluid as non-moving (a solid) or using
the lowest value of Rayleigh number. This table will be modified shortly because it
leads to a discontinuity at the point of transition (at Ra ¼ 109).
356 9 Nusselt Number Correlations

Table 9.1 Published values of constants used in the Nusselt number correlation for natural
convection on vertical plates
Rayleigh number range Flow character C m
(10)4–(10)9 Laminar 0.59 0.25
(10)9–(10)13 Turbulent 0.021 0.40

Fig. 9.2 Properties

of saturated liquid water
at three temperatures

9.2.1 Property Values and Their Temperature Dependency

All the dimensionless parameters depend on fluid properties, whose values depend
to some extent on the value of the temperature. There are a variety of sources of
fluid properties, and care must be taken to properly interpret the units and
exponents. Figure 9.2 lists fluid properties, for saturated liquid water at three
temperature values, taken from an online source.1 The compressed liquid approxi-
mation, namely that thermodynamic properties are generally functions of tempera-
ture only (not pressure), is invoked in using these parameter values (as long as the
pressure is above the saturation pressure for that temperature).
Figure 9.3 lists fluid properties for atmospheric air at the same temperatures.2
The only property value that depends significantly on pressure is the density, and
the ideal gas law can be used when needed for problems at other than atmospheric
pressure.
In both of these tables, the properties listed in the reference source are listed in the
top section, and those properties derived from the others are listed in the “DERIVED”
section. For air, properties were listed only at 60 and 80  C, and the values at 70  C
are from linear interpolations. Different sources may list different properties.
To organize these property values and their combinations, it is possible to
consider five of them as being primary property values (ρ, cp, k, μ, β) and the others

1
http://www.engineeringtoolbox.com/water-thermal-properties-d_162.html. Accessed 10/19/
2010.
2
http://www.engineeringtoolbox.com/air-properties-d_156.html. Accessed 10/19/2010.
9.2 Nusselt Number Example Calculation 357

Fig. 9.3 Property values for

air at atmospheric pressure

as derived from them (ν, α, Pr). These property values are shown for air and water
over a temperature range of 0–100  C in Fig. 9.4, designed to compare air against
water values using a semi-logy plot, and Fig. 9.5, which plots air and water on
separate linear y axes to observe the temperature dependence. All the property
values for water exceed those of air, except for the coefficient of thermal expansion.
Notice that liquid water is not truly incompressible in heat transfer applications.
Rather, if significant temperature variation occurs in water, the mild temperature
dependency of density results in a gentle motion which profoundly affects its heat
transfer behavior. A truly incompressible fluid would not experience a natural
convection flow because that flow is due to horizontal density variation in a
gravitational field. A rather interesting situation occurs when water is near its
freezing point because the density of the liquid increases ever so slightly from
0 to 4  C. The temperature dependency varies from property to property and
between air and water. The specific role each property plays on the convection
coefficient will be considered after they are calculated.
For most practical applications, property values are treated as suitably chosen
constants, not functions of temperature, in the evaluation of convection coefficients.
For natural convection problems, the convection coefficient depends directly on the
temperature difference driving the heat because of the Grashof number. For forced
convection, convection coefficients are generally considered to be independent of
temperature in most cases. This practice will be further assessed after all the factors
that contribute to convection coefficients are combined.

9.2.2 Rayleigh Number

Figure 9.6 shows the Rayleigh Number plotted (on semi-logy axes) against temper-
ature difference for liquid water and air. All properties are evaluated at 70  C. The
Raleigh number is some four orders of magnitude larger in water than air. The main
culprits for this difference are the density and thermal conductivity, with a minor
specific heat effect. The coefficient of thermal expansion and dynamic viscosity
358 9 Nusselt Number Correlations

Thermal Conductivity (W/m/K)

1
1000
H2O(L)
H2O(L)

Density (kg/m3)
100

0.1 10

Air Air
1

0.01 0.1
0 20 40 60 80 100 0 20 40 60 80 100
Film Temperature (oC) Film Temperature (oC)
10000 0.01

Viscosity (kg/m/s)
Specific Heat (J/kg/K)

H2O(L)
0.001 H2O(L)

1000
Air
0.0001

Air

100 0.00001
0 20 40 60 80 100 0 20 40 60 80 100
Film Temperature (oC) Film Temperature (oC)
Coefficient of Thermal Expansion (1/K)

0.01

Air
0.001

H2O(L)
0.0001

0.00001
0 20 40 60 80 100
Film Temperature (oC)

Fig. 9.4 Fluid properties for air and water between 0 and 100  C. Semi-logy axes used to compare
individual property values of liquid water and atmospheric pressure air

favor air. These comparisons will be revisited shortly. It is emphasized that the
x-axis represents the difference between the fluid and surface temperature, and not
on their values. A conceptual dilemma arises because the actual temperature of the
fluid in the thermal boundary layer changes from one extreme to the other, and so do
all the property values (some more than others). In practice, all property values are
taken at the film temperature, the geometric average of the surface and fluid
temperatures.
Technically, then, Fig. 9.6 applies to a specific temperature value. For example,
for a temperature difference of 100  C, the surface temperature would be 20  C and
the fluid temperature would be 120  C, or vice versa. In practice however, this
9.2 Nusselt Number Example Calculation 359

0.8 Thermal Conductivity (W/m/K) 0.035 1200 1.4

Density (kg/m3)
0.7 0.03 1000 1.2
0.6 1
H2O(L) 0.025
Liquid Water

800

Liquid Water
0.5 Air
0.02 0.8
600 H2O(L)
0.4

Air
Air
0.015 Air 0.6
0.3 400
0.01 0.4
0.2
0.005 200 0.2
0.1
0 0 0 0
0 20 40 60 80 100 0 20 40 60 80 100
Film Temperature (oC) Film Temperature (oC)
4600 Specific Heat (J/kg/K) 1200 1.6E-03 Viscosity (kg/m/s) 2.5E-05
4100 1.4E-03
1000
Liquid Water

3600 2.0E-05
1.2E-03

Liquid Water
3100 H2O(L) 800
1.0E-03 1.5E-05
2600 Air H2O(L)
600
Air

8.0E-04

Air
Air
2100
6.0E-04 1.0E-05
1600 400
1100 4.0E-04
200 5.0E-06
600 2.0E-04
100 0 0.0E+00 0.0E+00
0 20 40 60 80 100 0 50 100
Film Temperature (oC) Film Temperature (oC)

8.0E-04 Coefficient of Thermal 4.0E-03

7.0E-04 Expansion (1/K) 3.5E-03
Liquid Water

6.0E-04 3.0E-03
5.0E-04 2.5E-03
Air

4.0E-04 2.0E-03
3.0E-04 H2O(L) 1.5E-03
2.0E-04 Air 1.0E-03
1.0E-04 5.0E-04
0.0E+00 0.0E+00
0 20 40 60 80 100
Film Temperature (oC)

Fig. 9.5 Fluid properties of air and water between 0 and 100  C. Linear axes with air as a
secondary scale

distinction can usually be ignored, provided a reasonable film temperature for the
problem at hand is selected, using sound engineering judgment.

9.2.3 Nusselt Number

Application of the semi-empirical power law expression (Nu ¼ C(Ra)m) results in

the Nusselt number and is plotted against temperature difference in Fig. 9.7. The
Nusselt number increases with temperature difference and is about eight times
higher in water than in air. The downward jump in the curve for water at a Rayleigh
number of approximately 12 is due to the change in power law coefficients,
attributed loosely to a transition from laminar to turbulent flow within the thermal
360 9 Nusselt Number Correlations

10,000,000,000,000

1,000,000,000,000

Turbulent: 100,000,000,000
C = 0.021 H2O(L)
m = 0.20 10,000,000,000

Rayleigh Number
1,000,000,000

100,000,000
Laminar:
C = 0.59 10,000,000 Air
m = 0.25
1,000,000

100,000

10,000

1,000
0 20 40 60 80 100
Tfluid - Tsurface(°C)

Fig. 9.6 Rayleigh number vs. temperature difference between fluid and surface. Property values
evaluated at 70  C

Fig. 9.7 Nusselt number 200

vs. temperature difference 180
between fluid and surface
using published constants 160
140
Nusselt Number

120
H2O(L)
100
80
60
40 Transition Air
20
0
0 20 40 60 80 100
Tfluid-Tsurface(oC)

boundary layer. Conceptually, the Nusselt number (and thus convection coefficient)
should monotonically increase with temperature difference. It is conceivable that a
discontinuous jump UP could occur at the onset of turbulence, owing to the
enhanced mixing. A transition to turbulence, in any event, is very sensitive to
specific geometries. The intent here is to merely generate a monotonically increas-
ing function for modeling purposes. There is no such jump in air because the
Rayleigh number is always below the transition value for the specific case consid-
ered here.
9.2 Nusselt Number Example Calculation 361

Table 9.2 Revised table of correlation parameters for natural

convection on vertical plates
Rayleigh number range Flow character C m
(10)4–4.545(10)9 Laminar 0.59 0.25
4.545(10)9–(10)13 Turbulent 0.021 0.40

Fig. 9.8 Revised Nusselt 200

number vs. temperature 180
difference with modified
transition point 160
140
Nusselt Number

120 H2O(L)
Transition
100
80
60
40 Air
20
0
0 20 40 60 80 100
Tfluid-Tsurface(oC)

To address that problem, the critical transition Rayleigh number can be redefined
so that the transition occurs continuously (with a discontinuous slope). To do so
formally, the critical Rayleigh number can be calculated by setting the Nusselt
number equal using the two cases (1 and 2):

C1 ðRacrit Þm1 ¼ C2 ðRacrit Þm2

Rearranging and solving for the critical Rayleigh number:

m m
1
 1
C2 1 2 0:021 0:250:4
Racrit ¼ ¼ ¼ 4:545ð10Þ9
C1 0:59

Table 9.2 and Fig. 9.8 show the revised table and plot of Nusselt number
vs. temperature difference in which case a mild discontinuous slope is observed
at the transition point, but not a discontinuous jump.

9.2.4 Convection Coefficients

Finally, the convection coefficients are plotted against temperature difference in

Fig. 9.9, which is a linear axes, but different scales for air and water, and the same
data plotted in Fig. 9.10, on semi-logy axes. This latter view more visually
emphasizes the difference in magnitude between air and water. The convection
362 9 Nusselt Number Correlations

Fig. 9.9 Convection 1400 14

coefficients as a function of

Water Convection Coefficient

temperature difference. Film 1200 12

Air Convection Coefficient

temperature is 70  C H2O(L)
1000 10

(W/m/K)
(W/m/K)
800 Air 8

600 6

400 4

200 2

0 0
0 20 40 60 80 100
Tfluid - Tsurface (°C)

Fig. 9.10 Convection 10000

coefficient vs. temperature
Convection Coefficient (W/m2/K)

difference for water and air

H2O(L)
1000

100

Air
10

1
0 10 20 30 40 50 60 70 80 90 100
Tfluid - Tsurface (°C)

coefficients in liquid water are more than 100 times those of air. The Nusselt
number is eight times bigger, and the thermal conductivity, the last step in the
calculation, is 20 times bigger. In a future chapter, the underlying physics behind
this difference is explored further. This case study adds credence to the rule of
thumb case, except the water goes a bit above the value of 1,000. It is pretty hard to
maintain that temperature difference with water, however. Also, “Rules of Thumb”
are made to be broken.
For both fluids, the value of the convection coefficients drops steeply as the
temperature difference is reduced to zero. This effect is generally not observed for
forced convection situations. For natural convection, it is the temperature difference
which gives rise to a density difference which gives rise to a buoyant flow. The flow
in a forced convection does not require a temperature difference to drive it. This
effect is one reason that radiation (and evaporation, when relevant), becomes an
important, if not dominant, mode of heat transfer as an object cools (or warms) to an
ambient environment.
9.2 Nusselt Number Example Calculation 363

9.2.5 Natural Convection Film Property

A final step in this process is to develop a simple working formula that combines all
the fluid property values into a single term, except for the vertical height H and the
temperature difference. First, the definition of the Nusselt number is expressed in
terms of the convection coefficient:

kf NuH kf C
h¼ ¼ ðRaÞm
H H
Next, the Rayleigh number is expressed in terms of the primary property values:
 
β β
Ra ¼ gH 3 ΔT ¼   gH 3 ΔT
να μ k
ρ ρcp

Rearranging:
 
ρ2 c p β
Ra ¼ gH3 ΔT
μk

Inserting this form of the Rayleigh number into the convection coefficient correla-
tion and grouping property terms into brackets:
" #
k1m ρ2m cpm βm ΔT m
h ¼ Cg m
μm H 13m

The term in brackets is a fluid property (with very strange units) to be evaluated at
the film temperature, and will be referred to as the “film property” for natural
convection on a vertical plate. A similar approach can be taken for other
correlations in other situations. Figure 9.11 shows the value of this property (with
the factor Cgm included) as a function of film temperature (not the difference
between fluid and surface) for the laminar range (C ¼ 0.59, m ¼ 0.25), which covers
the problem in question. The water film property is more sensitive to temperature
than the air property.
Finally, a set of working formulas, strictly in SI units, that was used in the mug of
coffee problem (n-node transient case) are given by:

 
0:25
 
T outerSurface  T ambient

hair ¼ 1:5646  0:0015T film

H 0:25


T 
0:25

coffee  T innerSurface
hcoffee ¼ 32:38T film
0:441
H 0:25
To compare the relative effects of air and water further, it is useful to take
the ratio of the convection coefficients, and look at all the individual factors that
364 9 Nusselt Number Correlations

300 1 3

250
hnat L0.25
ΔT 0.25
= 0.59g0.25 [ k3r2cp b
m
] 4

H2O(L) 2.5
Liquid Water (W/m1.75/K1.25)

y = 32.381x0.441

AIR (W/m1.75/K1.25)
200 2

Air
150 1.5
y = -0.0015x + 1.5646
100 1

50 0.5

0 0
0 20 40 60 80 100
Film Temperature (∞C)

Fig. 9.11 Natural convection on a vertical plate film property vs. film temperature. Laminar flow
restriction

make it up. Entering values at 70  C from Figs. 9.2 and 9.3, keeping the individual
effects separate:
 0:75  0:5  0:25  0:25
kwater ρwater cp, water βwater
hH2O kair ρair cp, air βair
¼  0:25
hair μwater
μair

hH2O ð22:8Þ0:75 ð946Þ0:5 ð4:15Þ0:25 ð0:20Þ0:25

¼
hair ð19:6Þ0:25

hH2O ð10:4Þð30:7Þð1:43Þð0:67Þ
¼
hair ð2:1Þ
hH2O
¼ 145
hair
The natural convection coefficient of water is 145 times as large as that in air,
provided the same temperature difference is maintained (not so when the coffee
mug is revisited) and the wall height is the same. This analysis reveals that the
density difference has the largest effect (factor of 30), followed by the thermal
conductivity (k). The specific heat has a mild effect (1.43), and the thermal
expansion and viscosity favor the air. Further discussion of the underlying physics
behind these effects is deferred until convection coefficients models based on
governing laws are considered, rather than these empirical correlations.
9.3 Application to the Mug of Coffee 365

The combined fluid property effects for natural convection cases (and a similar
analysis could be conducted for forced convection cases) exhibits a strong temper-
ature dependency for liquid water, and a mild temperature dependency for air.
Therefore, the general practice of treating the property values as constants is shown
to be well justified for air (and could be extended to most gases), but care needs to
be taken when doing so with liquid water (or other liquids). For steady-state
applications, it seems reasonable to select a suitable film temperature. However,
for transient cases, recalculation of the convection coefficient as time proceeds may
be needed to obtain precise simulations.

9.3 Application to the Mug of Coffee

It is apparent that in the coffee/mug problem, there really is no such thing as “A”
convection coefficient because the temperature changes with time, and the natural
convection coefficients depend on the changing temperature differences between
the fluid and the corresponding wall. The convection coefficient rather is changing
with time, as the temperature difference changes. In addition, there are spatial
variations not addressed here. The Nusselt correlation used here give spatially
averaged convection coefficients, that is, averaged over the length of the vertical
plate. In reality, the convection coefficient is high near the leading edge of the flow
(near the top surface on the coffee side, near the bottom on the air side), and
decreases along the plate in the direction of induced flow as the boundary layer
thickens, reducing the local temperature gradient driving the heat.
To further complicate matters, the Nusselt number calculations are based on
steady-state experiments in which, for example, a vertical heated plate is
MAINTAINED at a surface temperature and investigated. The thermal boundary
layer has a thermal response time, and error is introduced if the fluid and/or surface
temperature is changing with time rapidly compared to that response time. It is
conceivable that this time constant can be estimated using methods developed in
this text, once a way to estimate the thickness of the boundary layer is developed.
A quasi-steady equilibrium is therefore implicitly assumed in the use of these
correlations to the coffee/mug problem.
On the other hand, there is no problem thinking about “A” coffee temperature,
even though it is now known that the coffee in the center is at a different tempera-
ture than the coffee right next to the mug. For modeling purposes, the choice of a
single value for convection coefficient simplifies the model implementation.

9.3.1 Value Selection for Constant Convection Coefficient Models

In the 2-node model of Chap. 4, a single convection coefficient was chosen for both
the coffee/inner mug surface and air/outer surface, even though it is clear that they
are both changing with time. So what value should be chosen? It seems logical to
select a temperature difference that exists during the period of time of most interest
366 9 Nusselt Number Correlations

Fig. 9.12 Thermal Rrad,ma

resistance network for the
side channel of a 1-node Tinner Touter
lumped model of the mug of Tcoffee T∞
coffee problem Rcm Rwall Rma

to the objective of the calculation. For example, if the objective is to calculate the
time it takes for the coffee to cool from an initial temperature (say 82  C) to a
defined drinking temperature (say 70  C), then it makes sense conceptually to
determine convection coefficient values when the coffee is at 76  C, the average
coffee temperature during the event. That convection coefficient will underestimate
its value when the coffee is above it, and overestimate it when the coffee is below,
and a degree of cancellation of errors will occur. On the other hand, if the time to
cool from a drinking temperature to a tepid temperature (say 50  C), then 60  C
would be the average coffee temperature during the event.
Figure 9.12 shows a quasi-steady state resistance network for the side channel
for the 1-node lumped capacity analysis (there is no capacitance attributed to the
mug). The coffee temperature is treated as if it were being maintained at a fixed
temperature, even though in the broader model, it gradually changes with time. At
the outer wall, convection (Rma) is modeled as acting in parallel with radiation
(Rrad.ma).
At any time in the simulation, the difference between the average temperature of
the coffee/mug system and ambient is well characterized (say 45  C, when ambient
is 25  C and the coffee is 70  C) . However, the Nusselt number correlation requires
the temperature difference between the fluid and the solid surface adjacent to it, not
the overall temperature drop from one fluid (coffee) to the other (ambient air). The
1-node lumped model allows the temperature of the two surfaces to be calculated,
but there is a circular argument. The surface temperatures depend on the convection
coefficients which depend on the surface temperatures. What can be done?
Develop an iterative scheme.
That is, make an intelligent guess (or a wild stab, if necessary) of both surface
temperatures. Use the Nusselt number correlation to determine a tentative convec-
tion coefficient based on those guesses. Calculate new inner and outer surface
temperatures based on these convection coefficients. Since these new temperatures
are different than the first guess (i.e., you guessed wrong), the convection
coefficients will be different from the first guess. But now the convection
coefficients can be recalculated based on these updated values. An iterative proce-
dure like this can often be done by hand, or it can be automated in a computer
program.
While there are actually different iterative algorithms that could be
implemented, the Method of Successive Substitution (MOSS), introduced in
Chap. 7, works well with most 2- and 3-dimensional steady-state heat transfer
problems, as already shown for the case where the resistance values are constant.
9.3 Application to the Mug of Coffee 367

The two wall surface temperatures in Fig. 9.12, Tinner and Touter are unknown.
Expressing the nodal energy balances on the inner and outer surfaces in the form
suitable for MOSS (using the Gauss–Seidel scheme where updates are used as soon
as available) yields the two recursion relations:
ðk Þ
T coffee T
ðk Þ
þ outer
ðkþ1Þ R Rwall
T inner ¼ cm NODE AT INNER WALL
1 1
þ
RðcmkÞ Rwall
ðkþ1Þ
T inner
þ ðkTÞ 1
ðkþ1Þ Rwall R
T outer ¼ ma, equiv
NODE AT OUTER WALL
1 1
þ
Rwall RðkÞ
ma, equiv

Superscripts (k + 1) and (k) have been applied to all unknowns that refer to the
iteration number. A subscript (k) means the “guessed” values, and (k + 1) refers to
the next guess. It is easy to confuse these iteration numbers with the “p” and “p + 1”
that refer to present and future in a numerical simulation of a transient behavior, but
they are fundamentally different things. A simulation step is a time prediction. An
iteration is a guess.
The resistor values are evaluated using the following formulas. Across the mug
wall:
w
Rwall ¼
kmug ðAinner þ Aouter Þ=2

where w is the mug thickness, kmug its thermal conductivity, Ainner ¼ πDH is the
inner surface area, and Aouter ¼ π(D + 2w)Ho is the outer surface area. This resis-
tance is a constant.
The inner convective resistance is:
1
Rcm ¼
hcoffee Ainner

where hcoffee depends on temperature as described previously.

The outer convective resistance is:
1
Rma ¼
hair Aouter

where hair depends on temperature as described previously.

368 9 Nusselt Number Correlations

The radiative coefficient is given by:

1
Rma, rad ¼
hrad Aouter
where the radiative coefficient is given by Kirchhoff’s Law (Chap. 3) as:
h i
hrad ¼ εσ ðT outer þ 273Þ2 þ ðT 1 þ 273Þ2 ðT outer þ 273 þ T 1 þ 273Þ

where ε is the emissivity of the surface and σ ¼ 5.669(10)8 W/m2/K4 is the Stefan–
Boltzmann constant. The temperatures are input in Celsius, and converted to Kelvin
in this equation, which must be in absolute units.
The equivalent resistance for the parallel convective/radiative channel is:

1
Rma, equiv ¼ 1
Rma þ Rma1, rad

Since the area is the same for the convection and radiation, the equivalent resistance
can be expressed as:

1
Rma, equiv ¼
ðhma þ hma, rad ÞAouter

Figure 9.13 shows a spreadsheet implementation of the MOSS for this problem.
Final converged results are shown in the top section. Fixed parameters are defined
below that. These cells refer to input parameters from a separate sheet (not shown).
The wall resistance and surface areas are from the mug of coffee problem. The outer
area is larger than the inner area because its radius is larger, and its height is higher.
The emissivity is a typical value for ceramic. The effect of radiation can be simply
tested by setting the emissivity to zero.
In the iteration table, the initial row (iteration 0) contains the initial guesses (set
at 50  C) for the inner and outer wall temperatures. On the same row, the film
temperatures (average of fluid and adjacent wall surface) are calculated, followed
by the inner convection, outer convection (from the Nusselt number correlation),
and radiative coefficient, and finally the thermal resistances. These formulas are
written once and copied down. On the iteration 2 row (k + 1), the recursion formulas
are entered for Tinner and Touter, which refer to the resistance values from the
previous row (k) or from the same row (k + 1), depending on the order the
calculations are done. These formulas are copied down. Ten iterations are shown
for this case, which is sufficient for successive values to converge to within 0.01  C.
Formal convergence of the method can be tested since the numbers continue to
change at a significant figure level less than shown.
The first “guess” of 50  C for the surface temperatures yielded a convection
coefficient of 744.5 W/m2/K on the inner (coffee) surface, and 6.028 W/m2/K on
the outer (air) surface. The radiative coefficient on the outer is coincidentally very
Workshop 9.1. Wind Chill Effect 369

RESULT
Rside= 2.63 oC/W
Uside= 12.56 oC/W

Fixed Parameters
emissivity = 0.9 Rrad,ma
Tcoffee (oC) = 70
Tinner Touter
Tamb (oC) = 25 Tcoffee T∞
Rwall (oC/W) = 0.094
Rcm Rwall Rma
Ainner (m2)= 0.0198
Aouter (m2)= 0.0303

Tfilm_outer(oC)
Tfilm_inner(oC)
Touter (oC)
Tinner (oC)

Rma,equiv
iteration

hrad,ma
hma

Rcm
hcm
0 50.00 50.00 60.00 37.50 744.5 6.028 6.119 0.0443 2.717
1 63.58 62.29 66.79 43.64 587.7 6.621 6.502 0.0562 2.515
2 67.11 65.60 68.55 45.30 486.9 6.752 6.608 0.0678 2.470
3 68.15 66.58 69.08 45.79 436.8 6.789 6.640 0.0755 2.457
4 68.47 66.88 69.24 45.94 417.0 6.800 6.650 0.0791 2.454
5 68.57 66.97 69.28 45.98 410.3 6.804 6.653 0.0804 2.452
6 68.60 67.00 69.30 46.00 408.2 6.805 6.653 0.0808 2.452
7 68.61 67.00 69.30 46.00 407.5 6.805 6.654 0.0810 2.452
8 68.61 67.01 69.31 46.00 407.3 6.805 6.654 0.0810 2.452
9 68.61 67.01 69.31 46.00 407.3 6.805 6.654 0.0810 2.452
10 68.61 67.01 69.31 46.00 407.3 6.805 6.654 0.0810 2.452

Fig. 9.13 Iteration table using the mug of coffee problem using the Gauss–Seidel method

close to the convection value. Based on these values, the recursion relations yield
improved “guesses” of 63.58 and 62.29  C for the inner and outer wall surface
temperatures, respectively. The film temperature, convection and radiation
coefficient, and thermal resistances are all recalculated, and the iteration scheme
is repeated for as many rows as desired. It is best to confirm convergence rather than
fixing the number of iterations and “hoping” it worked. This method is not
guaranteed to converge, and chaotic or divergent iteration “trajectories” are con-
ceivable. In that case, it is possible to isolate the unknown temperature in a different
way (i.e., isolate the temperature in the radiation coefficient).

Workshop 9.1. Wind Chill Effect

The temperature today is 20  C, but with the wind it feels like 5. What’s up with
that?
Physiologically, the sensation of cold is related more to the rate of heat loss
than directly to the temperature of the fluid that the skin is exposed to. The wind
chill effect (shown schematically) can be modeled by equating the actual rate of
heat loss from a person’s skin on a cold windy day (left sketch, with air
370 9 Nusselt Number Correlations

temperature T1 and actual convection coefficient hactual) to the heat loss on a

hypothetical colder day with no wind (right sketch, with hypothetical air temper-
ature T1,WindChill ¼ (T1  ΔTWindChill) and a hypothetical convection coefficient
hhypothetical that would exist under those conditions). Expressing the heat transfer
rate with an overall heat transfer coefficient U:

q ¼ U actual AðT skin  T 1 Þ ¼ U hypothetical AðT skin  T 1, Wind Chill Þ

Note that while the convection coefficient is the primary variable that changes with
the wind, the quantitative determination of wind chill effect involves several factors
and depends on whether exposed skin or clothed skin is to be compared. Radiation
acts in parallel with convection, there are conduction effects from the interior (Biot
number effects). Finally, physiological thermoregulatory responses (i.e., constric-
tion of blood vessels) are important. A good student project is to consider these
effects one by one. The main goal of this workshop is to gain experience in
estimating values of the convective heat transfer coefficient using Nusselt number
correlations. The effect of wind speed, size of the object (extended finger vs. arm),
and fluid type (air vs. water) will be considered.

Problem Statement
• Write a Worksheet or program code that takes as input parameters (in SI units) D
(outer diameter), wind speed V, and fluid properties (k, ρ, cp, μ) and uses the
Hilpert/Knudsen/Katz correlation to return the convection coefficient for forced
convection across a cylinder.
• Write a Worksheet or program code that takes as input parameters (in SI units) D
(outer diameter), surface temperature (Tskin), fluid temperature (T1), and fluid
properties (β, k, ρ, cp, μ) and uses the “Notsurewho” correlation to return the
convection coefficient for natural convection across a horizontal cylinder.
• Use this code to complete the Results table, and notice both the values of
convection coefficient, and the trends associated with size, wind speed, and
fluid type.

Forced Convection, External Flow

Convection coefficient experimental data for heat exchange between a cylinder
(outer diameter D) and a fluid (with properties k, ρ, cp, μ and derived properties
Workshop 9.1. Wind Chill Effect 371

 
ν ¼ μ=ρ, α ¼ k= ρcp ) with a cross-flow velocity (V ) have been correlated using
the Hilpert/Knudsen/Katz equation:

NuD ¼ CReDn Pr1=3


hD ρVD n  ν 1=3
¼C
k μ α

In selecting fluid properties as constants evaluated at the film temperature. The

constants C and n depend on the value of the Reynolds number (ReD) given in the
following table, modified from the published sources to give continuous functions
between ranges:

Reynolds number range C n

0.4–4 0.989 0.330
4–35 0.911 0.385
35–4,083 0.683 0.466
4,083–40,045 0.193 0.618
40,045–400,000 0.0266 0.805

Natural Convection, External Flow

Convection coefficient experimental data for heat exchange between a horizontal
cylinder (diameter D, with outer surface temperature Ts) in a still fluid at tempera-
 
ture T1 (with properties β, k, ρ, cp, μ and derived properties ν ¼ μ=ρ, α ¼ k= ρcp )
have been correlated by various authors (the “Notsurewho” correlation):

NuD ¼ C½Gr D PrN


 N
hD gβD3 jðT s  T 1 Þj  ν
¼C
k ν2 α

The constants C and N depend on the value of the Raleigh number (GrDPr).

Rayleigh (GrDPr) number range C N

104–2.12(10)7 0.53 0.25
2.12(10)7–1012 0.13 0.3333
372 9 Nusselt Number Correlations

Note that convection coefficient depends on the absolute value of temperature

difference.
 The coefficient
 of thermal expansion (β) for ideal gases is given by
β ¼ 1= T film þ 273 , where the temperature is in Celsius.
Assume the exposed skin temperature is 32  C (for natural convection cases).
For the diameters, use D ¼ 0.02 m for a finger, D ¼ 0.080 m for a forearm.
Use fluid properties evaluated at nominal room temperature (20  C) from this
table:

Workshop 9.2. Forced Convection for Internal Flows

There are many applications that involve the exchange of heat between a fluid
flowing in pipes (or ducts) and the inner walls of the pipe. For engineering
calculations, the rate of heat transfer (q) is determined using Newton’s Law of
Cooling, that is, q ¼ hðπDLÞðT wall  T bulk Þ, where D is the inner diameter, L is an
appropriate length of pipe (which could be a differential length, Δx), Twalls is the
temperature of the inner wall surface, and Tbulk is the average fluid temperature
over the length in question. The value of the convection coefficient (h) depends on
both the fluid type (in particular its properties such as thermal conductivity (k),
Workshop 9.2. Forced Convection for Internal Flows 373

density (ρ), specific heat (cp), and viscosity (μ)) and on the character of the flow (the
“wind chill” effect), and cannot be simply looked up in a Table. The goal of this
workshop is to gain experience in using published Nusselt number correlations for
internal, forced convection flow problems.
While many correlations have been published, the Dittus/Boelter equation for
turbulent flow is arguably the most general, and would be a good starting place even
if higher precision is warranted afterward. It is given by:

NuD ¼ 0:023R0:8 n
e, D P r

However, the Nusselt number cannot be less than a value of 3.66, the value obtained
for laminar flow. The exponent “n” equals 0.4 for heating (walls hotter than fluid),
and 0.3 for cooling (walls colder than fluid). Expressing the three dimensionless
variables directly in terms of system parameters:


hD ρVD 0:8  ν n
¼ 0:023
k μ α

where V is the flow velocity and fluid properties (k, μ, ν ¼ μ=ρ, α ¼ k=ρcp ) are
evaluated at the “film temperature,” taken to be the average of the bulk fluid and
wall temperatures. Also, this and even more sophisticated correlations yield
ESTIMATES of the convection coefficient, and this uncertainty must be borne in
mind in any application.

Pressure Drop
While thermodynamic analyses often neglect it, there is always a pressure drop in
internal flows that can be estimated using the Fanning friction factor, f, a nonlinear
function of Reynolds number and the relative roughness
 L 1 of2the
 pipe. The pressure
drop across a pipe of length L is given by: ΔP ¼ f D 2ρV The pressure drop is
not considered further here, except that it should be borne in mind that factors that
give rise to high convection coefficients (high velocity, small diameter) tend to also
give high pressure drops and therefore higher pumping costs.

Case Study
In a boiler application, liquid water enters a pipe of inner diameter (use D ¼ 0.02 m
as a default) whose inner walls are maintained at a higher temperature (a “heating”
event). At the exit of a preheating section, the fluid is a saturated liquid (x ¼ 0), and
the value of the convection coefficient at that point will be calculated. The fluid
continues to flow and phase change occurs in a boiler section until it is a saturated
vapor (x ¼ 1), and the value of the convection coefficient at that point will be
calculated. The flow continues through a superheating section and exits as a
superheated vapor. In a later chapter, the lengths of the heater sections will be
calculated, for which appropriate convection coefficient values are needed.
374 9 Nusselt Number Correlations

Hot Walls

Preheater Section Boiler Section Superheater Section

Properties of saturated Liquid Vapor Unit L/V ratio

H2O @ 1 atm
k 0.646 0.028 W/m/K 23.1
ρ 989 0.518 kg/m3 1,907.5
cp 4,175 2,000 J/kg/K 2.1
μ 5.50E04 1.40E05 kg/m/s 39.3

In a condenser application, the above case is reversed. The fluid enters as a

superheated vapor and exits as a compressed liquid. In this case, the inner walls are
maintained at a lower temperature (a “cooling” event). This hypothetical case
neglects pressure effects so that Nusselt number correlations can be used to
compare convection coefficients of the same fluid type but in different phases.
Problem Statement
• Write a spreadsheet or computer code that takes as input parameters D (tube
inner diameter), V (average flow velocity), and fluid properties (k, ρ, cp, μ) and
uses the Dittus/Boelter correlation to return the convection coefficient. Use an
“if-then-else” statement to determine whether the flow is laminar or turbulent
(do not use a transition Reynolds number, but rather ensure the Nusselt number
exceeds the laminar flow value of 3.66).
• Write a spreadsheet or computer code that compares convection coefficients for
the case study at the entrance and exit of the boiler section (properties given
below) over a wide range of mass flow rates, spanning both laminar and highly
turbulent flows. Prepare a plot of h vs. mass flow rate for saturated H2O liquid
and vapor at 1 atm with a diameter of 0.02 m (property values given). Put all
four curves (heating and cooling for liquid and gas) on the same well-formatted
log–log axes.

Workshop 9.3. Two-Node Mug of Coffee with Variable

Convection Coefficients

Modify the 2-node mug of coffee analysis (resistance network repeated here) to
include the effects of variable convection and radiation coefficients. As the mug and
its inner and outer surface temperatures change, so do the convection coefficients.
Use the Nusselt number correlations developed in this chapter (formulas of Fig. 9.9)
to recalculate the convection coefficients at the mug wall at each step. For the
Workshop 9.3. Two-Node Mug of Coffee with Variable Convection Coefficients 375

coffee free surface, use the Nusselt number correlation for a heated horizontal plate
of diameter D (from published or online source).
The simpler way to include these effects for this case study is to use the average
mug temperature to determine the difference between the fluid and solid tempera-
ture, which makes the simulation an explicit one. That is, a closed-form expression
for the future temperature is developed. This method is completely justified after
the mug heating phase for this case because there is very little spatial variation
across the mug wall (a low Biot number concept).

T∞
evaporation

convection
radiation

Rsa
Tsurface
mug
coffee Rcs T2 T∞
Tinside Toutside
T1 convection
radiation
Rcm Rmug Rmug
2 2 Rma,side
C1 C2
Rma,floor

However, during the mug heating phase, the mug surface temperature is sub-
stantially different than the average mug temperature (particularly the inside tem-
perature), and this effect can be accounted for by conducting an iterative scheme at
each time step to determine the surface temperature and convection coefficients.
This is a form of implicit formulation, which is not easily performed in a
spreadsheet.
In addition to the coffee and mug temperatures, plot the convective and radiative
coefficients vs. time, and consider the trends. Then, plot the dimensionless temper-
ature vs. time on semi-log x axes to investigate the overall first-order behavior. A
pure first-order behavior would result in a straight line. However, since the convec-
tion and radiative coefficients change, there will be some curvature. Define dimen-
sionless temperature as:
T coffee T 1 Ccoffee T coffee, initial þ Cmug T mug, initial
θ ¼ T LUMPED T 1 , where T LUMPED ¼ Ccoffee þ Cmug is the adiabatic
temperature after the mug heating phase.
Convection Fundamentals
10

To this point, heat transfer modeling methods have been developed for closed
thermodynamic systems, that is, for systems for which heat and work cross the
system boundary, but mass does not. For applications where fluids (liquids, gases,
slurries, etc.) flow into and/or out of the defined system, the mass that crosses the
system boundary carries its energy (internal, kinetic, and potential) with it. In
addition, a force (pressure times cross-sectional area) acting through a distance
per unit time (velocity) is required to push fluid in or out, called “flow work.” The
term “flow power” is technically more accurate, but that would often be interpreted
as the total head times the mass flow rate, a very different thing. The general mass
balance for an open thermodynamic system with multiple inlets and outlets can be
expressed as:

dMcv X X
¼ m_  m_
dt inlets outlets

where Mcv is the total mass inside the control volume (a triple integral), m_ ¼ ρVA is
the mass flow rate of a stream which could be an inlet or an outlet (a double integral,
with density and velocity being appropriately weighted averages). At steady-state,
the mass in the control volume remains constant, and the net mass flow rate is
balanced by the net mass flow rate out. That is:
X X
m_ ¼ m_ ðat steady stateÞ
inlets outlets

For a pipe or duct with a single inlet and single outlet, the mass flow rate of the inlet
equals that of the outlet. The volumetric flow rate (VA) is only the same if the
density is the same.

# Springer International Publishing Switzerland 2015 377

G. Sidebotham, Heat Transfer Modeling: An Inductive Approach,
DOI 10.1007/978-3-319-14514-3_10
378 10 Convection Fundamentals

The general energy balance for an open thermodynamic system with multiple
inlets and outlets can be expressed as:

dEcv X  P
 X 
P

¼ Q_ in  W_ shaft, out þ m_ e þ  m_ e þ
dt inlets
ρ outlets
ρ

where ððð ððð

   
2
Ecv ¼ ρedV ¼ ρ u þ V 2 =2 þ gz dV ¼ Mcv u þ V =2 þ gz is the total
CV CV
energy stored within the control volume,

u is the internal energy per unit mass

V2/2 is the kinetic energy of the fluid per unit mass
g is the potential energy relative to a reference height per unit mass
Q_ in is the rate of heat transfer crossing the boundary of the control volume
Ẇshaft,out is the net rate of shaft work (all non-flow work, including electric)
crossing the boundary
ṁP/ρ is the rate of flow work associated with the inlet and exit streams

All cases considered in this text will be at steady-state (the rate of energy stored
remains constant), there will be no mechanical devices exchanging work with the
environment, kinetic and potential energy changes will be considered to be negli-
gible. With these restrictions, the energy balance can be expressed in the form of
Energy In ¼ Energy Out:

X X
Q_ in þ _ ¼
mℏ _ þ W_ shaft, out ðat steady stateÞ
mℏ
inlets outlets

where ℏ ¼ u þ P=ρ is the enthalpy of the streams as they cross the boundary. The
enthalpy is a thermodynamic property that combines the internal energy carried in
or out with the work required to push fluid in or out. A line is drawn through it to
differentiate it from the convection coefficient, a fundamentally different quantity.

10.1 Boundary Layer Concepts

Convection coefficients are used as a boundary condition in transient and steady-

state conduction problems. They are estimated in practice using Nusselt number
correlations, a cookbook approach outlined in the previous chapter. An appreciation
for the underlying physics of Newton’s Law of Cooling requires a closer look at the
fluid in the immediate vicinity of a solid object at a different temperature. In this
chapter, a 1-node analysis of the flow across a thin flat plate is developed that
reveals the underlying physics of convection.
10.1 Boundary Layer Concepts 379

U¥ U¥

Freestream

ne
streamli Momentum
boundary d(L)
layer
x
d(x)

Fig. 10.1 Schematic of a momentum boundary layer on a flat plate placed in a uniform stream

The concept of a boundary layer is central to understanding convection

coefficients. A boundary layer is a layer of fluid immediately adjacent to a solid
that is generally thin compared to a characteristic dimension of the solid object.
Outside the boundary layer, the fluid has the properties of the free stream (temper-
ature and flow velocity). At the solid surface, the fluid properties are those of the
solid. In between, there are large spatial gradients in velocity and temperature that
occur across the boundary layer. For mass transfer applications, such as evapora-
tion, there are gradients in chemical composition as well.
In heat transfer, there are two boundary layers that develop simultaneously: a
momentum boundary layer and a thermal boundary layer. The momentum bound-
ary layer is often called hydrodynamic for historical reasons in which liquid water
was investigated, but momentum boundary layers occur for any fluid (gas or liquid).
The retarding effects of viscosity occur within the momentum boundary layer.
In fluid mechanics, the forces exerted by a fluid on an object are of primary interest
and these are determined by the momentum boundary layer. Figure 10.1 shows
a schematic of a momentum boundary layer on both sides of a thin plate placed in a
velocity field of uniform flow velocity (U1). The plate has a streamwise length L
and a width w into the page (not shown) and therefore a surface area Lw. The heavy
solid line is the solid plate. At the inlet plane (at x ¼ 0), a uniform velocity profile is
shown. At the exit plane (at x ¼ L), there is a velocity profile; the velocity is zero at
380 10 Convection Fundamentals

the solid surface (the no slip condition at y ¼ 0) and increases to the free stream
velocity at the edge of the boundary layer (at y ¼ δ(L)), with a zero gradient.
The lighter curved lines above and below the plate divide the free stream from
the boundary layer. In the free stream, the flow velocity in the x-direction equals the
free stream velocity. There is, however, a mild cross-stream flow velocity (in the
y-direction) as a consequence of mass conservation.
The dotted line shown on the top surface is a streamline that is coincident with
the edge of the boundary layer at the end of the plate. This streamline defines a
streamtube that has a constant mass flow rate at every cross-section. Since
the average velocity exiting the exit plane is less than the free stream velocity,
the area of the streamtube is larger than at the entrance. On the other hand, the line
that shows the boundary layer is not a streamline, and mass flows into the
boundary layer.
A velocity profile near the midpoint is shown underneath the plate. Here, the
boundary layer is thinner than at the exit, that is, δ(x) < δ(L). Therefore, the boundary
layer is a function of distance and generally increases with x. The velocity depends
on both x and y. The velocity gradient at the wall is higher at the midpoint of the
plate (x ¼ L/2, say), and therefore the local shear stresses on the plate are highest
near the leading edge and decrease with x.
In heat transfer, heat exchange between the solid object and the free stream
occurs across a thermal boundary layer. However, the thermal boundary layer
is influenced by the momentum boundary layer, so an understanding of both is
needed. Figure 10.2 shows a schematic of a thermal boundary layer growing on one
side of a flat plate fixed at a temperature Tw, with a uniform flow (not shown) of
temperature T1 approaching from the left. In this case, the line separating the
boundary layer from the free stream has a similar shape to the momentum boundary
layer, but is not generally coincident with it. Outside the thermal boundary layer,
the fluid temperature equals the free stream temperature. Temperature profiles are
shown at the trailing edge (x ¼ L) and at a midpoint. The temperature of the fluid
gradually changes from its value at the wall (at y ¼ 0), to the free stream value at the
edge of the boundary layer (at δT), with zero slope. The shape of the temperature
profile is similar, but the boundary layer is thicker at the exit than at the midpoint.
The horizontal dashed line in this case represents the boundary of a control volume
that will be used to analyze the thermal boundary layer. Notice that fluid crosses this
control surface with a mass flow rate equal to the difference between the mass
that flows into the control volume at the leading edge and the mass that exits at the
trailing edge. Also the temperature of the fluid that exits the top surface equals
the free stream value.
A powerful traditional method of analysis is the integral method, in which a
polynomial function is used to approximate the temperature and velocity profiles
across a boundary layer (y), and the x dependency is contained in the boundary
layer thickness. In practice, these functions are formally derived assuming a
polynomial fit to the velocity and temperature profiles that satisfy boundary
conditions at the wall and edge of the boundary layer. For a second-order fit,
three conditions are required to determine the coefficients of the function
f ¼ ay2 + by + c. These conditions are the no slip condition and fixed wall
10.1 Boundary Layer Concepts 381

T∞ T∞

dx
y δT @ L
δT @ x
TW x TW
dq

Fig. 10.2 Schematic of a thermal boundary layer on a flat plate in a uniform stream

temperature at the wall (y ¼ 0), and a fixed velocity and temperature with zero slope
at the edge of the boundary layer (x ¼ δ). The results are:
"  2 #
y y
uðx;yÞ ¼ U1 2 
δðxÞ δðxÞ
"  2 #
y y
T ðx;yÞ ¼ T wall þ ðT wall  T 1 Þ 2 
δT ðxÞ δ T ð xÞ

This velocity profile is reasonable for the case of forced convection only, where
there is an imposed free stream velocity, and for laminar flow, a difficult statement
to explain briefly. For natural convection, a third-order polynomial is the minimum
order needed. However, a second-order polynomial fit to the temperature profile is
reasonable for both natural and forced convection environments, with laminar flow.
In what follows, the 1-node boundary analysis is equivalent to the use of a linear
velocity and temperature profiles. That is:

y
uðx;yÞ ¼ U 1
δðxÞ
y
T ðx;yÞ ¼ T wall þ ðT wall  T 1 Þ
δ T ð xÞ

Notice that the velocity and temperature within the boundary layer are functions of
both x and y. The streamwise (x) dependency is contained in the boundary layer
thickness (momentum and thermal), and the cross-stream (y) dependency is
expressed as a polynomial approximation.
382 10 Convection Fundamentals

10.1.1 Newton’s Law of Cooling

The rate of heat transfer from the solid into the fluid can be evaluated at the solid
surface (on the fluid side of that discontinuity in material property, with solid below
and fluid above) for the differential slice of fluid of thickness dx of Fig. 10.2
extending across the boundary layer. Then, Newton’s Law of Cooling can be
defined for the average heat transfer coefficient. That is, a general expression for
the local heat transfer coefficient is:

∂T 
dq ¼ k  wdx  hðxÞ wdxðT w  T 1 Þ
∂y y¼0

Rearranging for the local convection coefficient:



k∂T
∂y  y¼0
h ð xÞ ¼
ðT w  T 1 Þ

In first-order modeling, the temperature gradient at the wall (at an arbitrary distance
from the leading edge, x) is considered to be proportional to the total change in
temperature divided by the local thermal boundary layer thickness:

∂T  ðT 1  T w Þ

∂y y¼0 δ T ð xÞ

This slope is what would be obtained for a linear temperature profile. Therefore, the
local convection coefficient is:

k
hðxÞ 
δ T ð xÞ

The total heat transfer rate and average convection coefficient are determined by
integrating the differential heat transfer rate across the length of the plate:

ðL 
∂T 
q¼ k  wdx  hLwðT w  T 1 Þ
∂y y¼0
x¼0

2 3
 ðL

∂T 
k 6
6 dx7
h¼ 6 ∂y y¼0 7 7
L 4x¼0 5
ðT w  T 1 Þ
10.1 Boundary Layer Concepts 383

This integration cannot be formally conducted until the variation of boundary

layer thickness with x is determined. However, in strict first-order modeling
(linear temperature profile), the average temperature gradient is taken to be the
overall temperature change divided by the geometric average of the boundary layer
thickness at the leading edge (δΤ(x¼0) ¼ 0) and at the trailing edge (δΤ(x¼L) ¼ δΤ).
That is:

ðL 
∂T  ðT wall  T 1 ÞL 2ðT wall  T 1 ÞL
 dx  1  ¼
∂y y¼0 2 0 þ δ T ð LÞ
δT ðLÞ
x¼0

That is, the average gradient over the length of the plate is twice that of the gradient at
the trailing edge. This result will be shown to be true after the boundary layer thickness
variation with x is approximated. The result, then, is that the average convection
coefficient on the plate equals twice its value at the trailing edge of the plate:

2k
h ¼ 2hðLÞ 
δT ð@LÞ

The averaged Nusselt number is a dimensionless measure of the convection coeffi-

cient defined by:

hL L
NuL ¼ 2
k δ T ð LÞ

The local Nusselt number is:

hx x
Nux ¼ 
k δ T ð xÞ

The Nusselt number is therefore shown to be a measure of the ratio of plate length to
boundary layer thickness. A thin boundary layer results in a large Nusselt number.
The convective thermal resistance is defined as:

1 δT
Rconvective ¼ 
hA kA

Convection is, at its heart, a conduction process, with a thermal resistance that
looks like the conductive resistance across a wall of thickness δ, and thermal
conductivity k. The boundary layer acts as an insulating barrier to convection.
The thicker the layer, the lower the heat transfer rate. As will be developed,
the thermal boundary layer thickness also depends on thermal conductivity,
so the temptation to say that the convection coefficient is proportional to the
fluid thermal conductivity must be suppressed. The temperature gradient is
steepest near the leading edge, and the local convection coefficient is high. The
temperature gradient, and thus local convection coefficient, decreases with
384 10 Convection Fundamentals

U∞ U∞

tum
momen
thermal δ
δΤ
x
L

Fig. 10.3 Schematic of thermal and momentum boundary layers when the momentum boundary
layer is thicker than the thermal boundary layer

U∞ U∞
al
therm

δΤ
tum
momen
δ

x
L

Fig. 10.4 Schematic of thermal and momentum boundary layers when the thermal boundary
layer is thicker than the momentum boundary layer

distance from the leading edge. This functional dependence will be obtained in a
simplified 1-node analysis of the thermal boundary layer.
The relative thickness of the two boundary layers will be shown to be related to
the ratio between the fluid kinematic viscosity (ν ¼ μ/ρ) and its thermal diffusivity
(α ¼ k/ρ/c). When the kinematic viscosity is greater than the thermal diffusivity, the
momentum boundary layer is thicker, as shown schematically in Fig. 10.3. In this
case, the thermal boundary layer is contained entirely within the momentum
boundary layer, and the velocity at the exit plane (at x ¼ L) is everywhere below
the free stream velocity. In the opposite case, the thermal boundary layer is thicker
than the momentum boundary layer (Fig. 10.4). The exit velocity is uniform over a
portion of the exit plane.
10.2 Forced Convection Across an Isothermal Flat Plate 385

10.2 Forced Convection Across an Isothermal Flat Plate

The basic mechanism of convection is due to conduction across an insulated layer of

fluid. All that needs to be done to obtain the heat transfer rate, and therefore define a
convection coefficient, is to relate the thickness of that layer to physical properties.
Traditionally, theoretical solutions (the Blasius exact solution to the boundary layer
approximation equations, and integral approximate methods) for the case of laminar
flow over a limited number of solid shapes (the flat plate being the simplest) are
developed as the only cases for which theoretical solutions are possible. The approach
taken here is to look at the boundary layer as a single node, and obtain a closed-form
solution that predicts the exact same functional dependencies as these more detailed
theoretical models. The final result differs only in the proportionality constant. All
these methods fail to produce predictive models for turbulent flows, or for flows over
more complicated shapes, but a deeper appreciation of the underlying physics
associated with convection coefficients is gained.
An analysis of the momentum equation (F ¼ ma) is first applied to obtain an
expression for the momentum boundary layer thickness, and an expression for the
drag coefficient, a dimensionless measure of the drag force. The thermal boundary
layer is influenced by the momentum boundary layer, and these results are used as
inputs to the thermal model.

10.2.1 Momentum Boundary Layer

Consider the control volume (CV) defined by the streamtube of Fig. 10.5. The top
surface coincides with the streamline that divides fluid that enters the boundary
layer, and the free stream. The free stream is gently displaced upward, but does not
experience a retarding force. The left (east) face has an unknown height, yeast. All of
the fluid that crosses this plane enters the boundary layer and then exits the CV
through the right (west) face at the exit plane. The velocity varies from zero at the
wall to the free stream velocity at the edge of the boundary layer (at the exit plane).

y
U∞

U∞
line
stream
δ(x) Wall Shear Stress δ(L)
yeast

Fig. 10.5 Control volume (dashed line) used for analysis of momentum boundary layer
386 10 Convection Fundamentals

The bottom surface is coincident with the wall, where there is no flow velocity
(no slip), but a shear stress is exerted by the wall on the fluid.
Mass balance. A mass balance will yield a relationship between the height of the
control volume at the entrance (yeast) and the boundary layer thickness at the exit
(δ). Taking a steady-state mass balance, the mass flux entering the left (east, at
x ¼ 0) plane equals the mass flux exiting the right (west, at x ¼ L) face. Integrating
across these planes:
2 3 2 3
yð
east ðδ
6 7 6 7
4 ρuwdy5 ¼ 4 ρuwdy5
y¼0 y¼0
x¼0 x¼L

The velocity is uniform at the entrance (x ¼ 0), and the integral is easily expressed,
but in terms of the unknown height of the streamtube, yeast :
2 3
yð
east
6 7
4 ρuwdy5 ¼ ρU1 wyeast
y¼0
x¼0

An average velocity at the exit plane is used at the exit plane:

2 3
δðLÞ
ð
6 7 1
4 ρuwdy5 ¼ ρUwδðLÞ  ρwU 1 δðLÞ
2
y¼0
x¼L

Using the geometric average between the wall and free stream is equivalent to
assuming a linear velocity profile with a discontinuous slope at the edge of the
boundary layer. Putting the pieces together, the mass flux entering equals the mass
flux leaving:
1
ρU1 wyeast  ρwU 1 δðLÞ
2

The height of the control volume entering is therefore simply half of the boundary
layer thickness at the exit, that is yeast ¼ δðLÞ =2. The result for a second-order
velocity profile is yeast ¼ 2δ(L )/3, that is, it has the same functional dependency,
but a different coefficient.
x-Momentum balance. An integral force balance in the x direction applied to the
control volume states that the rate of change of momentum stored (which is zero at
steady-state) equals the momentum flux entering minus the momentum flux exiting
plus the sum of the forces acting on the control volume. The only forces acting on
this control volume is the retarding friction at the wall, expressed as an integral
along the length. For flows across objects other than thin flat plates, pressure forces
10.2 Forced Convection Across an Isothermal Flat Plate 387

in the x-direction would also be present. In natural convection, gravity forces also
come into play. The momentum balance is:
2 3 2 3
yð
east ðδ ðL
dðx  MomÞ 6 7 6 7
¼4 ρu wdy5
2
4 ρu wdy5
2
 τwall wdx ¼ 0@SS
dt
y¼0 y¼0 x¼0
x¼0 x¼L

The first term, the momentum flux entering is:

2 3
yð
east
6 7 δðLÞ
4 ρu2 wdy5 ¼ ρU 21 wyeast  ρU 21 w
2
y¼0
x¼0

The momentum flux exiting is:

2 3
ðδ
6 7 2 U 21
4 ρu2 wdy5 ¼ ρU wδðLÞ  ρ wδðLÞ
4
y¼0
x¼L

The shear stress term is expressed by defining an average shear stress:

ðL
τwall wdx  τwall Lw
x¼0

Putting the pieces together:

1 1
0 ¼ ρU 21 wδðLÞ  ρU 21 wδðLÞ  τwall Lw
2 4
1 2 δðLÞ
τwall ¼ ρU 1
4 L

This momentum balance relates the average shear stress to the boundary layer
thickness.
For Newtonian fluids, the shear stress exerted by a fluid is proportional to the
velocity gradient, with the proportionality being the dynamic viscosity, a fluid
property. That is, the local shear stress at the wall (y ¼ 0) is:

∂u
τwall ¼ μ 
∂y y¼0
388 10 Convection Fundamentals

Therefore, an additional relationship between average shear stress and boundary

layer thickness can be obtained by integrating the shear stress across the entire
surface of the plate:
ðL 
1 ∂u
τwall ¼ μ  dx
L ∂y y¼0
x¼0

The velocity gradient at the exit plane is:


∂u U1
  for a linear velocity profile
∂y x¼L, y¼0 δðLÞ

The average velocity gradient is expected to be higher than at the exit, but to exhibit
the same functional dependency (i.e., the overall change in velocity across the
boundary layer (U1) divided by the geometric average distance across the bound-
ary layer δ(L )). Note that the shear stress at the entrance plane is infinite even in the
exact Blasius solution. In reality, the Blasius solution is the exact solution to
the boundary layer approximation equations, so that is not an exact model. The
functional dependency of the average shear stress is therefore:

1 2  δ U1
τwall ¼ ρU 1  μ
4 L δ

Solving for the boundary layer thickness:

sffiffiffiffiffiffiffiffiffiffi rffiffiffiffiffiffiffiffi
μL νL
δ2 ¼2
ρU1 U1

The dimensionless boundary layer thickness is:

rffiffiffiffiffiffiffiffiffiffiffiffiffi
δ μ 2
2 ¼ pffiffiffiffiffiffiffiffi
L ρU 1 L ReL

This result has exactly the same functional dependence as more detailed models.
The only difference is the constant, 2, compared to the Blasius solution of 5.0.
This model underestimates the magnitude of the boundary layer thickness and is
consistent with the use of the lower bound for the shear stress on the plate
(i.e., the value at the edge of the plate). In addition, the shear stress using a
linear velocity profile underestimates that of a more detailed model. The main
goal here is to demonstrate the key functional dependencies associated with
boundary layers.
10.2 Forced Convection Across an Isothermal Flat Plate 389

10.2.1.1 Drag Coefficient

The forces acting on the plate are expressed conventionally in terms of a drag
coefficient, which (for streamlined objects like a flat plate) is the force in the
direction of flow normalized by a hypothetical force equal to the stagnation pressure
times the planform area (shadow projected perpendicular to direction of flow).
That is:

ðL
τwall wdx
1 2 δ ð LÞ
Fdrag ρU1 Lw 1 δ
L ðLÞ
¼4
x¼0
CDrag ¼ ¼ ¼
Pstagnation Aplanform 1 2 1 2 2 L
ρU Lw ρU Lw
2 1 2 1

1
CDrag  pffiffiffiffiffiffiffiffi
ReL

10.2.1.2 Turbulence: When the Model Breaks Down

Fluid friction (shear stress) is fundamentally due to a transport of momentum at
the molecular level, and the dynamic viscosity (μ) is the associated fluid property.
Within the boundary layer, if the flow is laminar, then fluid particles travel
smoothly along well-defined streamlines (coincident with pathlines) that are gently
bumped away from the wall (a consequence of mass conservation).
If the flow within the boundary layer becomes turbulent, then another mecha-
nism for the transport of momentum arises, namely, random fluctuations in the
velocity field (in all three spatial dimensions). Streamlines and pathlines are no
longer coincident nor well defined. The flow field is fundamentally transient and
three dimensional, although a steady-state averaged velocity field is generally
considered in analysis. The shear stress at the wall is no longer controlled by the
viscous mechanism, and the model breaks down. So do the analytical models.
Largely from empirical data, for flows across a flat plate, there is a critical
Reynolds number of 5(10)5 above which boundary layers become turbulent.

10.2.2 Thermal Boundary Layer

A different control volume is used to analyze the thermal boundary layer, shown
in Fig. 10.6. The top (north) surface is horizontal (not a streamline) so mass crosses
this surface, and it carries the enthalpy (internal energy plus flow work) associated
with it. A 1-node model for the boundary layer is shown in which the node represents
an average temperature within the control volume (taken to be the geometric average
T ¼ ðT 1 þ T wall Þ=2). Temperature-dependent current sources (rather than thermal
resistances) are used to represent energy flows that cross the surface of the CV
since these energy flows are not driven by a temperature difference. Fluid enters
390 10 Convection Fundamentals

Fig. 10.6 Schematic of control volume and thermal resistance network for the thermal boundary
layer on an isothermal plate

the left (east) face with a uniform velocity at the free stream temperature (T1). The
fluid that exits the top surface is also at the free stream temperature (T1).
The velocity and temperature of the fluid that exits the right (west) face vary from
their value at the wall to their free stream values. The bottom surface, the wall, is
maintained at a fixed temperature (by unspecified means), Twall.

10.2.2.1 Mass Balance

A steady-state mass balance applied to the thermal boundary layer yields:

dM
¼ m_ west  m_ north  m_ east ¼ 0@SS
dt

Therefore, the mass flowing out the top (north) face is the difference between the
inlet (east) and that exit (west) mass fluxes: m_ north ¼ m_ west  m_ east
The velocity is uniform (at U1) as it enters the momentum boundary layer, and
therefore the mass flux into the west plane is:

m_ west ¼ ρU 1 wδT

At the exit plane (east), the velocity changes from zero (no slip) to some non-zero
value at the top edge of the thermal boundary layer. Defining an average velocity in
the thermal boundary layer (whose value depends in part on the relative thickness of
the momentum and thermal boundary layers, yet to be determined):
δðT

m_ east ¼ ρuwdy  ρU T wδT

y¼0
10.2 Forced Convection Across an Isothermal Flat Plate 391

This average velocity is given a subscript “T” to distinguish it from the average
velocity in the momentum boundary layer. An energy balance applied to the nodal
network, considered to be in a steady-state is:

dT    
C ¼ m_ west cp T 1  T ref  ðm_ west  m_ east Þcp T 1  T ref
dt
  T wall  T
 m_ east cp T east  T ref þ ¼ 0@SS
Rk

In this equation, each enthalpy stream is expressed relative to a reference tempera-

ture, and all the reference temperatures cancel. Mass conservation was used to
eliminate the mass flow rate leaving the top surface. The heat transfer rate between
the wall and the fluid in the thermal boundary layer is expressed with a conductive
thermal resistance. The energy balance simplifies to:

  T wall  T
m_ west cp T  T 1 ¼
Rk

This expression shows that the difference between the inlet and outlet enthalpy
flows equals the rate at which heat flows into the control volume from the wall.
Rearranging:

T wall  T
m_ west cp Rk ¼ ¼1
T  T1

The temperature ratio evaluates to 1 if the average temperature is taken to be the

geometric average between the wall and free stream, that is, T ¼ ðT wall þ T 1 Þ=2.
The conductive resistance is defined as the average distance between the
wall and the node representing the boundary layer divided by the thermal
conductivity and area. That is, the functional dependency of the conductive
resistance is given by:
 
δT ðLÞ
1
2 0 þ 2 δT
Rk  ¼
kwL 4kwL

The energy balance is:

 
  δT
m_ west cp Rk  ρU T wδT cp ¼1
4kLw

Solving for the thermal boundary layer thickness:

sffiffiffiffiffiffiffiffiffiffiffiffiffi s
ffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
 ffi
kL αL U1
δT  2 ¼2
ρcp UT U1 UT
392 10 Convection Fundamentals

The heat transfer rate from the wall to the fluid is driven by conduction, and can be
expressed using Newton’s Law of Cooling:

T wall  T T wall  ðT 1 þ T wall Þ=2 kLw

q  hLwðT wall  T 1 Þ ¼  ¼2 ðT wall  T 1 Þ
Rk δT =4 δT
kLw

The convection coefficient is therefore:

k
h2
δT
For a given fluid, the thicker the thermal boundary layer, the lower the convection
coefficient. Expressing the thermal boundary layer and rearranging:
sffiffiffiffiffiffiffiffiffiffiffiffiffi
 
pffiffiffiffiffiffiffiffiffi UT
h  kρcp
L

The convection coefficient increases (to the ½ power) with the thermal conductiv-
ity, density, and specific heat of the fluid. This combination of fluid properties
pffiffiffiffiffiffiffiffiffi
kρcp is the thermal inertia, or thermal hardness of the fluid. The convection
coefficient decreases with the length of the plate (because the boundary layer is
thicker, giving a larger conductive distance). It increases with the average velocity
at the boundary layer exit plane.
The Nusselt number (based on plate length, L) is given by:
ffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
s    sffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
 
hL U1 L ν UT UT
NuL ¼   ReL Pr
k ν α U1 U1

The ratio of the momentum and thermal boundary layers is:

sffiffiffiffiffiffiffiffiffiffi sffiffiffiffiffiffiffiffiffiffiffiffi
δ ν UT UT
  Pr
δT αU 1 U1

where the Prandtl number (ν/α), a combination fluid property, has been
introduced. The average velocity at the exit (east) plane depends on the relative
magnitude of the momentum and thermal boundary layers, but is always less than
the free stream velocity. The Prandtl number is shown for select fluids as a
function of temperature in Fig. 10.7. Air at atmospheric temperature has a
Prandtl number near unity and it is effectively constant. The thermal and
momentum boundary layer thicknesses are approximately the same. The Prandtl
number for liquid water at nominal room temperature is on the order of 10 and
decreases with temperature. The momentum boundary layer is somewhat thicker
10.2 Forced Convection Across an Isothermal Flat Plate 393

100000

10000
Prandtl Number = ν/α ∼ (δ/δΤ)2
1000
Unused Engine Oil

100

10
Sat'd Liquid Water
1
Air @ 1 atm
0.1

0.01
Mercury (a liquid metal)
0.001
0 50 100 150 200 250 300
Temperature (oC)

Fig. 10.7 Prandtl number of selected fluids as a function of temperature

than the thermal layer. Engine oil has a large Prandtl number and the thermal
boundary layer is thin compared to the momentum boundary layer. These three
cases are investigated in the next section. On the other hand, liquid mercury (and
other liquid metals) has a very low Prandtl number, and the thermal boundary
layer is much thicker than the momentum layer. These fluids are the simplest
because the average velocity in the thermal boundary layer is approximately
equal to the free stream velocity.

10.2.2.2 Limiting Case: Thermal Layer Thin Compared

to Momentum Layer
The average velocity at the exit plane is now addressed for the case where the
momentum boundary layer thickness is greater than the thermal boundary layer,
that is, δ > δT. Given the linear velocity profile, the velocity at the edge of the
thermal layer average velocity is:
  
∂u U1
uðy¼δT Þ   δT ¼ δT :
∂y y¼0 δ

The average velocity at the exit plane of the thermal boundary layer is half that:

U 1 δT
UT ¼
2δ
394 10 Convection Fundamentals

Therefore, the ratio of boundary layers is:

rffiffiffisffiffiffiffiffiffiffiffiffiffiffi rffiffiffisffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
 
δ 1 ν U 1 δT
¼ ¼ Pr
δT 2 α U1 2 δ

Rearranging and solving for the ratio of boundary layer thicknesses:

 1=3
δ 1 1= 1=
¼ Pr 3 ¼ 0:793 Pr 3
δT 2

The Prandtl number determines which boundary layer is thicker, momentum or

thermal. The momentum boundary layer grows with kinematic viscosity (ν ¼ μ/ρ),
while the thermal boundary layer grows with thermal diffusivity (α ¼ k/ρ/cp).
The Nusselt number is:
vffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
sffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
 ffi u  1=3
δT u pffiffiffiffiffiffiffiffi 1=
hL t 2
NuL ¼  ReL Pr  ReL Pr  1:12 ReL Pr 3
k δ Pr

The solution obtained from the integral method is:

hL pffiffiffiffiffiffiffiffi 1=
NuL ¼ ¼ 0:664 ReL Pr 3 for Pr < 1
k
The functional dependency is exactly the same.

10.2.2.3 Limiting Case: Thermal Layer Much Thicker

Than the Momentum Layer
In this case, there is a retarded layer very close to the wall relative to the thermal
boundary layer. The average velocity in the thermal boundary layer is therefore
approximately equal to the free stream velocity. The Nusselt number (based on
plate length, L) is given by:
ffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
s   ffi
hL U1 L ν UT pffiffiffiffiffiffiffiffiffiffiffiffi
NuL ¼ ¼ ¼ ReL Pr
k ν α U1
The ratio of the momentum and thermal boundary layers is:
sffiffiffiffiffiffiffiffiffiffiffi
δ ν UT pffiffiffiffiffi
¼ ¼ Pr
δT α U1
10.2 Forced Convection Across an Isothermal Flat Plate 395

10.2.2.4 General Case: Thermal Layer Thicker Than Momentum

Layer But of Comparable Magnitude
Recall the general expression for the thermal boundary layer:
sffiffiffiffiffiffi
αL
δT  2
UT

The average velocity in the thermal boundary layer is determined by integrating the
velocity profile (for a linear velocity in the momentum boundary layer):

δðT ðδ δðT  δ!
U1 y U 1 y2 
UT δT ¼ udy ¼ dy þ U 1 dy ¼ þ U1 ðδT  δÞ
δ δ 2 0
y¼0 y¼0 y¼δ
 
δ
¼ U 1 δT 
2

Squaring the thermal boundary layer, rearranging and inserting the average
velocity:
 
δ
U 1 δT  δT  4αL
2

Rearranging in the form of a quadratic equation:

1 2 δ 2αL
δ  δT  ¼0
2 T 4 U1
Solving and recognizing that the thermal boundary layer must be positive:
ffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
s 2 sffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi3

δ δ 2 4αL δ4 43 αL 5
δT ¼  þ ¼ 1þ 1þ
4 4 U1 4 U 1 δ2

Recall the momentum boundary layer:

rffiffiffiffiffiffiffiffi
νL
δ2
U1
396 10 Convection Fundamentals

Inserting into the thermal boundary layer:

" rffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi# " rffiffiffiffiffiffiffiffiffiffiffiffiffiffi#
δT 1 16α 1 16
¼ 1þ 1þ ¼ 1þ 1þ
δ 4 ν 4 Pr
0qffiffiffiffiffiffiffiffi1 !
U1 L
2L L δ ν 4
NuL ¼ ¼2 ¼ 2@ A pffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
δT δ δT 2 1 þ 1 þ 16=Pr
pffiffiffiffiffiffiffiffi
4 ReL
NuL ¼ pffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi for Pr > 1
1 þ 1 þ 16=Pr

10.3 Natural Convection Across an Isothermal Vertical Plate

When a solid object is surrounded by a stationary fluid at a different temperature,

gravity will induce a flow as shown schematically in Fig. 10.8, which shows a
rectangular plate (heavy line) of vertical length L, and width w (into the page). In
the left schematic, the plate temperature (Twall) is maintained at a higher tempera-
ture than the ambient fluid (T1). In the right schematic, the plate is colder than
ambient. The coordinate system (x, y) is oriented with x vertical (in the direction of
the main flow direction) and y horizontal (distance from the plate into the fluid).
The origin is placed at the leading edge of the plate. Profile sketches of temperature
(left side of plate) and velocity (right side) are shown at the trailing edge (at x ¼ L)
and at a midpoint. The light curved line is the outer edge of the boundary layer that
grows from the leading edge. Unlike forced convection, there is only one boundary
layer because the temperature field creates the density variation that results in a
vertical buoyant force that drives the flow. That is, the momentum and energy
equations are coupled.
For a hot plate in a cold fluid, the fluid in the boundary layer is usually at a lower
density than that of the free stream, and the induced flow is in the upward direction
(left schematic). The induced flow is downward if the plate is colder than the fluid.
One important exception is the case of liquid water below 4  C where the density is
maximum. A plate fixed at the melting point (0  C) surrounded by liquid water at
4  C would experience an upward flow. Water is unusual that way.
Another difference between forced and natural convection, shown in Fig. 10.9, is
that while the main direction of flow is along the direction of the plate (vertical),
there is a horizontal component of flow toward the plate, not away from the plate as
in forced convection. The streamline shown defines a streamtube whose cross-
sectional area decreases in the direction of flow. The flow accelerates within the
streamtube and coincides with the edge of the boundary layer at the end of the plate
(at x ¼ L). The mass flow that enters the streamtube does so with negligible momen-
tum. All the momentum exiting the streamtube is acquired by a buoyant force (which
is a pressure force), and there is a retarding frictional force exerted by the wall.
10.3 Natural Convection Across an Isothermal Vertical Plate 397

Twall
u(L,y)
T∞ y

Twall
u(x,y) T∞
T∞ L u(x,y)
Twall

y T∞
u(L,y)
Twall

Hot Plate Cold Plate

Fig. 10.8 Schematic of natural convection boundary layer on a vertical plate

u(L,y)
yer
ary la

stream
bound

L
line

˙
m
induced
Fig. 10.9 Schematic of the flow
boundary layer and
streamline on a hot plate
398 10 Convection Fundamentals

10.3.1 Mass

The mass flow rate exiting the streamtube (which, at steady-state, equals the mass
flow rate entering it in a much larger area) is:

ðδ
m_ ¼ ρuðL;yÞ wdy  ρU wδ
y¼0

In this expression, a characteristic flow velocity, U*, has been defined as the mass-
average velocity at the exit plane of the boundary layer, and is to be determined.

10.3.2 x-Momentum

Taking a rectangular control volume that contains the boundary layer, the forces
and momentum fluxes (in the x-direction) are shown in Fig. 10.10. The momentum
flux entering the bottom and side surfaces are neglected in comparison to the
momentum flux exiting, which is shown as a “virtual force” acting downward,
equal to the mass flow rate times the average velocity. The pressure acting on the
top and bottom surfaces are the hydrostatic forces and give rise to a net buoyant
force. The weight of the fluid in the CV is a downward gravity force, and a shear
stress on the wall acts to oppose the motion. This schematic is shown for a hot plate.
For a cold plate, the shear stress would act in the opposite direction, and the exiting
momentum flux would act upward on the bottom surface. The pressure and gravity
forces would be the same as in this schematic.

Fig. 10.10 Boundary layer

control volume used for
momentum balance on a hot
plate with forces shown
10.3 Natural Convection Across an Isothermal Vertical Plate 399

At steady-state, the sum of the upward forces equals the sum of the downward
forces:
 
Fpressure  Fgravity ¼ FNet ¼  Ffriction  }F}momentum

The pressure difference between top and bottom yields a net upward force, which is
the buoyant force (equal to the weight of the displaced fluid, that is, Archimedes’
principle). Gravity acts downward. The net upward force on the fluid in the
boundary layer is the difference between the buoyant and gravity forces. The
friction force and momentum virtual force are upward for a hot plate and downward
for a cold plate. The analysis for a hot plate will be performed (+ signs), and the
results are the same for a cold plate, with the direction of flow reversed.
Expressing each term further:
  1
Pbottom  Ptop wδ  ρgðLwδÞ ¼ τwall Lw þ ρU 2 wδ
2

The pressures are the hydrostatic pressures in the far field that are imposed across
the boundary layer (which could be proven
 formally with  a dimensional analysis of
the y-momentum equation). That is, Pbottom  Ptop ¼ ρ1 gL. The momentum
equation becomes:
1
ρ1 gðLwδÞ  ρgðLwδÞ ¼ τwall Lw þ ρU 2 wδ
2

Rearranging:
1
ðρ1  ρÞgδ ¼ ρU 2 ðδ=LÞ þ τwall
2

The average shear stress at the wall is taken to be proportional to the viscosity and
the average velocity gradient taken to be the overall change in velocity over the half
width of the boundary layer:

ðL 
1 ∂u U 2μU 
τwall ¼ μ  μ ¼
L ∂y y¼0 δ=2 δ
x¼0

The x-momentum equation, in terms of two unknowns (δ and U*) is:

1 2 δ 2μU 
ðρ1  ρÞgδ ¼ ρU þ
2 L δ
The underlying physics of this equation is that there is a net upward buoyant force
(left hand side) that drives the flow. This force gives rise to a change in momentum
of the heated fluid and is opposed by a frictional force on the wall.
400 10 Convection Fundamentals

10.3.2.1 Coefficient of Thermal Expansion, b

The difference in density between the fluid in the boundary layer and that in the free
stream is what drives the flow. That difference is related to the difference in
temperature through a fluid property, the coefficient of thermal expansion.
The difference in density between the boundary layer and free stream can be
expressed in terms of temperature by taking a Taylor-type expansion of density,
which, being a thermodynamic property, is in general a function of two other
thermodynamic variables, namely:
   
∂ρ  ∂ρ 
ρðT; PÞ ¼ ρ1 þ  ð T  T 1 Þ þ  ðP  P1 Þ þ O ðT  T 1 Þ2
∂T P ∂P T
 
þ O ð P  P1 Þ 2

The coefficient of thermal expansion is a fluid property defined as:


1 ∂ρ 
β
ρ ∂T P

Therefore, neglecting the pressure effect on density in comparison to the tempera-

ture effect (a justifiable approximation, from a more detailed dimensional analysis)
and the second-order terms:

∂ρ 
ρ1  ρ    ðT  T 1 Þ ¼ ρβðT  T 1 Þ
∂T P

The temperature for the control volume used is taken to be the geometric average:
 
T wall þ T 1 1
ρ1  ρ  ρβ  T 1 ¼ ρβðT wall  T 1 Þ
2 2

The momentum equation is therefore expressed in terms of the temperature, not

density, difference:

δ 4μU 
ρgβðT wall  T 1 Þδ ¼ ρU 2 þ
L δ
There is, in general, a balance of three effects: net buoyancy, momentum change,
and friction. Buoyancy drives the flow, adding momentum, which is opposed
by friction.

10.3.3 Energy

A thermal resistance network for a 1-node steady-state model of the thermal

boundary is shown in Fig. 10.11. The dashed line shows the control volume, with
10.3 Natural Convection Across an Isothermal Vertical Plate 401

Fig. 10.11 1-Node energy

model of natural convection
boundary layer

ambient temperature fluid entering the bottom (south) and side (east) surfaces,
modeled as current sources. The mass flow exiting the top (north) does so with an
average temperature (taken to be the geometric average of the wall and ambient).
The wall exchanges heat with the average temperature in the boundary layer
through a conductive resistance.
The steady-state energy balance (energy in ¼ energy out), recognizing the east
and south mass flows both enter the control volume at ambient temperature, and that
their sum equals the mass flowing exiting the top:

  T wall  T  
_ p T 1  T ref þ
mc _ p T  T ref
¼ mc
Rk
402 10 Convection Fundamentals

Upon rearrangement, the reference temperature cancels:

T wall  T  
_ p T  T1
¼ mc
Rk
This expression shows that heat conducted from the hot wall into the boundary
layer equals the change in enthalpy flux between inlet and outlet flows. The
temperature average is midway between the wall and ambient, and therefore the
temperature terms are equal, and the underlying relationship for a natural convec-
tion boundary layer is:

_ p Rk ¼ 1
mc

Expressing the conductive resistance as an average difference to the midpoint of the

boundary layer (¼ ½*(0 + δ/2) ¼ δ/4) divided by the thermal conductivity and
surface area:

_ p ðδ=4Þ ρcp U  wδ2

mc
¼ ¼1
kLw 4kLw

The energy equation yields a simple relationship between the boundary layer
thickness and the characteristic velocity, namely:

U  δ2
¼1
4αL

10.3.3.1 Combined Momentum and Energy Equations

The momentum equation can be rearranged as:


δ2 gβðT wall  T 1 ÞL  U 2 ¼ 4νU  L

Using the energy equation to eliminate the boundary layer thickness:

4αL

gβðT wall  T 1 ÞL  U2 ¼ 4νU  L
U

Rearranging:

gβðT wall  T 1 ÞL
U 2 ¼
Pr þ 1
10.3 Natural Convection Across an Isothermal Vertical Plate 403

Back substituting the energy equation and rearranging:

 1=4
pffiffiffi ðPr þ 1ÞL
δ2 α
gβðT wall  T 1 Þ

The boundary layer thickness normalized by the plate length (and multiplying and
dividing by the square root of the kinematic viscosity):

rffiffiffi" #1=4 1=
δ α ðPr þ 1Þ ðPr þ 1Þ
4

2 ¼ 2 pffiffiffiffiffi
L ν gβðT wall T 1 ÞL
3
PrðGrÞ
1=
4
ν2

10.3.3.2 Newton’s Law of Cooling

As before, Newton’s Law of Cooling defines a convection coefficient:
 
T w  T kwL T wall  T 1
qLw ¼  ¼ hLwðT wall  T 1 Þ
Rk δ=4 2

2k
h
δ
The average Nusselt number is:
1 pffiffiffiffiffi
=
hL L GrL4 Pr
NuL ¼ 2 
k δ ð1 þ PrÞ1=4

The ¼ power dependency on Grashof number is demonstrated. There is a more

complicated dependency on Prandtl number than Grashof number. There are
limiting cases, however:
1 pffiffiffiffiffi
=
NuL  Gr L4 Pr for Pr 1 ði:e:, liquid metalsÞ
1= 1=
NuL  ðGr L
PrÞ ¼ RaL 4 for Pr  1 ði:e:, cold liquid water, unused engine oilÞ
4

As with forced convection, this model applies as long as the flow remains laminar.
In reality, as the Raleigh number exceeds approximately 109, the flow within the
boundary layer becomes turbulent, and an additional mechanism for the transport of
momentum and energy becomes important. This turbulence mechanism is difficult
to model simply.
404 10 Convection Fundamentals

Workshop 10.1. Boundary Layer Thickness for the Mug of Coffee

Calculate the thickness of the boundary layer and the corresponding characteristic
flow velocity on the air and the coffee sides of the mug wall for the base coffee
case at a time when the coffee is at 70  C (results from Chap. 9, Figs. 10.9–10.11).
Compare the boundary layer thickness to the diameter of the mug to determine
whether the boundary layer is truly thin compared to the relevant physical
dimension.
Internal Flows Models
11

It is common for engineering devices to be designed with pipes or ducts in which

fluids (gas or liquid) flow. The fluid enters at an inlet temperature and the walls of
the duct are at a different temperature. This chapter addresses internal flows in
which a single ducted stream is involved. The next chapter considers heat
exchangers, where one stream transfers heat to another. When mass flows into a
control volume, it carries the energy associated with it. This mechanism of energy
transfer is termed advection, and will be modeled as a fluid temperature-dependent
current source. There are also momentum considerations, as the flows are driven by
pressure (or gravity) and opposed by friction, so that pumps or compressors are
generally needed to move the fluid.
In this chapter, straight sections of pipe are considered, but in practice, an
internal flow frequently consists of straight sections followed by turns or coils.
Conceptually, the flow path is treated here as if the actual flow path were straight-
ened out into a single straight section of pipe with the same total length. The turns in
practice can introduce turbulence, and thereby enhance the convection coefficients,
as well as introduce additional pressure drops.

11.1 Thermal Length

The concept of a thermal length is useful for internal heat transfer situations and a
tool for heat exchanger design. Consider Fig. 11.1 in which a fluid enters a pipe of
inner diameter Di with an average flow velocity V (or mass flow rate m_ ¼ ρVA,
where ρ is the fluid density and A ¼ πD2i =4 is the cross-sectional area) at a uniform
temperature Tin. Its environment (the fluid in the outer stream) is maintained at a
temperature T1. If no phase change occurs inside, the fluid exits the pipe of length
L at a different average temperature Tbulk,out. The mass flow rate is the same at any
cross-section, but the flow velocity changes inversely with the fluid density. The
bulk temperature is formally defined shortly.

# Springer International Publishing Switzerland 2015 405

G. Sidebotham, Heat Transfer Modeling: An Inductive Approach,
DOI 10.1007/978-3-319-14514-3_11
406 11 Internal Flows Models

Fig. 11.1 Schematic of isolated tube of flowing fluid surrounded by a gas (or other fluid) at a
different temperature. End view (left) and side view (right)

Fig. 11.2 Temperature profiles for thermally short (left) and thermally long behaviors (right)

There are two temperature differences that need to be carefully distinguished.

On the one hand, there is the change of the inner fluid temperature as it flows in the
pipe ðT bulk, out  T in Þ. On the other hand, there is the temperature difference that
drives the heat transfer, which varies along the length of the duct (from T 1  T in at
the inlet section to T 1  T bulk, out at the exit). This distinction is demonstrated by the
sketches of temperature profiles for a heating event in Fig. 11.2, which shows axial
profiles of the bulk average (defined shortly) fluid temperature plotted along the
pipe. The temperature difference that drives the heat transfer is shown by the
arrows. The effect it has on the fluid temperature is evidenced by the temperature
as it approaches ambient. In the thermally short case, the temperature of the fluid
does not change substantially, and the temperature difference that drives the heat
changes very little with position. In contrast, for thermally long behavior, the fluid
attains thermal equilibrium with ambient prior to exiting. The heat transfer rate is
highest at the entrance and decreases to zero once equilibrium is attained.
11.1 Thermal Length 407

Formally, a thermally short stream can be defined as one in which the exit
temperature is much closer to the inlet temperature than it is to ambient temperature.
That is:

jT out  T in j  jT 1  T in j Thermally Short Streams

There is heat exchanged, but it does not substantially change the temperature of the
fluid as it passes through the pipe. For example, in a car radiator, the air that is
actively blown through (by a fan and/or the speed of the vehicle) is intended to
behave as a thermally short stream. The inlet air temperature is typically 25  C (Tin),
while the inlet coolant flowing into the heat exchanger is typically 90  C (which is
considered T1 from the perspective of the air). The air exiting is not intended to exit
at a substantially higher temperature and could be maybe 30  C (Tout). So the change
in air temperature as it flows through (Tout  Tin) is 5  C, while the temperature
difference driving the heat (T1  Tin) is 65  C at the inlet. The air is intended to act as
close to an infinite reservoir as possible.
On the other hand, a thermally long stream is one in which the temperature exits
much closer to the ambient temperature than to the inlet temperature:

jT out  T in j  jT 1  T out j Thermally Long Streams

For thermally long streams, the fluid flowing in the pipe attains thermal equilibrium
with the outer fluid.
The “thermal length” of a stream is defined when the change in temperature of
the stream is comparable to the difference between the outlet temperature and the
ambient temperature. That is:

jT out  T in j  jT 1  T out j Defines Thermal Length

The functional dependency of thermal length can be obtained using a transient heat
transfer analysis, by adopting a Lagrangian viewpoint, that is, one which follows a fluid
element as it passes through. The thermal length (Lt) is defined when the time the fluid
particle spends inside the pipe (its residence time) equals its thermal response time:

Residence Time ¼ Thermal Response Time

Lt
¼ RC
V
Applied to an element of fluid of length Δx, the thermal resistance can be expressed
in terms of an overall heat transfer coefficient Ui (based on the inner diameter),
which consists generally of three resistances in series: inner convection, across the
walls of the pipe, and an outer convection. The capacitance of the element is the
mass of the element times its specific heat:



Lt 1 πD2i
¼ ρcp Δx
V U i πDi Δx 4
408 11 Internal Flows Models

Expressing the thermal length in terms of the average flow velocity:

ρcp VDi
Lt ¼
4U i
Expressing it in terms of the mass flow rate:

_ p
mc
Lt ¼
πU i Di
Notice that the physical length of the pipe (L) is not involved in the thermal length
(Lt). Rather, a comparison of the actual length of pipe to the thermal length will
determine whether a stream is thermally long, short, or intermediate. If the pipe is
much longer than the thermal length (L >> Lt), the pipe is thermally long. Making it
any longer will not increase the heat transfer rate. Notice that a pipe of a given length
can behave either as a thermally short pipe or a thermally long pipe merely by
changing the flow rate. At low flows, the fluid spends a long time in the pipe and will
reach thermal equilibrium (thermally long behavior) and vice versa at high flows.
The same relationship for thermal length will arise in the subsequent steady-state
(Eulerian) analysis.

11.1.1 Basic Thermodynamic and Heat Transfer Relations

Internal flow and heat exchanger design theory involves a combination of basic
open-system energy balances (thermodynamics) and rate equations (heat transfer).
A thermodynamic specification would be, for example, that the fluid entering
should be at a specific temperature, given an input mass flow rate, and should exit
at another specific temperature, and the energy balance would yield the required
heat transfer rate. A heat transfer calculation would be to determine the length of
pipe required for a given pipe diameter, flow rate, and temperature of the ambient
(outer). Alternatively, given the temperature of the ambient and a given pipe length,
the exiting temperature could be sought. There are many different ways problems
can be formulated, and it possible to inadvertently either over-specify or under-
specify a problem.
An overall steady-state energy balance in the form enthalpy flow in ¼ enthalpy
flow out is:

_ hin þ Q_ in ¼ m
m _ hout ðþKE þ PEÞ

The unfortunate nomenclature conflict in which “h” is used to represent convection

coefficients in heat transfer applications and enthalpy in thermodynamic
applications is addressed here by drawing a line through h to represent enthalpy.
11.1 Thermal Length 409

Kinetic and potential energy changes are generally considered to be negligible in

heat transfer problems, but could be included when needed, such as in high speed or
gravity-driven flows.
When there is no phase change, the enthalpy can be related to temperature
through the specific heat (at constant pressure). For a uniform inlet stream (for
which the temperature does not change with radial position), the enthalpy, relative
to a reference temperature, is:

Tðin
 

hin ¼ cp dT ¼ cp T in  T ref
T ref

where the average specific heat is defined.

Internal flows are generally two dimensional. The temperature changes with
distance from the entrance (x), and the temperature also varies with radial position
(r). For a heating event (ambient hotter than inlet), the temperature of the fluid near
the inner wall is higher than the fluid near the centerline. However, at a given
distance downstream (x) there is an average temperature, called the bulk tempera-
ture, which yields the correct total enthalpy flux. The bulk temperature at any
position x is therefore defined by:

Dð
i =2
   
_ p T bulk  T ref 
mc ρucp T  T ref 2πrdr
r¼0

Assuming constant properties (ρ and cp):

Dð
i =2
2πρ
T bulkðxÞ ¼ uðrÞ T ðrÞ rdr for a single phase process
m_
r¼0

The energy balance equation, rearranged for the outlet temperature, becomes:

Q_ in
T bulk, out ¼ T bulk, in þ for a single phase process
_ p
mc

For an internal flow, there is no such thing as an “ambient” temperature outside of a

thermal boundary layer. The bulk temperature is an average temperature within the
duct, intermediate to the wall temperature and fluid temperature at the centerline.
This thermodynamic equation places an overall energy balance constraint. It
does not address how it is accomplished, for example, how long a pipe is needed.
However, the rate of heat transfer between the inner and outer fluids is expressed in
terms of system parameters, including flow rate, geometry, and thermal properties
410 11 Internal Flows Models

of both fluids and the walls separating them. All of these factors can be boiled down
to a Newton’s Law of Cooling type expression, with an overall heat transfer
coefficient (U ) substituted for a single convection coefficient (h):

ðL
   
Q_ in ¼ UdA T 1  T bulkðxÞ  U i ðπDi LÞ T 1  T bulk
x¼0

The differential area (dA ¼ πDi dx) represents a specific surface area that separates the
two fluids, for example, the inner surface of a differential element. The overall heat
transfer coefficient, U, quantifies the effectiveness of the heat exchange and can be
determined with thermal resistance methods developed in earlier chapters. In the
integral, the bulk temperature, which varies with distance x along the duct, represents
the mass flow weighted average temperature of the fluid at any cross-section.
The main objective of this chapter is to develop a modeling capability to
estimate the rate of heat transfer between a fluid flowing in a duct and the walls
of the duct. The principles will be introduced with a water-tube boiler example
(inspired by Cooper Union student Julien Caubel, who designed and built a steam
boiler for his senior project). A long, straight section of pipe of constant diameter is
considered to be broken up into three sections (Fig. 11.3): a preheater (sometimes
called an economizer) in which feedwater (as a subcooled liquid) is heated to its
boiling point, a boiler section where the phase change occurs, and a superheater
section where the steam is raised above its boiling point. Figure 11.4 shows, to
scale, a plot of the temperature plotted against heat addition (per unit mass) for
water at a pressure of 2.0 bar that enters at 100  C, is heated to its boiling point
(120.2  C), then boiled completely, and finally heated to a target design temperature
of 164  C. This exit condition is that which, if delivered to a turbine or reciprocating
piston, would exit at 1 atm as a saturated vapor, and is calculated assuming an
isentropic expansion of a perfect gas from 2 to 1 bar, with cp ¼ 2.02 kJ/kg/K and gas
constant RH2O ¼ 0.461 kJ/kg/K. Note that the energy required to change the phase is
much larger than that required to change the temperature of the liquid (in the
preheater) and the vapor (in the superheater).
In practice, these three functions are often accomplished with very different
design configurations due in part to the different fundamental heat transfer behavior
and to the large change in fluid density associated with phase change. However, the
response of water in a straight pipe from a compressed liquid all the way to a

Hot “air”

Preheater Section Boiler Section Superheater Section

Fig. 11.3 Schematic of water-tube boiler

11.1 Thermal Length 411

180

170

160
Temperature (oC)

150 Preheater
140

Superheater
130
Boiler
120

110

100
0 500 1000 1500 2000 2500
Heat Transferred to Water (kJ/kg)

Fig. 11.4 Water temperature as a function of heat added to the water (entering at 100  C) at a
constant pressure of 2.0 bar (pressure drop neglected)

superheated vapor is a simple design configuration, and allows for a good example
of the principle objectives of this chapter. The goal of the heat transfer analysis is to
turn the x-axis from one of heat exchanged to physical length.
The first example considered is the boiler section, because the two fluid
temperatures can be taken as the inlet temperatures, and there is no ambiguity
about the average fluid temperature. On the water side, the fluid is changing phase
along the entire length, and therefore does not change temperature as it is heated.
On the air side, the flow is considered as an external flow, where the fluid is heated
in a thermal boundary layer, but outside that layer, the temperature remains equal to
the inlet temperature around the perimeter of the pipe. The cooling experienced by
the air is considered to show up in the wake region. The second and third examples
will treat the preheater and superheater, where the inner fluid temperature changes
with position, while the external fluid (hot “air”) is the same along the length. Given
a boiler length in Example 1, the lengths of the preheater and superheater will be
determined.
The approach to solving internal flow problems depends on how the problem
statement is posed. In the following example, which has several parts, the geometry
of the pipe (length, diameter, and wall thickness) in the boiler section is specified, as
is the thermodynamic state of the entrance and exit. The internal mass flow rate (and
therefore flow velocity) is to be determined. Since Nusselt number correlations will
be used to estimate convection coefficients (rather than specifying them in the
problem statement), an iterative procedure will be required to obtain an estimate of
the mass flow rate. Then, with that mass flow rate, the required lengths of the
preheater and superheater are determined.
412 11 Internal Flows Models

11.2 Example: Water-Tube Boiler

In a water-tube boiler operating at steady-state (Fig. 11.1), saturated liquid water at

an absolute pressure of 2.0 bar enters a 2.0 m long (L) K-type ¼00 copper pipe (with
actual outer diameter D0 ¼ 9.5 mm and wall thickness t ¼ 0.89 mm). The exterior of
the pipe is subjected to a cross-flow of combustion projects at 1 atm, T1 ¼ 950 K
with an approach velocity of Vgas ¼ 1.0 m/s. Determine the mass flow rate of water
(ṁw) required to completely boil the water. Then determine the lengths of the
preheat and superheat sections, assuming the inlet to the preheater is at 100  C, and
the exit of the superheater is at 164  C.

11.2.1 1-Node Models with Advective Flow as a Current Source

A thermal resistance model for this problem is shown in Fig. 11.5. This model
applies to each section (preheater, boiler, and superheater), provided the enthalpies
and temperatures are suitably chosen. Space has been discretized by considering the
entire inside section of the pipe as a control volume represented by a single node.
The fluid entering the pipe is modeled as a current source, since enthalpy (internal

Fig. 11.5 Resistance network. Top image is superimposed on a side view pipe schematic
11.2 Example: Water-Tube Boiler 413

energy plus flow work) is carried into the control volume with the entering mass.
It is not modeled as a resistance, since the energy crossing the boundary is not
driven by a temperature difference. It is not heat transfer; it is mass transfer of a
bulk flow (as opposed to a diffusional mass flow driven by concentration gradients).
This mechanism of energy transport is termed advection. The fluid exiting is
modeled as an advective current sink.
The main node represents an average temperature of the water inside the pipe (T).
It is in thermal contact with the hot combustion product gases through a series
resistance channel which consists of a convective resistance on the inside (Rinner), a
conductive resistance across the wall of the tube (Rwall), and a convective resistance
on the outside surface (Router). The temperature of the gases will be lower in the wake
region, but that has little influence in the thermal boundary layer region on the
outside of the pipe. There are no capacitors shown in the model because the system is
at steady-state. It would not be difficult to estimate a response time to a sudden
change in inner or outer fluid temperature.

11.2.2 Energy Balance

Considering the general nodal equation for the main water node, with E
representing the total energy stored in the control volume:

dE X _
¼ Q in ¼ 0 at Steady State
dt
T1  T
m_ w 
hin  m_ w 
hout þ ¼0
R
Solving for the desired mass flow rate of water:

T1  T
m_ w ¼
R
hfg

In this expression, the conventional heat of vaporization nomenclature  hfg ¼ 

hout
hin has been used, recognizing that hout is specified as saturated vapor and 
hin is
specified as saturated liquid for this particular problem. The water temperature,
assuming thermodynamic equilibrium exists, is the saturation temperature
corresponding to the internal pressure, which is assumed to be constant. For
P ¼ 2.0 bar ¼ 2.0(10)5 N/m2, the corresponding saturation temperature is
120.4  C, obtained from steam tables. 1If a significant pressure drop occurred, the
change in corresponding saturation temperature would have to be considered.

1
http://www.engineeringtoolbox.com/saturated-steam-properties-d_101.html.
414 11 Internal Flows Models

11.2.3 Thermal Resistance

The thermal resistance, given an inner convection coefficient hi and inner surface
area Ai, pipe thickness t, thermal conductivity kp, and average area (between inside
and outside), outer convection heat transfer coefficient ho, and outer surface area
Ao, is expressed as:

1 t 1
R¼ þ þ
h i Ai k p A h o Ao

The exact solution for the conduction across the pipe could be used, but the thin-wall
approximation is valid in this case.
It is useful to express the resistance in terms of an overall heat transfer coeffi-
cient (U) based on a particular area. Using the pipe inner surface area as the basis,
the resistance is expressed as:


1 1 1 tAi Ai
R¼ ¼ þ þ
U i Ai Ai h i k p A h o Ao

For the cylindrical geometry:

Ai π ðD  2tÞL 2t
¼ ¼1
Ao πDL D
Ai π ðD  2tÞL 1  2t=D
¼ ¼
A πDLþπ2ðD2tÞL 1  t=D

For a thin-walled pipe (t/D << 1), the overall heat transfer coefficient is:

1 1 t 1
¼ þ þ thin wall pipe
U i hi k p ho

Here, the three terms represent the three resistances in series. If any one of them is
significantly larger than the other two, it will control the heat transfer rate, and the
other two can be neglected, in practice. The conductive resistance is easily calcu-
lated, and in this case is negligible in comparison to both convection coefficients
(but still included in calculations). The outer convection environment is one of a
mildly forced external convection of a gas, and a direct Nusselt number correlation
can be used. The inner convection environment is one of forced internal convection.
However, the inner flow velocity is not known from the problem statement.
In addition, the fluid enters as a liquid, changes phase within the boiler section,
and exits as a vapor (at the same mass flow rate, but very different flow velocity)
with very different thermal flow rate. Therefore, an iterative solution method will
be developed.
11.2 Example: Water-Tube Boiler 415

11.2.4 Nusselt Number Correlations

11.2.4.1 External Flow

The result of this particular problem is heavily dependent on the convection
coefficient used on the outside of the pipe. A popular Nusselt number for forced
convection across a cylinder of outer diameter D is the Knudsen/Katz correlation:

hD
NuD ¼ ¼ CReDn Pr 1=3
k
The values of the constant “C” and exponent “n” depend on the value of the
Reynolds number of the cross-flow, and are presented in Table 11.1, with a
modification of the published ranges to allow a smooth transition between ranges
(as was done in the Chap. 9).
There are published correlations for a bank of tubes, which are more complicated
once the flow patterns begin to interfere, but modeling the flow as an isolated
cylinder is expected to provide a reasonable first approximation for the problem
at hand.
The relevant properties for air at a film temperature of (Tw + T1)/
2 ~ 400  C ¼ 673 K were obtained online2 and input into the following equations:

ρVDi ð0:524 kg=m3 Þð1 m=sÞð0:00772 mÞ

ReD ¼ ¼   ¼ 123:3
μ 3:28ð10Þ5 kg=s=m

The constants from Table 11.1 are C ¼ 0.683 and n ¼ 0.466. The convection
coefficient is therefore:

k 0:0515 W=m=K
ho ¼ C ReDn Pr 1=3 ¼ 0:683 ð123:3Þ0:466 0:681=3
D 0:00772 m
W
ho ¼ 26:7
m2 C
That number should seem reasonable by this time for forced convection of a gas.

Table 11.1 Correlation Reynolds number range C n

parameters for forced flow
0.4–4 0.989 0.330
across a cylinder
4–35 0.911 0.385
35–4,083 0.683 0.466
4,083–40,045 0.193 0.618
40,045–400,000 0.0266 0.805

2
http://www.engineeringtoolbox.com/air-properties-d_156.html.
416 11 Internal Flows Models

11.2.4.2 Internal Flow

Estimating an appropriate convection coefficient value for the inside of the pipe
requires use of Nusselt number-type correlations for the case when the fluid is
boiling. Published approaches to this scenario model the convection coefficient as
the sum of that obtained for a forced convection without boiling and a component
attributed to the boiling. The boiling component depends on the viscosity, surface
tension, heat of vaporization, liquid-specific heat, densities of saturated liquid and
vapor, and Prandtl number of the liquid and can be difficult to interpret and
implement. However, neglect of the boiling component results in an underestimate
of the inner convection coefficient. In this case, it is anticipated that the convection
coefficient on the outside will be much lower than even this underestimate, and
therefore will be the controlling factor in the overall heat transfer coefficient.
The approach taken here will be to calculate the convection coefficient for forced
convection without the boiling contribution at both the inlet (where it is liquid) and
outlet (where it is steam). It might be expected that the former will be much larger
than the latter (because it is a liquid vs. gas comparison), and that the latter will be
comparable to the outer convection. However, the flow velocity at the outlet will be
much larger than at the inlet (due to the large decrease in fluid density), making a
strong forced convection, so judgment on that is deferred.
The flow inside the pipe is classified as forced convection for internal flows. The
Dittus/Boelter correlation for fully developed turbulent flow in a pipe of inner
diameter D is:

NuD ¼ 0:023Re0:8
D Pr
n
for turbulent flow

The exponent on the Prandtl number (n) is 0.4 for heating and 0.3 for cooling.
For laminar flow, the Nusselt number is a constant:

NuD ¼ 3:66 for laminar flow

Rather than defining a transition Reynolds number, it is recommended that the

turbulent flow formula be evaluated (once the Reynolds and Prandtl numbers have
been calculated). If that value is below 3.66, then set the Nusselt number to 3.66
(using an if statement, for example).
For the problem at hand, the mass flow rate is the same at the inlet and exit of the
boiler section, but the flow velocity is different due to the change in density.
Therefore, it is useful to express the Reynolds number in terms of the mass flow
rate. Multiplying and dividing by the cross-sectional area A ¼ πD2/4:

ρVD ðρVAÞD 4m_ w

ReD ¼ ¼  2 ¼
μ μ πD =4 πμ D

For the problem as posed, a guess for mass flow rate will be made, which allows the
Reynolds number, hence Nusselt number, hence convection coefficient to be
evaluated. It is intriguing that the Reynolds number depends only on the viscosity
of the fluid for this particular problem (same mass flow and diameter at the entrance
11.2 Example: Water-Tube Boiler 417

and exit). The entering liquid has a much higher viscosity and therefore a lower
Reynolds number than the exiting vapor. Any difference in Reynolds number is not
due to a different velocity.
It is instructive to express the convection coefficient (for turbulent flows)
in terms of the fluid properties, and then determine the major contributing factors
that distinguish liquids from vapor.

 " #
k 4m_ 0:8 μcp n k1n cpn m_ 0:8
hinner ¼ 0:023 ¼ 0:028 0:8n
D πμD k μ D1:8

For a fixed mass flow rate, there is no direct dependence of convection coefficient
on density, while it increases with thermal conductivity and specific heat and
decreases with viscosity.
Property values for water for saturated liquid and vapor at 2 bar pressure are
shown in Table 11.2. The ratio of the liquid to vapor values is:
2 1n  n 3
kliquid cp, liquid " #
hLiquid 66
kvapor cp, vapor 7
7 ð42:8Þ1n ð2:15Þn
¼ ¼ for turbulent flow in both
hVapor 4 μ 0:8n 5 ð23:5Þ0:8n
liquid
μvapor
streams.

For heating (n ¼ 0.4), the results are (showing effect of each fluid property):

hLiquid ð9:52Þð1:36Þ
¼¼ ¼ 3:66
hVapor ð3:53Þ

That this ratio is precisely equal (to two significant figures) to the laminar Nusselt
number is a complete coincidence. The thermal conductivity effect is almost a
factor of 10, there is a mild specific heat effect, and a stronger viscosity effect
(factor of 3.5). The higher viscosity of the liquid tends to suppress turbulent
fluctuations and thereby reduce the effectiveness of the heat transfer between the
tube inner wall and inner fluid.

Table 11.2 Fluid properties for saturated liquid and vapor water at 2 bar pressure (and their ratio)
H2O properties
Property Sat’d liquid @ 2 bar Sat’d vapor @ 2 bar Ratio (liquid to vapor)
k (W/m/K) 0.685 0.016 42.8
ρ (kg/m3) 959 1.23 779.8
cp (J/kg) 4,248 1,970 2.2
μ (kg/s/m) 2.82E04 1.20E05 23.5
α (m2/s) 1.68E07 6.60E06 0.025
ν (m2/s) 2.94E07 9.76E06 0.030
Pr 1.75 1.48 1.2
418 11 Internal Flows Models

Fig. 11.6 Spreadsheet used to calculate convection coefficients at the entrance and exit of the
boiler tube, and then iteratively determine the mass flow rate

The spreadsheet used to iteratively determine the mass flow rate for the boiler is
shown in Fig. 11.6. Fixed parameter values are compiled at the top. In the iteration
table, the mass flow rate for the first iteration is set to zero. The Nusselt number
correlation is calculated at the entrance and at the exit (vapor), the difference being
in the fluid properties. An if-then-else statement is entered for the Nusselt number
that using the Dittus/Boelter correlation with a lower limit of 3.66, the laminar flow
value. The geometric average of the entrance and exit, and then the overall heat
transfer coefficient (Ui, based on the inner diameter) is calculated. The mass flow is
determined from the 1-node energy balance:

T 1  T w U i D i ðT 1  T w Þ
m_ w ¼ ¼
Rhfg hfg

For the first iteration, where there is no inner flow, the corresponding convection
coefficient values are 325 W/m2/K for the inlet (liquid) and 7.6 W/m2/K for the
outlet (vapor). The average of these (166.2) is approximately six times as large as
the outer coefficient so that the predicted overall heat transfer coefficient is con-
trolled by the external convection. The mass flow rate is calculated using this
overall coefficient for the next iteration. Now, the Reynolds number in the liquid
region is 198, and the corresponding Nusselt number evaluates as the laminar flow
limit (3.66). In the exit region, the Reynolds number is much larger (due to the
lower absolute viscosity), and there is a much larger Nusselt number due to the
strong forced convection. However, the convection coefficient (47.8) is still low
11.2 Example: Water-Tube Boiler 419

800

700 677.0

600

Tube Wall
Temperature (8C)

500 Boiling Combustion

Water Products
400

300

200
120.2

203.5

203.6
100

0
0 2 4 6 8 10 12 14 16
Radial Position (mm)

Fig. 11.7 Temperature vs. radial position in the boiler section

compared to the liquid side. The overall heat transfer coefficient for the second
iteration differs by only a few percent from the first iteration because the outer
convection dominates and is independent of flow rate inside the pipe. The iterative
system converges very quickly to a self-consistent result with a final flow rate of:

4 
3, 600 s

m_ w ¼ 3:44ð10Þ kg=s

¼ 1:24 kg=h
h

This seems like a low practical value (about 1.25 l/h of liquid water fed) and shows
the challenge of designing practical boilers.
The radial temperature profile for the final converged solution is shown in
Fig. 11.7. The thermal boundary layers are shown as straight lines (with a boundary
layer thickness k/h). The conductive resistance across the tube wall is four orders of
magnitude lower than the convective resistances, and therefore there is very little
temperature change across the tube wall, which is maintained at 203.5  C.
The boiling action, conceptually, occurs in the inner boundary layer region, where
the indicated temperature is above the boiling point and is a very complex phenom-
enon. The tube wall temperature is closer to the water temperature because the
thermal resistance on the outside is approximately six times greater than the inside.
As a practical matter, the temperature of combustion product gases is frequently
at a temperature well above the melting temperature of the containing vessel. This
condition is only possible if the walls are cooled by the inner fluid, so that there is a
tremendous jump in temperature across the thermal boundary layer between the hot
gases and the containment wall. If, for some reason, the water flow was interrupted, a
transient heating of the tube wall would occur, and could lead to rupture of the tube.
420 11 Internal Flows Models

11.3 Momentum Considerations: Pressure Loss

An important secondary consideration for internal flows is the pressure drop due to
fluid friction associated with the flow. Generally, the smaller the diameter (for a
fixed mass flow), the higher the convection coefficient, hence heat transfer rates, but
the higher the pressure drop, hence pumping requirements. The trade-off between
pumping costs and heat transfer effectiveness can lead to an economic optimization
problem. A brief outline is presented here of the pressure drop estimation, and fluid
mechanics resources should be consulted if more detail on pressure effects is
needed.
Pressure losses are categorized as major (for losses associated with a straight
section of constant diameter) and minor (associated with changes in flow path area
or direction). Considering the major losses, an overall force balance on a straight
section of pipe (internal hydraulic diameter D and length L) with the inlet at state
1 (pressure P1) and the exit at state 2 (P2) is:
   
P1 πD2 =4 ¼ τwall ðπDLÞ þ P2 πD2 =4

where τwall is the average shear stress that acts along the inner walls of the pipe
(Fig. 11.8). Rearranging for the pressure drop:

L
P1  P2 ¼ ΔP ¼ 4τwall
D
The shear stress at the wall is expressed in terms of the Fanning friction factor
(f) defined by:

1 2
τwall ¼ f ρV
2
where ρV2/2 is the dynamic pressure (the hypothetical pressure if the fluid were
brought to rest without friction). The friction factor is a function of the Reynolds
number of the flow, for incompressible flows and the relative roughness of the pipe
(ε, the average roughness divided by the diameter). For laminar flows (Reynolds

Fig. 11.8 Forces acting on a control volume in a straight section of pipe

11.3 Momentum Considerations: Pressure Loss 421

Table 11.3 Pressure drop mdot 3.44E04 kg/s

calculation for boiler
Di 0.00772 m
section
ε, pipe roughness 2.00E06 m
ε/D, relative roughness 0.00026
L 2m
Entrance Exit
ρ (kg/m3) 959 1.23
Re 201 4,729
f (from Moody Chart) 0.080 0.010
dP/dx (Pa/m) 1.2 113.8
dP/dx*L (atm) 2.29E05 2.25E03

numbers below approximately 2,000), the friction factor is a closed-form

expression:
f ¼ 16=ReD for laminar flows, independent of roughness.

For fully turbulent flows, the friction factor is a function of the relative rough-
ness and is independent of Reynolds number. An implicit function, the Colebrook
equation, can be used, or the friction factor can be obtained from a Moody chart.
The pressure drop is therefore:
L 32f m_ 2 L
ΔP ¼ 2f ρV 2 ¼ 2 5
D π ρD
The second expression converted the velocity in terms of mass flow rate. The
pressure drop is extremely sensitive to the inner diameter (fifth power).
In the boiler section, the density, velocity, and Reynolds number are very
different at the exit than at the entrance. Therefore, it makes sense to consider the
rate of change of pressure by setting the length L to a differential length (dx) and
letting the finite change in pressure (ΔP) shrink to a differential change (dP).
The rate of change of pressure with x is therefore:

dP 2f ρV 2 32f m_ 2
¼ ¼ 2 5
dx D π ρD
Results of a basic spreadsheet pressure drop calculation for the boiler section are
shown in Table 11.3. The top block shows quantities that are invariant. The bottom
block shows the relevant quantities at the entrance and exit sections. The friction
factor was obtained from the laminar formula for the entrance case and from a
Moody chart online (the Fanning-type, not the Darcy type) for the exit case. The
pressure gradient (dP/dx) in units of Pascals per meter is shown to be higher at the
exit than at the entrance, despite the lower friction factor, ultimately a density
effect. However, when multiplied by the actual length, and converted to units of
atmosphere, the pressure drop is of the order of 103 atm. The conclusion for this
422 11 Internal Flows Models

Fig. 11.9 Resistance

. .
network for preheater and m cpTin T m cpTout
superheater sections

1
R UA
T∞

case study is that the assumption of negligible pressure drop across the boiler is
justified.
When there are changes in the flow path (turns or diameter changes), minor flow
losses are introduced with a coefficient, K, multiplied by the dynamic pressure entering
the element. Note that the magnitude of the minor losses can be larger than the major
losses in an actual flow path. The total pressure drop is the sum of the pressure drops in
all straight sections (major losses) and flow path elements (minor losses):

X L ρV 2 X ρV 2
ΔP ¼ f þ K
D 2 2
major minor
section element

The minimum input power required to overcome this pressure drop, from an energy
analysis, is:


P 1 P2
W shaft, min ¼ m_ 
ρ1 ρ2

11.3.1 Preheater and Superheater Lengths

Figure 11.9 shows a resistance network for the preheater and superheater sections
considered as control volumes. Figure 11.5 also applies to this case. The central
node, at temperature T, represents a suitable average temperature of the water in the
pipe in this section that is needed to calculate the rate of heat transfer between the
outer fluid (hot gases) and inner fluid (liquid water in this section of pipe).
Revisiting the problem statement for this application, water enters the preheater
at 2 bar pressure as a compressed liquid (the inlet temperature, Tin ¼ 100  C, is
below the boiling point at that pressure of 120  C). It carries flow energy into the
pipe at the rate m_ w ð
hin þ KE þ PEÞ, where the enthalpy consists of the internal
energy of the inlet fluid (u) and the flow work (P/ρ) required to push it in. The fluid
11.3 Momentum Considerations: Pressure Loss 423

also has kinetic and potential energy. Since there is no phase change, the enthalpy
(relative to a reference) can be related to the temperature through the specific heat.
Neglecting
 kinetic
 and potential energy, the rate of energy flow in is
m_ w cp T in  T ref . The reference temperature is not shown in Fig. 11.9 as it will
be cancelled by the rate at which this reference enthalpy exits. The rate of energy
flow out is expressed similarly. These energy flows refer to bulk temperatures at
precise locations, where the fluid crosses the defined control volume.
Along the pipe length, heat is transferred from the hot gases outside (at T1)
across the pipe wall through three thermal resistances in series (convection to outer
wall, conduction across pipe, and convection to inner fluid). From a steady-state
energy balance, the rate of energy flow out of the control volume is higher than the
inlet energy flow rate by that amount. Unlike the boiler section, where the water
temperature remains at the boiling temperature, the temperature varies with posi-
tion, and so does the local rate of heat transfer. Treating the entire section as a single
lumped element, the heat transfer rate can be approximated based on a suitably
chosen average temperature.
Formally, the energy balance is:

dEstored     T1  Tw
C ¼ m_ w cp T in  T ref  m_ w cp T out  T ref þ
dt R
where C is the total heat capacitance of the system, and Estored represents the energy
stored within the control volume. For steady-state operation, the rate of change of
energy stored is zero (or there is a balance between rates of energy crossing the
boundary), and the energy balance can be rearranged in the following format,
introducing an overall heat transfer coefficient (U):

1 ðT out  T in Þ
UA ¼ ¼ m_ w cp  
R T1  Tw

The thermal resistance (this time based on the outer area, Ao, for variety) is the sum
of the inner convective, pipe wall conductive (using the exact cylindrical form this
time), and outer convective resistances:


1 lnðDo =Di Þ 1 1 1 Do lnðDo =Di Þ Do
R¼ þ þ ¼ þ þ
ho πDo L 2πkpipe L hi πDi L πDo L ho 2kpipe hi D i

The term in parentheses is the inverse of the overall heat transfer coefficient, based
on the outer surface area.


1 1 Do lnðDo =Di Þ Do
¼ þ þ
Uo ho 2kpipe hi D i
424 11 Internal Flows Models

A closed-form expression for the pipe length is therefore:

m_ w cp ðT out  T in Þ
L¼  
πDo U o T 1  T w

In this expression, for the defined problem, three of the four temperatures are well
defined, while the average fluid temperature along the pipe is not. The fluid
temperature is bound between the inlet and outlet temperatures, however.
Figure 11.10 shows a calculation of the preheater and superheater lengths that
determines a minimum, a maximum, and an average length, depending on what is
chosen for the average water temperature. An underestimate, Lmin, sets the average
water temperature equal to the inlet water temperature. That model uses the largest
possible temperature difference driving the heat all along the pipe length.
An overestimate, Lmax, sets the average temperature to the outlet temperature. An
average length uses the geometric average of the inlet and outlet temperatures.
What is apparent for this case study is that the predicted length changes by only a
few percent from minimum to maximum. That is not always the case.
The only differences between the preheater and superheater sections are that the
inner convection coefficient is substantially lower in the latter (making U lower,
and the length longer), the heat capacity of vapor is approximately half that of liquid

Fig. 11.10 Spreadsheet calculation of preheater (left) and superheater (right) lengths
11.4 Internal Heat Transfer Flow Models 425

(making the length shorter) and the inlet and outlet temperatures are different
(a larger temperature change, making the length longer). The average temperature
difference driving the heat is also lower in this section, since the water temperature
is closer to T1. The lengths of the preheater and superheater are both short
compared to the length of the boiler section (2 m, an input to the problem) and
that is primarily due to the much lower enthalpy change required in these sections
(Fig. 11.4).
The way that open system problems are approached depends on the way a given
problem is formulated. In this case, there were three sections of pipe. The length of
the boiler section (2 m) was an input that was defined by a specific design objective.
The analysis of that section results in a determination of the required mass flow rate.
That flow rate became an input to the other two sections, whose inlet and outlet
temperatures were also defined. Also, all the inlet and outlet temperatures were
specified, and the required lengths determined. It is common for a given length of
pipe and inlet temperature and flow rate to be specified and the resulting exit
temperature determined. This example is intended to show several, but not all
possible, variations.

11.4 Internal Heat Transfer Flow Models

In this section, a 1D steady-state heat transfer flow situation is defined in which fluid
steadily enters a pipe of constant inner diameter Di, wall thickness t, and length L at
a mass flow rate ṁ and inlet temperature Tin (Fig. 11.11). The pipe is exposed to an
outer fluid at temperature T1 (constant along entire length). The fluid exit temper-
ature is not specified in the problem statement (in contrast to the preheater and
superheater of the previous section). The fluid does not undergo phase change, so
any change in the fluid enthalpy changes its temperature. Posed with these inputs,
there are three unknown quantities: the fluid exit temperature, the total heat transfer
rate, and the average temperature of the fluid in the pipe.
Two of the three required relationships to solve for the three unknowns are an
overall thermodynamic energy balance that relates the energy flows to the total heat
transfer rate, and a heat transfer relationship that relates the total heat transfer rate to

L
Tin Tout = ?

. T=?
m .
Q=?
x
x+dx

T∞

Fig. 11.11 Schematic diagram for internal flow heat transfer models
426 11 Internal Flows Models

the average fluid temperature. The third relationship needed for closure requires an
assumption for the average fluid temperature. Three 1-node models are developed, and
their model predictions are then compared to a differential 1D flow model, which is
considered to be an exact solution for comparison. That model is developed first.

11.4.1 Differential 1D Flow Model

A section of pipe of differential length dx is considered. A steady-state energy

balance in the form energy in ¼ energy out is given by:
     
_ p T ðxÞ  T ref þ UπDi dx T 1  T ¼ mc
mc _ p T ðxþdxÞ  T ref

In this expression, T(x) is the bulk temperature at location x (the inlet); T(x+dx)
 is the
bulk temperature at the exit of the differential section; T  T ðxÞ þ T ðxþdxÞ =2 is the
average temperature of the fluid in the section; and Tref is a reference temperature,
which cancels. The overall heat transfer coefficient is that based on the inner
diameter (unlike the previous example, which based U on the outer diameter).
Expressing the temperature at the exit as a Taylor expansion from the inlet:

dT

T ðxþdxÞ ¼ T ðxÞ þ dx þ . . .
dx
x

Expanding the energy balance:



  1 dT dT
_ p T ðxÞ þ UπDi dx T 1  T ðxÞ þ T ðxÞ þ
mc dx _ p T ðxÞ þ
¼ mc dx
2 dx dx

Rearranging:


 

dT 1 UπDi UπDi
1 dx ¼ ðT 1  T Þ
dx _ p
2 mc _ p
mc

The second term on the left hand side (from the expansion of the average tempera-
ture in the heat transfer term) is a higher order term than the other two terms in the
equation and becomes negligible in comparison to unity for a differentially sized
element (i.e., as dx goes to zero). Another way of saying that is that the average
temperature is approximately equal to the incoming fluid, which is going to be
defined in the next section as a “thermally short” section. The preheater and
superheater in the previous section behaved that way because the prescribed change
in temperature through the pipe was small compared to the temperature difference
between T1 and the entrance fluid temperature.
11.4 Internal Heat Transfer Flow Models 427

For the problem as stated, the outer fluid temperature is constant along the entire
length of the duct (as opposed to heat exchangers where this temperature also varies
with position), and therefore dT1/dx ¼ 0, and the energy balance can be expressed
with dT ¼ d(T  T1) as:


d ðT  T 1 Þ UπDi
¼ ðT 1  T Þ
dx _ p
mc

The term in parentheses has units of inverse length and is defined as the thermal
length, the same expression that was derived with a Lagrangian viewpoint:

_ p
mc
Lt ¼
UπDi
Separating variables and applying limits from the entrance to an arbitrary point x
within the pipe:

ðT ðx
d ðT  T 1 Þ dx
¼
ðT  T 1 Þ Lt
T in 0

Assuming the thermal length to be constant (the heat transfer coefficients could
vary with position if entrance effects are important or fluid properties change
substantially), the final result for temperature as a function of position (x) is:

T  T1 x
¼ e Lt
T in  T 1
This solution is considered to be the exact solution of the 1D flow situation, but not
a complete solution to the flow situation. The temperature T represents the average
(bulk) temperature of the fluid at a particular position along the pipe. There must
be 2D effects because this average fluid temperature is different from the tempera-
ture of the inner wall surface in order for heat to flow across the inner resistance.
This expression has a very similar appearance and underlying physics to the lumped
capacity transient temperature, but with time replaced with position, and time
constant replaced with thermal length.
Consider the limits. The exit temperature is obtained by setting x to the pipe
length (L):

T out  T 1 L
¼ e Lt
T in  T 1
If the pipe length is much shorter than the thermal length (L << Lt), then the exit
temperature is approximately equal to the inlet temperature. There is heat transfer,
but it does not change the temperature of the inner fluid appreciably. That is
428 11 Internal Flows Models

“thermally short” behavior. If the pipe length is much longer than the thermal
length, then the exit temperature approaches the outer fluid temperature (T1). The
fluid has achieved thermal equilibrium with the outer fluid and is considered to be
“thermally long” behavior. These concepts will be applied to the 1-node models in
the next section.
The total heat transfer rate is determined by:
 
Q_ ¼ mc
L
_ p ðT out  T in Þ ¼ mc
_ p T 1 þ ðT in  T 1 Þe Lt  T in
 L

_ p ðT 1  T in Þ 1  e Lt
¼ mc

In order to compare models, it will prove useful to normalize the actual heat transfer
rate by the maximum possible heat transfer rate, which is what would occur if the
inner fluid were everywhere equal to the inlet temperature:

^_ ¼ Q_ ¼ Q_ _ p
mc L
 L
t L

Q ¼ 1  e Lt ¼ 1  e Lt
Q_ max UAi ðT 1  T in Þ UAi L

Defining a dimensionless length as the pipe length normalized by the thermal

length:

UπDi L UA L
L^ ¼ ¼ ¼
_ p
mc _ p Lt
mc

An expression for the dimensionless heat transfer rate is therefore a function of the
dimensionless length:
 
^
1  e L
Q^_ ¼
L^

This function, along with the results of the following 1-node models to follow, is
shown in Fig. 11.12 (top figure). The dimensionless temperature is shown in the
bottom figure.

11.4.2 1-Node Flow Models

The 1-node resistance model shown in Fig. 11.9 applies to this case. An overall
energy balance is given by:

Q_ ¼ mc
_ p ðT out  T in Þ

This thermodynamic relation has two unknowns ( Q_ and Tout). It is useful, and
common, in heat exchanger applications to define the maximum possible heat
transfer rate that could be obtained, and that occurs simply when the fluid exits at
11.4 Internal Heat Transfer Flow Models 429

.
Q
UA (T∞− Tin )
Thermally Long (low flow)
2
1.8
1.6
1.4
1.2 Thermally Short (high flow)
1
0.8
Average Temp. model
0.6 Well-mixed
0.4
Differential
0.2
0
0.01

0.1

100
10
1
UA
L̂ = .
mc p

Tout − T∞ 0.8
Tin − T∞ 0.6 Well-mixed

0.4 Thermally Long

(low flow) Differential
0.2
UA
0 L̂ = .
mcp
0.01

0.1

100
10
1

−0.2

−0.4 Average Temp. model

Thermally Short
−0.6
(high flow)
−0.8

−1

Fig. 11.12 Dimensionless heat transfer rate (top) and exit temperature (bottom) vs. dimensionless
pipe length for several models

the ambient temperature, that is, for thermally long behavior. The maximum
possible heat transfer rate is:

Q_ max ¼ mc
_ p ðT 1  T in Þ

This concept will apply to the “number of transfer units” (NTU) approach in heat
exchanger applications in the next chapter.
430 11 Internal Flows Models

The heat transfer rate between the outer and inner fluids is given by an overall
heat transfer coefficient (U) across a defined surface area (A) as:
 
Q_ ¼ UA T 1  T

This relation introduces the average temperature of the fluid. In the differential
solution, this temperature was taken to be the geometric average between the inlet
and outlet, but upon Taylor expansion, and shrinking the element size (dx ! 0), the
average temperature becomes approximately equal to the inlet temperature, for
the purposes of calculating the heat transfer rate. In the 1-node model, the entire
pipe length is taken as the element. In other words, the element is not differentially
small, and the average temperature can be treated differently, as is shown in the
following four cases:

11.4.2.1 Thermally Short Model

In the thermally short model, the temperature change of the fluid  (Tout Tin) is
small compared to the temperature difference that drives heat T 1  T . Heat is
transferred to the fluid, but it does not appreciably change its temperature. In this
case, a good approximation is that the average temperature equals the inlet temper-
ature. The overall heat transfer rate, then, can be expressed directly as:

Q_ ¼ UAðT 1  T in Þ

Expressed in dimensionless form, normalized by the maximum possible heat

transfer rate:

Q_ Q_
¼ ¼1
_
Q max UAi ðT 1  T in Þ

Or, simply

^_ ¼ 1
Q

Note that there is in fact a change in fluid temperature predicted by this model.
The average temperature was set equal to the inlet temperature for the purpose of
estimating the heat transfer rate. The thermodynamic relationship yields the exit
temperature:

Q_ UA
T out ¼ T in þ ¼ T in þ ðT 1  T in Þ
_ p
mc _ p
mc
11.4 Internal Heat Transfer Flow Models 431

In terms of the dimensionless length:

T out ¼ T in þ L^ ðT 1  T in Þ
T out  T 1
¼ 1  L^
T in  T 1

Consider the limits. For a pipe length much less than the thermal length (L^ << 1),
the exit temperature approaches the inlet temperature. For a pipe length equal to the
thermal length, the exit temperature equals the outer fluid temperature, and the
model is not expected to be very good. The predicted outlet temperature overshoots
the outer fluid temperature for any pipe length greater than the thermal length. In
short, the thermally short model is expected to be reasonable for situations where the
pipe length is actually short compared to its thermal length, but violates the Second
Law of Thermodynamics for pipe lengths that are longer than the thermal length.

11.4.2.2 Thermally Long Model

In the thermally long model, the fluid inside the pipe is considered to reach a thermal
equilibrium with the outer fluid. This condition is attained by setting the exit
temperature equal to the outer fluid temperature, and the thermodynamic balance is:

Q_ ¼ mc
_ p ðT 1  T in Þ

In dimensionless terms:

Q_ Q_ _ p
mc
¼ ¼
_
Q max UA ðT
i 1  T in Þ UA

Or simply,

Q^_ ¼
1
L^
The dimensionless heat transfer exceeds the maximum possible for pipe lengths that
are shorter than the thermal length, in violation of the Second Law. For pipe lengths
greater than the thermal length, a reasonable trend is apparent. So the thermally long
model gives reasonable trends for pipes that are long compared to the thermal length.

11.4.2.3 Average Temperature Model

It might seem that an improved model would be to set the average fluid temperature
equal to the average of the inlet and exit temperatures, that is:


T in þ T out
Q_ ¼ UA T 1 
2
432 11 Internal Flows Models


T in þ T out
UA T 1  _ p ðT out  T in Þ
¼ mc
2

Solving for the exit temperature (and skipping details, a good exercise. . .):
 
L^ T 1 þ T in 1  L^ =2
T out ¼
1 þ L^ =2

The dimensionless exit temperature is:

T out  T 1 2  L^
¼
T in  T 1 2 þ L^
For thermally short pipes, the outlet temperature approaches inlet temperature. For
thermally long pipes, the outlet temperature does not approach ambient. That is:

lim T out ¼ 2T 1  T in
L^ >1

The outlet temperature overshoots the outer fluid, but remains finite. The dimen-
sionless heat transfer rate is:

^_ ¼
Q
2
2 þ L^
This model gives good agreement with the exact model across thermally long and
short cases, but does not give a good prediction for the exit temperature for
thermally long cases. This situation is similar to the case in transient numerical
analysis when a central difference approximation is used to express the first
derivative of temperature with time. It is a more accurate numerical approximation
than a backward difference when the numerical step size is sufficiently small, but it
gives wrong limits for large step sizes.

11.4.2.4 Well-Mixed Thermal Model

In this model, the exit temperature is set equal to the average temperature. This
model is called a well-mixed model (borrowing from stirred reactor chemical
models) because it mimics the behavior of the type of case shown in Fig. 11.13
which is typical of, say, conditioned air entering a room through inlet vents, with
the air in the room being well mixed (actively or passively), and exiting through
open areas (i.e., underneath doors, through cracks in the wall) that are far removed
from the inlet. There is a region of mixing near the inlet, but the fluid in the room is
at a mixed average temperature. There is simultaneous heat exchange with the
walls, but the temperature of the fluid as it exits equals the average well-mixed
temperature.
11.4 Internal Heat Transfer Flow Models 433

Fig. 11.13 Schematic of a well-mixed internal flow

The heat transfer rate is therefore:

 
Q_ ¼ UA T 1  T ¼ UAðT 1  T out Þ

Combining with the thermodynamic balance and rearranging for the outlet
temperature:

_ p T in L^ T 1 þ T in
UAT 1 þ mc
T out ¼ ¼
_ p
UA þ mc L^ þ 1
Expressed as a dimensionless temperature:

T out  T 1 1
¼
T in  T 1 ^
L þ1
The outlet temperature approaches the outer fluid for long pipes, and the inlet for
short pipes, that is, the temperature exhibits the correct limits.
The dimensionless heat transfer rate is

Q^_ ¼
1
1 þ L^
As shown in Fig. 11.12, this model underestimates the heat transfer rate and
overestimates the dimensionless exit temperature, but gives reasonable trends over
the entire range, and reasonable trends for the exit temperature. It is therefore
considered to be the best general 1-node model to use for internal flow problems.
It also can be advantageous in more detailed numerical simulations, as numerical
stability is assured.

11.4.3 Pipe Flows with Constant Heat Flux

A common internal flow situation is to force heat into a fluid by, for example,
wrapping an electrical heating tape around the outside of the pipe. Electrical energy
is converted to thermal energy and heat is driven into the fluid. The same thermo-
dynamic and heat transfer relations apply, but heat transfer rate is a known input
434 11 Internal Flows Models

quantity. The exit temperature and the temperature of the heating element (the role
of ambient temperature) is unknown (and changes with position).
_ and it is uniformly
The total heat transfer rate forced into the fluid is given by Q,
delivered across the entire length (L) of the pipe. For a differential element of length
dx, a differential heat rate is added ( d Q_ ), which is a constant along the pipe.
It follows that a heating intensity (heat transferred per unit length) is given by:

d Q_ Q_
¼
dx L
The thermodynamic balance for this element is:
 
d Q_ ¼ mc
_ p T bulkðxþdxÞ  T bulkðxÞ

Expanding in a Taylor expansion:


dT bulk dT bulk
dQ_ ¼ mc
_ p T bulkðxÞ þ _ p
dx þ , , ,  T bulkðxÞ ¼ mc dx
dx dx

Rearranging:

dT bulk dQ_ 1 Q_
¼ ¼
dx dx mc _ p
_ p Lmc

The rate of change of bulk temperature with distance is a constant, and therefore, on
integration, the bulk temperature increases linearly with x:

_
Qx
T bulkðxÞ ¼ T inlet þ
_ p
Lmc

The heat transfer balance is:

 
dQ_ ¼ U i πDi dx T element, ðxÞ  T bulk, ðxÞ

The overall heat transfer coefficient (Ui) based on the inner diameter (Di) represents
a series thermal resistance channel across any insulation around the heating tape,
the wall of the pipe, and inner convection (assuming the temperature of the element
itself is uniform, although conductive resistance of the heating element could be
included). Solving for the temperature of the heating element:

Q_
T element, ðxÞ ¼ T bulk, ðxÞ þ
LU i πDi
The element temperature is always hotter than the fluid. The more effective the heat
transfer (Ui) and the larger the diameter, the closer the element temperature is to the
fluid temperature.
Workshop 11.1. Finned Pipe 435

Workshop 11.1. Finned Pipe

Water at temperature T1, pressure P1 enters an aluminum-finned steel pipe

(dimensions defined in sketch) with a mass flow rate of ṁ and flows with negligible
pressure drop. The pipe is exposed to air at T1 with a cross-flow velocity (not
shown in sketch) of Vair. For the parameter values given, determine the heat transfer
_ length of pipe (L), and number of fins (nfins) if the exit is a saturated vapor.
rate (Q),
Treat the external flow as a forced convection over a flat plate of length wfin and
neglect radiation effects. Treat the internal flow as forced convection with fully
developed flow.

Default parameter values

Quantity Symbol Value Unit
Mass flow rate ṁ 0.02 kg/s

Ambient temperature T1 25 C

Inlet temperature T1 150 C
Internal fluid pressure P1 1.0 Atm
Air cross-flow velocity Vair 10 m/s
Pipe thermal conductivity kpipe 43 W/m/ C
Fin thermal conductivity kfin 212 W/m/ C
Pipe outer diameter Dpipe 0.05 m
Pipe thickness tpipe 0.00655 m
Fin thickness tfin 0.00159 m
Fin width wfin 0.19 m
Fin spacing Δx 0.01 m

Results
Quantity Symbol Value Unit
Overall heat transfer coefficient (based on inner area) Ui W/m2/K
Total heat transfer rate Q_ Watts
Pipe length L m
Number of fins (round up) nfins –

A detailed model development follows. It is a good idea to think about this

problem and tackle it first before reading on.
436 11 Internal Flows Models

Fig. 11.14 Section detail.

Note: the average fluid
temperature is the fluid
temperature at this section,
which varies with distance
along the pipe

Fig. 11.15 Overall

equivalent circuit. Note: the
average fluid temperature
represents the average
between inlet and outlet of
the pipe

Model Development
Use a 1-node fin resistance model (Fig. 11.14) and a 1-node overall flow model
(Fig. 11.15). Since the inlet and exit temperature are specified (the exit is a saturated
vapor), the average temperature can be set equal to the geometric average for the
most accuracy. A single closed-form expression can be obtained that includes all
the fins acting in parallel, or a more brute force computer simulation that steps
through the pipe fin by fin until the final desired temperature is achieved.

Workshop 11.2. Solar Pipe in Continuous Mode: 2-Node Model

Water at temperature T1 ¼ 15  C flows into a copper pipe (kpipe ¼ 400 W/m/K) of

outer diameter D ¼ 0.0155 m, thickness tpipe ¼ 0.002 m, length L ¼ 1.35 m and is
exposed to solar radiation (concentrated with a parabolic mirror) with an insolation
value of I0 ¼ 500 W/m2 and a solar collection width to rod diameter ratio of
n ¼ 19.1. An optical efficiency ηopt ¼ 0.9 (defined as the ratio of the solar radiation
that strikes the pipe to the collected radiation) accounts for collected solar energy
Workshop 11.2. Solar Pipe in Continuous Mode: 2-Node Model 437

not absorbed by the pipe. The angle between the incoming solar rays and the pipe
is θ ¼ 20 . The absorption coefficient of incident radiation is α ¼ 0.85. The rod
is exposed to air at a temperature T1 ¼ 25  C with a natural convection
coefficient (from a Nusselt number correlation) that has a mild temperature depen-
dency given by:


T  T 1 1=4
hc, outer ¼ 1:32 W=m2 =K
D

where T is the surface temperature of the pipe. The emissivity of the rod surface is
ε ¼ 0.9. The average temperature of the walls with which the rod exchanges thermal
radiation is Tw1 ¼ 30  C. The convection coefficient on the inside of the pipe can
be approximated using the larger of Dittus/Boelter correlation or laminar flow
lower limit. That is, the Nusselt number (SI units) is the higher of:


hinner ðD  2tÞ ρV ðD  2tÞ 0:8 μcp 0:4
NuD ¼ ¼ 0:023 OR NuD ¼ 3:66
k μ k

For water properties, use ρ ¼ 1,000 kg/m3, cp ¼ 4,200 J/kg/K, k ¼ 0.6 W/m/K, and
μ ¼ 5.0(10)4 kg/m/s. Assume the system is pressurized sufficiently to prevent
water from changing phase.
Calculate, as a function of volumetric flow rate (AV):

• The steady-state average temperature of the pipe (Tpipe, in  C).

• The exit temperature of the water (T2, in  C).
• The minimum pressure required to prevent phase change.
• The rate of heat transfer to the water (in Watts).
• The thermal efficiency, defined as the ratio of heat transfer rate to the water
divided by the collected solar flux (solar radiation rate entering top plane).

Think about and tackle yourself before reading on. . .

438 11 Internal Flows Models

2-Node Well-Mixed Flow Model

The thermal resistance network shown for this system defines an absorbing surface
node (the outer surface of the pipe, at an average temperature Tpipe), which receives
solar radiation (modeled as a current source). This surface is in thermal contact with
the ambient air surrounding it (at temperature T1) and the surrounding walls
(at Tw1). Note that there are two very different radiation modes at play; the
absorption of solar radiation (modeled as a current source), and thermal radiation
of the pipe surrounded by the local environment (modeled as a resistance). The
absorbing surface is in thermal contact with the fluid flowing through it (at an
average temperature T w ) through a conductive resistance across the pipe in series
with an inner convective resistance. The energy flux associated with the water flow
is modeled as current sources entering and leaving. A nodal energy balance on the
absorbing surface node is:

T 1  T pipe T w1  T pipe T w  T pipe

0 ¼ Q_ abs þ þ þ
Rh1 Rr1 Rpipe þ Rh, inner

In the well-mixed flow model (which is numerically stable for all cases, and yields
correct high and low flow limits), the average water temperature for the heat
transfer term is set equal to the exit temperature (T w ¼ T 2 ).

Using Newton’s Law of Cooling and the Stefan–Boltzmann law for radiation
(with σ ¼ 5.669(10)8 W/m2/K4), the thermal resistances are defined as:

1
Rh1 ¼
hc, outer πDL
Workshop 11.2. Solar Pipe in Continuous Mode: 2-Node Model 439

1
Rr1 ¼
hr πDL
1
¼ h 2 i 
εσ T pipe þ 273 þ ðT w1 þ 273Þ2 T pipe þ 273 þ T w1 þ 273 πDL

1
Rhinner ¼  
hc, inner π D  2tpipe L

t
Rpipe ¼ k pipe
(thin-wall approximation, which is valid here)
ðDtpipe ÞL
pipe π

The rate of heat absorption is defined by:

Q_ abs ¼ αηopt nDLI 0 cos θ

The nodal equation can be rearranged into a form suitable for the method of
successive substitution, with a “k” superscript to indicate iteration number, noting
that the thermal resistances between the outer pipe wall and ambient are functions
of temperature:
ðk Þ
Q_ abs þ
T
T1
ðk Þ þ T w1
ðk Þ þ R
2
pipe þRh, inner
ðkþ1Þ Rh1 Rr1
T pipe ¼ 1
ðk Þ þ 1
ðk Þ þ Rpipe þR
1
h, inner
Rh1 Rr1

An energy balance on the water node is:

T pipe  T w
_ p ðT 2  T 1 Þ ¼
mc
Rpipe þ Rh, inner

where the mass flow is m_ ¼ ρAV. Invoking the well-mixed model (T w ¼ T 2 ), and
rearranging for the exit water temperature in a form suitable for Method of
Successive Substitution (note that the Jacobi method uses “k” and Gauss–Seidel
uses “k + 1”):
ðk Þ or ðkþ1Þ
T
ðkþ1Þ
T1
Rflow þ Rpipe
pipe
þRh, inner
T2 ¼ 1
Rflow þ Rpipe þR
1
h, inner

where a new type of thermal resistance has been introduced: Rflow ¼ mc 1

_ p . This
thermal resistance only works in this way for the well-mixed model.
The thermal efficiency of the panel is defined as the heat transferred to the water
divided by the incident solar flux entering the top:

Heat to Fluid _ p ðT 2  T 1 Þ
mc
ηthermal ¼ ¼
Heat Collected nDLI 0 cos θ
440 11 Internal Flows Models

Workshop

In teams of three or four, write program code with:

• A commented title section.

• A commented “INPUT PARAMETER” section in which all parameters are
defined and enter values from the problem statement.
• A commented “DERIVED PARAMETERS” in which quantities that do not
depend on the flow rate or unknown pipe temperature are calculated (i.e., Rpipe).
• A commented “ITERATION LOOP” section that uses an outer FOR loop to step
through AV (flow rate), and an inner WHILE loop in which the method of
successive substitution is implemented, and tested for convergence.
• A commented “RESULTS” section that plots pipe outer temperature, pipe inner
temperature, and fluid exit temperature vs. flow rate (report in units of liters per
hour), thermal efficiency vs. flow rate, and thermal efficiency vs. exit
temperature.

Workshop 11.3. Solar Pipe in Continuous Mode: Multi-node

Model

The problem statement is the same as Workshop 11.2. A multi-node mode is

developed and applied.

Multi-node Well-Mixed Flow Model

A more complex model is required to account for spatial variations in the pipe
surface temperature. On the one hand, such a variation would seem to be important
for flow rates, where there is a substantial change in water temperature with
position. On the other hand, conduction axially in the pipe would tend to maintain
a constant wall temperature, especially for highly conductive pipes (like copper).

Consider the thermal resistance network shown (larger versions shown subse-
quently), in which the pipe and fluid spaces are broken into discrete elements
(represented by nodes) and connected thermally by appropriate thermal resistances.
Workshop 11.3. Solar Pipe in Continuous Mode: Multi-node Model 441

Each pipe surface node absorbs solar radiation, exchanges heat with the environment
(by convection and radiation acting in parallel), with the adjacent inner fluid (by a
conductive resistance across the pipe in series with a convective resistance inside the
pipe), and with the neighboring pipe nodes (by a conductive resistance through the
pipe wall). The end segments (Tp1, TpN, T1, and TN) differ from the interior nodes
(Tp2. . .TpN and T2. . .TN) in that they communicate with only one neighbor. Each
fluid node is in thermal contact with the pipe outer surface through a series thermal
resistance across the pipe wall and convection on the inner surface. There is an inlet
flow and outlet flow to each fluid node and the well-mixed model sets the average
temperature equal to the outlet temperature. Finally, heat is conducted forward and
backward through an axial conductive resistance (in parallel with the flow). This
latter mechanism can be important for highly conductive pipes like copper.
The forward and backward conduction in the fluid is in parallel with the flow and
is a mechanism whose relative importance will be shown to depend on the size of the
spatial grid.
For a numerical choice of N segments (2N nodes), the grid spacing is given by:

L
Δx ¼
N
All the resistance values and solar absorption term are the same as that in the
2-Node model, except that the dimension L is replaced with the grid spacing, Δx.
The axial pipe resistance and axial resistance for the inner fluid are new to this
model and are given by:

Δx Δx
Rp, axial ¼   2  ¼  
kpipe ðπ=4Þ D  D  2tpipe
2 k pipe πDt pipe 1  tpipe =D

Δx
Raxial ¼
kf π ðD  2tÞ2 =4

A nodal energy balance on each of the 2N unknown temperatures, written in the

form suitable for a Gauss–Seidel iteration, results in the following system of
algebraic equations (nonlinear when the outer convection and radiation resistances
are functions of temperature):
For the first pipe node:

ΔQ_ abs þ RT 11 þ RT w1
T
þ Rp, p2axial þ RpipeTþR
1
h, i
T p1 ¼ w1
1
R1 þ Rw1
1
þ Rp,1axial þ Rpipe 1þRh, i

For the interior pipe nodes (i ¼ 2 to N  1):

ΔQ_ abs þ RT 11 þ TRw1

T T
þ Rpp,ði1 Þ
þ Rpp,ðiþ1 Þ
þ RpipeTþR
i
h, i
T pi ¼ w1 axial axial
1
R1 þ Rw1
1
þ Rp,2axial þ Rpipe 1þRh, i
442 11 Internal Flows Models

For the last pipe node:

ΔQ_ abs þ RT 11 þ RT w1
T
þ Rppð,N1 Þ
þ RpipeTþR
N
h, i
T pN ¼ w1 axial
1
R1 þ Rw1
1
þ Rp,1axial þ Rpipe 1þRh, i

An energy balance on the first fluid node, rearranging into a form suitable for
MOSS, and invoking the well-mixed flow model, with Rflow ¼ 1=mc _ p and:
T
T IN
Rflow þ Rpipe þR
p1
h, i
þ RTaxial
2

T1 ¼ 1
Rflow þ Rpipe 1þRh, i þ Raxial
1

For interior fluid nodes (i ¼ 2 to N  1):

T
T i1
Rflow þ Rpipe þR
pi
h, i
þ RTaxial
i1
þ RTaxial
iþ1

Ti ¼ 1
Rflow þ Rpipe 1þRh, i þ Raxial
2

For the last (exit) fluid node:

T
T N1
Rflow þ Rpipe þR
pN
h, i
þ RT N1
TN ¼ axial
1
Rflow þ Rpipe1þRh, i þ Raxial
1

This system of equations can be programmed into an iterative loop to solve for all
nodal temperatures, and then any desired quantity.
Consider the relative importance of the axial conduction and the axial advection:
Δx
_ p Δx
Raxial kf πðD2tÞ2 =4 4mc
¼ ¼
RFlow =m_ cp
1
πkf D2i
The axial conductive mechanism becomes more important than the advective flow
resistance when the spacing between nodes becomes sufficiently small.
Workshop 11.3. Solar Pipe in Continuous Mode: Multi-node Model 443
Heat Exchanger Fundamentals
12

The previous chapter focused on the flow of a single stream that exchanges heat
with its environment. This chapter considers the exchange of heat between two
streams at different temperatures separated by a solid barrier. The aim is to develop
a simplified modeling approach that can be used as a preliminary design tool. A
closed-form 2-node model is developed that captures the key relationships between
fluid types, their flow rates and temperatures, and the heat transfer surface between
them. A more detailed model of a specific design configuration (a tube in a tube, or
double-pipe) is developed as a case study for a more detailed investigation.
There are many different configurations in engineering practice, and there are
numerous resources available to obtain more comprehensive treatments of heat
exchangers. In the end, practitioners and designers must learn how to interpret
manufacturer specifications, which can be presented in various forms. A good
fundamental understanding of the key principles involved facilitates that practice.

12.1 Functional Requirements of a Heat Exchanger

A heat exchanger can be defined as an engineering device whose primary function is

to transfer heat from one fluid to another. Heat exchangers are hidden everywhere.
For example, nearly every household refrigerator has two heat exchangers. An
evaporator (usually hidden from view) is located inside the refrigerated space and
is designed so that heat from the cold space will transfer across a barrier to an even
colder refrigerant that flows through it. A condenser (usually placed in back of or
under the refrigerator) is designed to transfer heat from a hot refrigerant being
pumped through a cycle to ambient air. The heat rejected by the condenser consists
of the sum of the heat picked up by the evaporator, and the work added to the
refrigerant as it is compressed from a low pressure vapor to a high pressure vapor.
As another example, in nearly every automobile, the engine block is cooled by a
liquid coolant (termed “water” cooled, although it is not water) that flows into
spaces in the engine block to keep it from overheating. The engine block itself is

# Springer International Publishing Switzerland 2015 445

G. Sidebotham, Heat Transfer Modeling: An Inductive Approach,
DOI 10.1007/978-3-319-14514-3_12
446 12 Heat Exchanger Fundamentals

technically not a heat exchanger because its primary function is to convert chemical
energy into mechanical energy, not to transfer heat from one fluid to another. On the
other hand, the heat picked up by the coolant must be “dissipated” somehow, and
therefore the radiator is a true heat exchanger that transfers the heat picked up in the
engine block to ambient air. The hot coolant that leaves the engine block enters the
radiator and passes through a long pipe (not straight but doubled back and forth)
that has heat transfer fins placed on it (to increase the effective overall heat transfer
coefficient). Air is blown across the fins and heat is transferred from the hot coolant,
through the fins and to ambient. The coolant exits the radiator colder than its inlet
(but hotter than ambient) and returns to perform its main function of keeping the
engine block cool.
Other examples include heat exchangers designed to create high pressure steam
from a high temperature source fluid (i.e., from hot combustion products). These are
widely used for the generation of mechanical power (which can be used to drive an
electric generator) or merely as a means of distributing heat to an industrial process.
After passing through a turbine, the low pressure steam must be condensed in
another heat exchanger that uses ambient air, or river water as the second fluid.
Cooling towers involve evaporation of some condensing water.
A schematic of a typical “water in tube” boiler design is shown in Fig. 12.1.
In contrast, a “fire tube” boiler is one in which hot combustion products flow inside
tubes that are surrounded by liquid water which cause boiling on the outside
surface. The left sketch shows an end view, with hot “air” (i.e., combustion
products are hot, and have thermal properties nearly identical to those of air)
entering from below and encountering a bank of tubes that contain water at a
lower temperature. Heat is transferred from the gas to the tubes and the gases exit
the top at a lower temperature. How much lower depends on many factors such as
thermal properties of both fluids, flow rates of both streams, geometry and thermal
properties of the walls that separate the two fluid streams. The right sketch shows a
side view, with water entering each tube row as a liquid and emerging from the right
as a vapor. In practice, each tube could be delivered fluid with the same inlet
conditions (and the tubes are connected in parallel, as shown in Fig. 12.1), or the
tubes could be connected in series into a single long pipe with multiple passes (with
connections outside the field of view).

Warm “Air” Warm “Air”

H2o(L) H2o(V)

H2o(L) H2o(V)

H2o(L) H2o(V)

Hot“Air” Hot“Air”

Fig. 12.1 Schematic of a boiler section consisting of a bank of tubes in a cross-flow of hot gases
12.2 Double-Pipe Heat Exchangers 447

In the boiler example considered in the previous chapter, the outer fluid was
considered to be at a single temperature (T1) along the entire length of the pipe.
The inner fluid enthalpy changed with position. This case is considered a heat
exchanger, in the sense that heat is exchanged from one fluid to another across a
wall. However, the temperature of only one stream changed with position (for the
preheater and superheater), and the outer fluid temperature was constant along the
entire length of the duct. In contrast, both fluids in a heat exchanger often experi-
ence a significant change in temperature with position, each with an inlet tempera-
ture, and a different outlet temperature. This chapter considers means to estimate
the performance of such engineering devices. There are many designs, but most of
them can be analyzed by the methods developed here, at least approximately.
Broadly, heat exchanger problems are often formulated by specifying all four
temperatures (two inlets, two outlets) and the desired heat transfer rate, and then it
becomes a matter of determining required flow rates, or the size of the heat
exchanger. Alternatively, the flow rates, inlet temperatures, and the size of the
heat exchanger can be specified, and the heat transfer rate and exit temperatures
need to be determined. The key parameter that defines the size of the heat exchanger
is the surface area that separates the two fluids. A large surface area will bring one
or both of the fluids to near thermal equilibrium (thermally long behavior). A small
surface area will exchange heat, but not substantially change the temperature
(thermally short behavior).

12.2 Double-Pipe Heat Exchangers

A schematic representation of the so-called double-pipe heat exchangers

(a common configuration) is shown in Fig. 12.2. In both the co-flow and counter-
flow arrangements, one fluid (A) enters a pipe at an inlet temperature (TAin),
receives (or rejects) heat, and exits at a different temperature (TAout). This inner
fluid is like those considered in the previous chapter, except that the temperature of
the fluid outside the inner pipe varies with position. Fluid B enters through a larger,
concentric pipe and flows through an annular space. The Second Law of Thermo-
dynamics (heat flows from hot to cold) requires that the hotter fluid must be at a
higher temperature than the colder fluid at every location within the pipe and that
the temperature change along the path must be monotonic (increasing or decreas-
ing, or remaining constant for phase changing systems). These constraints yield
very different temperature profiles for the co-flow and counter-flow arrangements.
In the co-flow arrangement, fluid B enters from the same side as A and flows in
the same direction. Assuming that A is the hotter fluid, its temperature must
decrease monotonically in the direction of flow, as B increases along the same
direction. Therefore, the exiting temperature of A must be greater than the exiting
temperature of B. The sketch of temperature vs. position shows that the lowest
temperature of fluid A (at its exit) is greater than the highest temperature of fluid B
(at its exit).
448 12 Heat Exchanger Fundamentals

Fig. 12.2 Schematic representation of co-flow (left) and counter-flow (right) double-pipe heat
exchangers. Representative temperature profiles (temperature vs. position) are shown below each
schematic for non-phase changing operation

In the counter-flow arrangement, fluid B enters where fluid A exits, and vice
versa, and flows in the opposite direction. Assuming A is the hotter fluid, its
temperature must decrease monotonically from entrance to exit (left to right), as
B increases from its entrance to exit (right to left). At every location along the pipe,
A must be at a higher temperature. The Second Law restriction on A is that it must
exit at a temperature above the inlet of B (not the exit, as in the co-flow case).
Similarly, the exit of B must be below the inlet of A. That restriction yields a
temperature profile that is very different from the co-flow arrangement.
In the co-flow arrangement, the local temperature difference between A and B is
a maximum near the entrance, and decreases along the length of the pipe. In the
counter-flow arrangement, the temperature difference changes less, and in fact can
be a constant throughout the entire length of the pipe.

12.2.1 Averaged Temperature Differences

The heat transfer rate (q, in Watts) between the two fluid streams can be expressed
in a Newton’s Law of Cooling form, namely:


q ¼ U i Ai T A  T B

In this expression, the overall heat transfer coefficient Ui is based on the inner
surface area (Ai), but could be based on the outer (or any other surface area).
The temperature difference between the fluids is given as averages, and there are
different possible approaches. The most obvious and easily derived is the average
12.2 Double-Pipe Heat Exchangers 449

mean temperature difference (AMTD) in which the geometric average of the inlet
and outlet fluid temperatures are used. That is:
   
T Ain þ T Aout T Bin þ T Bout
q ¼ U i Ai AMTD ¼ U i Ai 
2 2

This result is the same for the co-flow and counter-flow arrangement. In the next
section, 2-node models will be developed that yield closed-form solutions for the exit
temperatures and heat transfer rates. A differential analysis, not developed here, yields
a similar expression for the heat transfer rate, but with a less intuitive expression for
the averaged temperature difference, called the log mean temperature difference
(LMTD). The LMTD is different for the co-flow and counter-flow arrangements:

q ¼ U i Ai LMTD

where


T A  T B ¼ LMTDCOFLOW
ðT Ain  T Bin Þ  ðT Aout  T Bout Þ
¼ h i for co‐flow heat exchangers
ln ððTTAout
Ain T Bin Þ
T Bout Þ


T A  T B ¼ LMTDCOUNTERFLOW
ðT Ain  T Bout Þ  ðT Aout  T Bin Þ
¼ h i for counter‐flow heat exchangers
ln ððTT Ain T Bout Þ
Aout T Bin Þ

The three different types of averages are shown for a specific case in Fig. 12.3. The
inlet temperature of A is set at 100  C and the inlet of B is set at 0  C. The exit of A
is set at 50  C, and the average temperature from the three forms is shown as a
function of all possible exit temperatures for B. It is apparent that the AWTD and
LMTDCOUNTER-FLOW give similar results, with significant deviations for the
LMTDCO-FLOW case.
For problems in which all four temperatures are specified, the averaged temper-
ature difference can be calculated upfront. Figure 12.3 shows that the AMTD,
which is easily remembered, can be used for counter-flow arrangements, and that
it is generally preferable to use the LMTD for co-flow arrangements.
The following general problem statement is written for those cases, where the inlet
temperatures and flow rates of both streams are specified, as is the effective heat
transfer coefficient Ui. The exit temperatures and heat transfer rates are to be deter-
mined as a function of total effective length (surface area/perimeter) of the pipe.
450 12 Heat Exchanger Fundamentals

100
Average Temperature Difference (oC) 90 TA,in = 100 ⬚C
TA,out = 50 ⬚C
80
TB, in = 0 ⬚C
70
60
50
40
30
AMTD
20
LMTD: COUNTER-FLOW
10
LMTD: CO-FLOW
0
0 20 40 60 80 100
Exit Temperature of B (oC)

Fig. 12.3 Comparison of averaged temperature differences for the case, where fluid A enters at
100  C and exits at 50  C, and fluid B enters at 0  C. The average mean temperature difference
(AMTD) and two log mean temperature difference (LMTD) expressions are shown for all possible
exit temperatures of fluid B. Note that for the co-flow arrangement fluid B cannot exit at a higher
temperature than the exit of A (50  C in this case)

12.2.1.1 Problem Statement

In a double-pipe heat exchanger (Fig. 12.2) of total length L, fluid A (with density
and specific heat) enters a pipe (inner diameter Dinner, wall thickness t, thermal
conductivity k) with a volumetric flow rate (AV)A at a temperature TA,in. In the
annular region of a concentric pipe with inner diameter D, fluid B (with density and
specific heat) enters with a volumetric flow rate (AV)B at a temperature TB,in.
Neglect stray heat loss between the outer pipe and its ambient and assume no
phase change in either stream. Calculate, as a function of pipe length, the exit
temperatures of the two streams, the internal heat transfer rate between the two
streams, and the heat exchanger effectiveness (defined as the actual heat transfer
rate normalized by the maximum possible heat transfer rate, where one fluid exits at
the inlet temperature of the other, and the other fluid does not overshoot the inlet).
Consider both co-flow and counter-flow arrangements.

12.2.2 2-Node Analysis: Well-Mixed Model

Invoking a 1-node model for each fluid stream, and using a well-mixed model, a
closed-form solution for the exit temperatures can be obtained. A thermal resistance
model (Fig. 12.4) indicates that heat exchange between the two fluids takes place
12.2 Double-Pipe Heat Exchangers 451

Fig. 12.4 Thermal resistance network for the 2-node heat exchanger model

with an overall heat transfer coefficient (U ) based on a defined area of the wall
separating the streams. The overall heat transfer coefficient for the double-pipe
arrangement, referenced to the inner surface area of the pipe is defined by the three
thermal resistances in series between the two fluids (convection at the inner wall,
conduction across the wall of thickness t, and convection at the outer wall):

1 1
Ui πDi L ¼ ¼  
Rinner þ Rwall þ Router 1
þ t
þ 1
hinner πDi L kπ ðDi þtÞL houter π ðDi þ2tÞL

where the thin-wall approximation has been used (the exact formula for conduction
across a cylindrical shell could be used instead). Solving for the overall heat transfer
coefficient:
1
Ui ¼ 1
hinner þ tDi
kðDi þtÞ þ houter ðDDi i þ2tÞ

Nusselt number correlations can be used for the convection coefficients,

recognizing that the outer flow region is an annular region and the Reynolds number
is based on an equivalent hydraulic diameter (that gives the same cross-sectional
area as the annular region):

2
πDH2 π D 2
 ð D i þ 2tÞ
¼
4 4
Solving for the hydraulic diameter:
qffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
DH ¼ D2  ðDi þ 2tÞ2
452 12 Heat Exchanger Fundamentals

Nodal equations applied to nodes A and B are:


m_ A cpA ðT Ain  T Aout Þ þ U i Ai T B  T A ¼ 0


m_ B cpB ðT Bin  T Aout Þ þ U i Ai T A  T B ¼ 0

The temperatures in the heat transfer terms represent the average temperatures in
each section. Treating the two inlet temperatures, mass flow rates and specific heats
as inputs, there are four unknown temperatures for these two equations (two exit
temperatures and two average temperatures).
Invoking the well-mixed model, the average temperature of the fluid in the pipe
is set equal to the exit temperature, adding two more equations to close the set. This
model will underestimate the average temperature difference between the
two streams, and therefore the model underestimates the total heat transfer rate.
Dividing each equation by the mass flow rate and specific heat, and rearranging:
   
U i Ai U i Ai
T Aout 1þ ¼ T Ain þ T Bout
m_ A cpA m_ A cpA
   
U i Ai U i Ai
T Bout 1 þ ¼ T Bin þ T Aout
m_ B cpB m_ B cpB

It is convenient and insightful to define two dimensionless lengths as the physical

length (L) normalized by thermal length for each stream:

U i Ai U i πDi L L
L^ A ¼ ¼ ¼
m_ A cpA m_ A cpA Lt, A

U i Ai U i πDi L L
L^ B ¼ ¼ ¼
m_ B cpB m_ B cpB Lt, B

Notice that there are two thermal lengths, one for each stream. The longer of the two
(with the lower thermal flow rate) is considered the controlling stream because it is
the closest to achieving thermal equilibrium with its partner, and its value is
referred in practice as number of transfer units (NTU). Combining the two nodal
equations:
 


^

^ T Bin þ L^ B T Aout
T Aout 1 þ L A ¼ T Ain þ L A
1 þ L^ B
Rearranging:
   
^ L^ B L^ B L^ A
T Aout 1 þ L A  ¼ T Ain þ T Bin
1 þ L^ B 1 þ L^ B
12.2 Double-Pipe Heat Exchangers 453

Finally, the exit temperature

 of fluid A is expressed in terms of input
 parameters:

^ ^
T Ain þ 1þL LA^ T Bin T Bin þ 1þL LB^ T Ain
T Aout ¼  B
. Similarly for fluid B:T Bout ¼  A
.
^ L^ A L^ B ^ L^ A L^ B
1 þ L A  1þL^ 1 þ L B  1þ L^
B A

The heat transfer rate between the two streams is obtained from one of three
forms:

qA!B ¼ m_ A cpA ðT Ain  T Aout Þ ¼ m_ B cpB ðT Bout  T Bin Þ ¼ U i Ai ðT Aout  T Bout Þ

The heat exchanger effectiveness is defined as the actual heat transfer rate
normalized by the maximum possible heat transfer rate. The maximum possible
heat transfer rate of either stream is that which would occur if the stream exited at
the inlet temperature of the other stream. That is:

qmax, A ¼ m_ A cpA ðT Ain  T Bin Þ

qmax, B ¼ m_ B cpB ðT Ain  T Bin Þ

The quantity (ṁcp) is considered to be a thermal mass flow rate. The smaller of
these defines the maximum possible heat transfer rate. A higher heat transfer rate
would cause the other stream to overshoot the inlet of the other, in violation of the
second Law. That is:


_ p min ðT Ain  T Bin Þ
qmax ¼ mc

The heat exchanger effectiveness is therefore:

q
ε¼
qmax

This 2-node model can give surprisingly good results, and it can be applied to any
heat exchanger type (when properly interpreted). However, it cannot distinguish
between co-flow and counter-flow behavior. The well-mixed model was used
because it works well across all ranges of flow. If, as is common, the inlet and
exit temperatures and mass flow rates of one stream are specified, then the geomet-
ric average temperature can be used instead, and then the required mass flow of the
other stream and the effective UA required can be determined more directly.
The NTU method is commonly used in design calculations, where NTU is a
dimensionless measure of the effective length of a double-pipe heat exchanger,
defined as the longer of the two thermal lengths:

UA
NTU ¼
 ¼ L^ max
_ p min
mc
454 12 Heat Exchanger Fundamentals

1
0.9
(m. cp )min
Heat Exchanger Effectiveness,ε
0.8 C= =0
(m. cp )max
0.7 0.25
0.6 0.5
0.75
0.5
1
0.4
0.3
0.2
0.1
0
0 1 2 3 4 5 6
NTU

Fig. 12.5 Heat exchanger effectiveness vs. number of transfer units (NTU) (longer of the two
thermal lengths) for 2-node well-mixed heat exchanger for selected values of the ratio of the
thermal mass flow rate

The heat exchanger effectiveness is plotted against pipe length (as NTU) in
Fig. 12.5 for fixed ratios of the minimum to maximum thermal flow rates (C) for
the 2-node well-mixed heat exchanger model: Use of this plot will give good initial
estimates of heat exchanger performance, and reveal key trends that can guide
design decisions. For example, for a situation where the thermal mass flow rates of
the two streams are comparable (C ¼ 1), then there is no need to select a heat
exchanger with an NTU of greater than approximately 2. The uncertainty in
estimating UA is generally considerable in any case.
The results and temperature profiles for three cases are considered in which the
temperature of the hotter fluid enters at 100  C and the colder fluid enters at 0  C.
The calculated average temperatures for the appropriate model are shown below the
plot of temperature vs. position. In the first case (Fig. 12.6 with NTU ¼ 1, C ¼ 0),
the effectiveness is 0.5, and the co-flow and counter-flow results are identical. The
cold exhibits high flow, or thermally short, behavior. Its temperature does not
change in the heat exchanger, and the direction of flow does not influence the
response of the hotter fluid. For longer heat exchangers (higher NTU) but with
C ¼ 0, the exit temperature of the hotter fluid will be closer to the cold fluid, but the
shapes will be similar to this case.
Figure 12.7 shows a case with a longer heat exchanger (NTU ¼ 3) and with the
thermal mass flow rate of the hotter fluid half that of the colder fluid (C ¼ 0.5). In
this case, both fluid temperatures change, but the hotter fluid changes more due to
its lower thermal flow rate. The counter-flow temperature profiles are not parallel,
12.2 Double-Pipe Heat Exchangers 455

NTU= 1
C=0
ε= 0.5
CO-FLOW COUNTER-FLOW
x TA (oC) TB (oC) x TA (oC) TB (oC)
0 100 0 0 100 0
1 50 0 1 50 0

100 100
90 CO-FLOW 90 COUNTER-FLOW

Temperature (oC)
Temperature (oC)

80 80
70 70
60 60
50 50
40 40
30 30
20 20
10 10
0 0
0 0.5 1 0 0.5 1

Distance (dimensionless) Distance (dimensionless)

AMTD 75 oC AMTD 75 oC
LMTD 72.13 oC LMTD 72.13 oC

Fig. 12.6 Temperature profiles from the well-mixed 2-node model for the case of NTU ¼ 1
(thermal length of hotter fluid), and C ¼ 0 (infinite thermal flow rate of colder fluid, or thermal
length of zero)

and the restriction of the well-mixed model does not allow the exit of the cold fluid
to exceed the exit of the hotter.
Figure 12.8 shows a case with a very long heat exchanger (NTU ¼ 6) and with
equal thermal flow rates of the two streams (C ¼ 1). In this case, the exit streams of
both streams are close together. In the counter-flow case, the temperature profiles
are parallel (which will be true for all cases with C ¼ 1), and the temperature
difference is the same along the entire length of the pipe. The response of either
stream is that of a constant heat flux.
The well-mixed model forces the outlet temperature of each stream to be equal
to the average temperature. Therefore, the outlet temperature of the hotter stream
must be greater than the outlet temperature of the colder stream. For this reason, for
very long heat exchangers, the well-mixed model is better suited to the co-flow
arrangement than the counter-flow arrangement, in which case the exit of the colder
fluid can exceed the exit of the hotter fluid.
The model considered next is a better model for the counter-flow case, where the
thermal mass flow rates are comparable, but the length is sufficiently long that the
exit of the cold fluid is significantly higher than the inlet of the hotter fluid.
456 12 Heat Exchanger Fundamentals

NTU=3
C = 0.5
ε= 0.545
CO-FLOW COUNTER-FLOW
x TA (oC) TB (oC) x TA (oC) TB (oC)
0 100 0 0 100 27.27
1 45.45 27.27 1 45.45 0

100 100
90 CO-FLOW 90 COUNTER-FLOW

Temperature (oC)
Temperature (oC)

80 80
70 70
60 60
50 50
40 40
30 30
20 20
10 10
0 0
0 0.5 1 0 0.5 1

Distance (dimensionless) Distance (dimensionless)

AMTD 59.09 oC AMTD 59.09 oC

LMTD 47.99 oC LMTD 58.03 oC

Fig. 12.7 Temperature profiles from the well-mixed 2-node model for the case of NTU ¼ 3
(thermal length of hotter fluid), and C ¼ 0.5 (thermal flow rate of hotter fluid half that of colder fluid)

12.2.3 2-Node Analysis: Average Temperature Model

The well-mixed model underestimates the heat transfer rate, but has the advantage
of not violating the Second Law of Thermodynamics. It is expected to work better
for co-flow arrangements than counter-flow arrangements. As with the well-mixed
model, the heat transfer rate can be expressed three different ways, based on the
direct heat transfer, or as the thermodynamic change:




_ p A ðT Ain  T Aout Þ ¼ mc
q ¼ UA T A  T B ¼ mc _ p B ðT Bout  T Bin Þ

For the average temperature model, the temperature difference is expressed as the
geometric average of the temperature difference at the inlet and that at the outlet.
Expressing that difference in the form for the counter-flow arrangement:

1
T A  T B ¼ ½ðT Ain  T Bout Þ þ ðT Aout  T Bin Þ
2
Expressing this difference for the co-flow arrangement yields the same result.
Algebraic manipulation (a good exercise) results in the following expression for
12.2 Double-Pipe Heat Exchangers 457

NTU=6
C=1
ε= 0.461
CO-FLOW COUNTER-FLOW
x TA (oC) TB (oC) x TA (oC) TB (oC)
0 100 0 0 100 46.15
1 53.85 46.15 1 53.85 0

100 100
90 CO-FLOW 90 COUNTER-FLOW

Temperature (oC)
Temperature (oC)

80 80
70 70
60 60
50 50
40 40
30 30
20 20
10 10
0 0
0 0.5 1 0 0.5 1

Distance (dimensionless) Distance (dimensionless)

AMTD 53.84 oC AMTD 53.84 oC

LMTD 35.99 oC LMTD 53.86 oC

Fig. 12.8 Temperature profiles from the well-mixed 2-node model for the case of NTU ¼ 6 (thermal
length of hotter fluid) and C ¼ 1 (thermal flow rate of hotter fluid equal to that of colder fluid)

the outlet temperature of A in terms of the inlet temperatures and the two thermal
lengths, defined as with the well-mixed model.


2 þ L^ B  L^ A T Ain þ 2L^ A T Bin
T Aout ¼
2 þ L^ A þ L^ B þ L^ A L^ B


2  L^ B T Bin þ L^ B ðT Ain  T Aout Þ
T Bout ¼
2 þ L^ B
The outlet of B is expressed in terms of the outlet for A, which must therefore be
calculated first. The effectiveness of the heat exchanger as a function of the NTU
(longer of the two thermal lengths) for various ratios of the thermal flow rates (C) is
shown in Fig. 12.9 which exhibits the same general shape as the well-mixed model.
However, for lower values of C, the heat exchanger effectiveness exceeds unity, in
violation of the Second Law. This result should not be surprising, based on the
limiting case for a single stream model from the previous chapter. For that reason,
this model should be used with caution.
One temperature profile case is shown in Fig. 12.10, with NTU ¼ 3 and C ¼ 1.
The exit temperature of the hotter fluid is lower than the exit of the colder fluid. This
behavior violates the second Law for the co-flow arrangement but is acceptable for
the counter-flow arrangement.
 1.5
1.4
1.3
.
(m c p)min

Heat Exchanger Effectiveness, e

1.2 0.75
C= . =0
1.1 (m c p)max
0.25
1 0.5
0.9 1
0.8
0.7
0.6
0.5
0.4
0.3
0.2
0.1
0
0 1 2 3 4 5 6
NTU

Fig. 12.9 Heat exchanger effectiveness vs. NTU (longer of the two thermal lengths) for 2-node
average temperature heat exchanger model for selected values of the ratio of the thermal mass
flow rate. Cases that exhibit an effectiveness greater than unity violate the Second Law of
Thermodynamics

NTU=3
C=1
ε= 0.882
CO-FLOW COUNTER-FLOW
x TA (oC) TB (oC) x TA (oC) TB (oC)
0 100 0 0 100 52.94
1 11.76 52.94 1 11.76 0

100 100
90 CO-FLOW 90 COUNTER-FLOW
Temperature (oC)
Temperature (oC)

80 80
70 70
60 60
50 50
40 40
30 30
20 20
10 10
0 0
0 0.5 1 0 0.5 1

Distance (dimensionless) Distance (dimensionless)

AMTD 29.4 oC AMTD 29.4 oC

LMTD NaN oC LMTD 25.46 oC

Fig. 12.10 Temperature profiles from the average temperature 2-node model for the case of
NTU ¼ 3 (thermal length of hotter fluid ¼ 3) and C ¼ 1 (thermal flow rate of hotter fluid equal to
that of colder fluid). The co-flow case violates the Second Law of Thermodynamics
12.2 Double-Pipe Heat Exchangers 459

A general conclusion or recommendation is that the well-mixed model can be

used for all cases to establish the basic performance of most, if not all, heat
exchangers. The average temperature model gives improved accuracy in certain
areas (short heat exchangers), but can give results that violate the Second Law. An
important practical case where the average temperature model is advantageous is
for counter-flow arrangements with thermally long pipes with comparable thermal
flow rates of the two streams. This arrangement is common in practice.

12.2.4 2n-Node Model

More quantitative accuracy and a distinction between co-flow and counter-flow

arrangements is obtained by discretizing space, as expected by now. The pipe is
broken into n equally spaced sections of length Δx ¼ L/n. A resistance model for an
interior cell “i” is shown in Fig. 12.11 for both the co-flow and counter-flow
arrangements. There is a node for each fluid (A and B). An energy balance applied
to node A(i) using the well-mixed model (in which the exit temperature from a cell

Fig. 12.11 Thermal resistance model for node “i” in the co-flow (top) and counter-flow (bottom)
arrangements. The well-mixed model is used (in which the average temperature of each fluid is set
equal to the exit temperature from that cell)
460 12 Heat Exchanger Fundamentals

equals the average temperature used to calculate the heat exchange between the
corresponding fluid B cell) is:

m_ A cpA ðT A, i1  T Ai Þ þ U i πDi ΔxðT Bi  T Ai Þ ¼ 0

where Ui is an overall heat transfer coefficient based on the inner surface area.
Rearranging for TAi (a form suitable for Method of Successive Substitution, with
superscript “k” to indicate iteration number):

ðk Þ ðkÞ
ðkþ1Þ m_ A cpA T A, i1 þ U i πDi ΔxT Bi
T Ai ¼
m_ A cpA þ U i πDi Δx

Similarly, an energy balance applied to cell B(i) yields:

ðkÞ ðkÞ
ðkþ1Þ m_ B cpB T B, i1 þ U i πDi ΔxT Ai
T Bi ¼ ðco‐flow arrangementÞ
m_ B cpB þ Ui πDi Δx

For the counter-flow arrangement, with fluid A flowing left to right and B flowing
right to left, the nodal equation for fluid B must recognize that the flow term
entering the cell comes from the right (TB+1) and the nodal equation is:

ðkÞ ðkÞ
ðkþ1Þ m_ B cpB T B, iþ1 þ Uinner πDinner ΔxT Ai
T Bi ¼ ðcounter‐flow arrangementÞ
m_ B cpB þ U inner πDinner Δx

Workshop 12.1. Double-Pipe Heat Exchangers

The following instructions for an in-class workshop were followed to create an

excel file. The base case input parameter values are for a 1 m long water to water
heat exchanger. The flow rates were adjusted (using “goal seek”) so that the thermal
lengths of both streams were set to unity. What follows are results for several cases,
varying pipe length, flow rates of one stream and substituting air for water. These
results are not explicitly discussed. Rather, the reader is encouraged to practice
technical writing skills by describing the trends observed for each case shown and
to run different case studies.

Suggested Spreadsheet Solution

• In a worksheet entitled “params,” write a separate block for input parameters (from
problem statement) that a user can change, and a block for relevant derived
parameters. Add a block to display final results (that confirms convergence).
Consider inner and outer convection coefficients to be input parameters (but
separate worksheets could be used to calculate them based on Nusselt number
correlations).
Workshop 12.1. Double-Pipe Heat Exchangers 461

• In a worksheet1 entitled “Iterations_co-flow,” for the heat exchanger broken into ten
equal sized sections, define a row with ten columns for fluid A and ten columns for
fluid B. Enter the inlet fluid temperatures as initial “guessed” values (reference to
“params”). In the next row, enter the recursion formulas for the fluid temperatures
that reference the row above it. Add two columns to check for convergence of the exit
streams. Copy this iterative row as far as needed to obtain a reasonable convergence.
• Copy the “Iterations_co-flow” worksheet, rename it “Iterations_counter-flow”
and modify it for the counter-flow arrangement.
• In a worksheet entitled “Results” lay out a block for the heat exchanger with four
rows of temperatures (fluid A, inner wall, outer wall, fluid B) for co-flow.
Reference the fluid temperatures to the final converged values from “iterations”
and enter voltage divider formulas for the wall temperatures. Repeat for counter-
flow. Make well-formatted plots of all four temperatures vs. position for both
arrangements (Figs. 12.12, 12.13, 12.14, 12.15, 12.16, 12.17, 12.18, 12.19,
12.20, 12.21, 12.22, 12.23, 12.24, and 12.25).
• Use the worksheet as a design tool. . .
CO-FLOW ARRANGEMENT
Max Change = 0.00E+00
dx=0.1 m
node 1 2 3 4 5 6 7 8 9 10
x 0 0.05 0.15 0.25 0.35 0.45 0.55 0.65 0.75 0.85 0.95 1
TA 100.00 91.67 84.72 78.93 74.11 70.09 66.74 63.95 61.63 59.69 58.07
Tinner 40.44 42.03 43.36 44.47 45.39 46.16 46.80 47.33 47.78 48.15 48.46
Touter 39.71 41.42 42.85 44.05 45.04 45.87 46.55 47.13 47.61 48.01 48.34
TB 0.00 8.33 15.28 21.07 25.89 29.91 33.26 36.05 38.37 40.31 41.93

COUNTER-FLOW ARRANGEMENT
Max Change = 0.00E+00
dx=0.1 m
node 1 2 3 4 5 6 7 8 9 10
x 0 0.05 0.15 0.25 0.35 0.45 0.55 0.65 0.75 0.85 0.95 1
TA 100.00 95.24 90.48 85.71 80.95 76.19 71.43 66.66 61.90 57.14 52.38 52.38
Tinner 68.80 66.88 62.11 57.35 52.59 47.83 43.06 38.30 33.54 28.78 24.02 24.02
Touter 68.42 66.53 61.77 57.01 52.24 47.48 42.72 37.96 33.19 28.43 23.67 23.67
TB 47.62 47.62 42.86 38.10 33.34 28.57 23.81 19.05 14.29 9.52 4.76 0.00

Fig. 12.12 Worksheet entitled “Results” which checks for convergence, reports the converged
fluid temperatures and calculates the surface temperatures of the inner pipe

A A A A A A A A A A B B B B B B B B B B
check A check B 1 2 3 4 5 6 7 8 9 10 1 2 3 4 5 6 7 8 9 10
100 100 100 100 100 100 100 100 100 100 0 0 0 0 0 0 0 0 0 0
9.092581 9.092581 90.9 90.9 90.9 90.9 90.9 90.9 90.9 90.9 90.9 90.9 9.1 9.1 9.1 9.1 9.1 9.1 9.1 9.1 9.1 9.1
7.43908 7.43908 91.7 83.5 83.5 83.5 83.5 83.5 83.5 83.5 83.5 83.5 8.3 16.5 16.5 16.5 16.5 16.5 16.5 16.5 16.5 16.5
6.086271 6.086271 91.7 84.9 77.4 77.4 77.4 77.4 77.4 77.4 77.4 77.4 8.3 15.1 22.6 22.6 22.6 22.6 22.6 22.6 22.6 22.6
4.979473 4.979473 91.7 84.7 79.2 72.4 72.4 72.4 72.4 72.4 72.4 72.4 8.3 15.3 20.8 27.6 27.6 27.6 27.6 27.6 27.6 27.6
4.073948 4.073948 91.7 84.7 78.9 74.5 68.3 68.3 68.3 68.3 68.3 68.3 8.3 15.3 21.1 25.5 31.7 31.7 31.7 31.7 31.7 31.7
3.333094 3.333094 91.7 84.7 78.9 74.0 70.6 65.0 65.0 65.0 65.0 65.0 8.3 15.3 21.1 26.0 29.4 35.0 35.0 35.0 35.0 35.0
2.726965 2.726965 91.7 84.7 78.9 74.1 70.0 67.4 62.3 62.3 62.3 62.3 8.3 15.3 21.1 25.9 30.0 32.6 37.7 37.7 37.7 37.7
2.231062 2.231062 91.7 84.7 78.9 74.1 70.1 66.6 64.7 60.0 60.0 60.0 8.3 15.3 21.1 25.9 29.9 33.4 35.3 40.0 40.0 40.0
1.82534 1.82534 91.7 84.7 78.9 74.1 70.1 66.8 63.7 62.5 58.2 58.2 8.3 15.3 21.1 25.9 29.9 33.2 36.3 37.5 41.8 41.8

Fig. 12.13 Worksheet entitled “iterations_co-flow” which shows the first few rows of the
iteration. A similar worksheet “iterations_counter-flow” is not shown

1
It is advised to set up this iteration table in the “params” worksheet, and once it is working and
debugged, cut and paste it to its own sheet. Alternatively, name key cells in “params” and use the
named cells in the formulas.
462 12 Heat Exchanger Fundamentals

INPUT PARAMETERS
L= 1 m pipe length
rhoA= 1000 kg/m^3 density of A
cpA = 4200 J/kg/K specific heat of A
rhoB= 1000 kg/m^3 density of B
cpB= 4200 J/kg/K specific heat of B
Dinner = 0.02 m inner diameter of inner pipe
t= 0.005 m pipe thickness
D= 0.2 m inner diameter of outer pipe
k= 40 W/m/K thermal conductivity of pipe
AV,A= 8.91E-06 m^3/s volumetric flow rate of A
AV,B= 8.91E-06 m^3/s volumetric flow rate of B
Tain= 100 oC inlet temperature of A
Tbin= 0 oC inlet temperature of B
hinner = 1000 W/m2/K inner convection coefficient
houter = 1000 W/m2/K outer convection coefficient

DERIVED PARAMETERS
dx= 0.1 m length of numerical segment (for 10 section grid)
Rinner = 0.159155 oC/W thermal resistance between A and inner wall
Rwall= 0.001941 oC/W thermal resistance across pipe wall
Router= 0.106103 oC/W thermal resistance between B and outer wall
Uinner= 595.6416 W/m2/K overall heat transfer coefficient
mdotA= 8.91E-03 kg/s mass flow rate of A
mdotB = 8.91E-03 kg/s mass flow rate of B
(mdot*cp)A= 3.74E+01 oC/W Flow capacitance of A
(mdot*cp)B= 3.74E+01 W/K Flow capacitance of B
LthermalA= 0.999798 m thermal length of A
LthermalB= 0.999798 m thermal length of B
A= 1.000202 L/LthermalA, dimensionless length of A
B= 1.000202 L/LthermalB, dimensionless length of B
qmaxA = 3741.77 W maximum possible heat transfer rate of A
qmaxB = 3741.77 W maximum possible heat transfer rate of B
qmax 3741.77 W maximum possible heat transfer rate

RESULTS (change = 0.00E+00 0.00E+00)

2-Node co-flow counter-flow
TAout= 66.7 58.1 52.4 oC exit temperature of A
TBout= 33.3 41.9 47.6 oC exit temperature of B
qA= -1247.3 -1568.8 -1782.0 W heat transfer rate to A
qB= 1247.3 1568.8 1782.0 W heat transfer rate to B
effectiveness= 0.333 0.419 0.476 heat exchanger effectiveness

NTU= UA/Cmin 1.00 "Number of Transfer Units" = Longer of the 2 thermal lengths
Cmin/Cmax= 1.00 ratio of thermal lengths
AV,A= 5.35E-01 LPM volumetric flow rate of A
AV,B= 5.35E-01 LPM volumetric flow rate of B
Lt,A= 0.999798 m Thermal length of A
Lt,B= 0.999798 m Thermal length of B
AREA= 0.062832 m^2 Inner Surface Area

Fig. 12.14 Parameters for the base case: 1 m long water to water heat exchanger with thermal
lengths equal unity
Workshop 12.1. Double-Pipe Heat Exchangers 463

120
TA
100 Tinner
Touter
80 TB
Temperature (oC)

60

40

20

0
0 0.2 0.4 0.6 0.8 1
Axial Position (m)

CO-FLOW

120

TA
100
Tinner
Touter
80
Temperature (oC)

TB

60

40

20

0
0 0.2 0.4 0.6 0.8 1
Axial Position (m)

COUNTER-FLOW

Fig. 12.15 Temperature profiles for base case water to water heat exchanger with unity thermal
lengths
464 12 Heat Exchanger Fundamentals

INPUT PARAMETERS
L= 2 m pipe length
rhoA= 1000 kg/m^3 density of A
cpA = 4200 J/kg/K specific heat of A
rhoB= 1000 kg/m^3 density of B
cpB= 4200 J/kg/K specific heat of B
Dinner = 0.02 m inner diameter of inner pipe
t= 0.005 m pipe thickness
D= 0.2 m inner diameter of outer pipe
k= 40 W/m/K thermal conductivity of pipe
AV,A= 8.91E-06 m^3/s volumetric flow rate of A
AV,B= 8.91E-06 m^3/s volumetric flow rate of B
Tain= 100 oC inlet temperature of A
Tbin= 0 oC inlet temperature of B
hinner = 1000 W/m2/K inner convection coefficient
houter = 1000 W/m2/K outer convection coefficient

DERIVED PARAMETERS
dx= 0.2 m length of numerical segment (for 10 section grid)
Rinner = 0.079577 oC/W thermal resistance between A and inner wall
Rwall= 0.00097 oC/W thermal resistance across pipe wall
Router= 0.053052 oC/W thermal resistance between B and outer wall
Uinner= 595.6416 W/m2/K overall heat transfer coefficient
mdotA= 8.91E-03 kg/s mass flow rate of A
mdotB = 8.91E-03 kg/s mass flow rate of B
(mdot*cp)A= 3.74E+01 oC/W Flow capacitance of A
(mdot*cp)B= 3.74E+01 W/K Flow capacitance of B
LthermalA= 0.999798 m thermal length of A
LthermalB= 0.999798 m thermal length of B
A= 2.000404 L/LthermalA, dimensionless length of A
B= 2.000404 L/LthermalB, dimensionless length of B
qmaxA = 3741.77 W maximum possible heat transfer rate of A
qmaxB = 3741.77 W maximum possible heat transfer rate of B
qmax 3741.77 W maximum possible heat transfer rate

RESULTS (change = 0.00E+00 0.00E+00)

2-Node co-flow counter-flow
TAout= 60.0 51.7 37.5 oC exit temperature of A
TBout= 40.0 48.3 62.5 oC exit temperature of B
qA= -1496.8 -1806.2 -2338.8 W heat transfer rate to A
qB= 1496.8 1806.2 2338.8 W heat transfer rate to B
effectiveness= 0.400 0.483 0.625 heat exchanger effectiveness

NTU= UA/Cmin 2.00 "Number of Transfer Units" = Longer of the 2 thermal lengths
Cmin/Cmax= 1.00 ratio of thermal lengths
AV,A= 5.35E-01 LPM volumetric flow rate of A
AV,B= 5.35E-01 LPM volumetric flow rate of B
Lt,A= 0.999798 m Thermal length of A
Lt,B= 0.999798 m Thermal length of B
AREA= 0.125664 m^2 Inner Surface Area

Fig. 12.16 Parameters and results for a pipe length twice that of the base case
Workshop 12.1. Double-Pipe Heat Exchangers 465

120
TA
100 Tinner
Touter
TB
80
Temperature (oC)

60

40

20

0
0 0.5 1 1.5 2
Axial Position (m)

CO-FLOW

120

100 TA
Tinner
Touter
80
Temperature (oC)

TB

60

40

20

0
0 0.5 1 1.5 2
Axial Position (m)

COUNTER-FLOW

Fig. 12.17 Temperature profiles for a pipe length twice as long as the base case
466 12 Heat Exchanger Fundamentals

INPUT PARAMETERS
L= 0.5 m pipe length
rhoA= 1000 kg/m^3 density of A
cpA = 4200 J/kg/K specific heat of A
rhoB= 1000 kg/m^3 density of B
cpB= 4200 J/kg/K specific heat of B
Dinner = 0.02 m inner diameter of inner pipe
t= 0.005 m pipe thickness
D= 0.2 m inner diameter of outer pipe
k= 40 W/m/K thermal conductivity of pipe
AV,A= 8.91E-06 m^3/s volumetric flow rate of A
AV,B= 8.91E-06 m^3/s volumetric flow rate of B
Tain= 100 oC inlet temperature of A
Tbin= 0 oC inlet temperature of B
hinner = 1000 W/m2/K inner convection coefficient
houter = 1000 W/m2/K outer convection coefficient

DERIVED PARAMETERS
dx= 0.05 m length of numerical segment (for 10 section grid)
Rinner = 0.31831 oC/W thermal resistance between A and inner wall
Rwall= 0.003882 oC/W thermal resistance across pipe wall
Router= 0.212207 oC/W thermal resistance between B and outer wall
Uinner= 595.6416 W/m2/K overall heat transfer coefficient
mdotA= 8.91E-03 kg/s mass flow rate of A
mdotB = 8.91E-03 kg/s mass flow rate of B
(mdot*cp)A= 3.74E+01 oC/W Flow capacitance of A
(mdot*cp)B= 3.74E+01 W/K Flow capacitance of B
LthermalA= 0.999798 m thermal length of A
LthermalB= 0.999798 m thermal length of B
A= 0.500101 L/LthermalA, dimensionless length of A
B= 0.500101 L/LthermalB, dimensionless length of B
qmaxA = 3741.77 W maximum possible heat transfer rate of A
qmaxB = 3741.77 W maximum possible heat transfer rate of B
qmax 3741.77 W maximum possible heat transfer rate

RESULTS (change = 0.00E+00 0.00E+00)

2-Node co-flow counter-flow
TAout= 75.0 69.3 67.7 oC exit temperature of A
TBout= 25.0 30.7 32.3 oC exit temperature of B
qA= -935.5 -1149.7 -1207.2 W heat transfer rate to A
qB= 935.5 1149.7 1207.2 W heat transfer rate to B
effectiveness= 0.250 0.307 0.323 heat exchanger effectiveness

NTU= UA/Cmin 0.50 "Number of Transfer Units" = Longer of the 2 thermal lengths
Cmin/Cmax= 1.00 ratio of thermal lengths
AV,A= 5.35E-01 LPM volumetric flow rate of A
AV,B= 5.35E-01 LPM volumetric flow rate of B
Lt,A= 0.999798 m Thermal length of A
Lt,B= 0.999798 m Thermal length of B
AREA= 0.031416 m^2 Inner Surface Area

Fig. 12.18 Parameter values for a pipe length half that of the base case
Workshop 12.1. Double-Pipe Heat Exchangers 467

120
TA
100 Tinner
Touter
TB
80
Temperature (oC)

60

40

20

0
0 0.1 0.2 0.3 0.4 0.5
Axial Position (m)

CO-FLOW

120
TA
100 Tinner
Touter
TB
80
Temperature (oC)

60

40

20

0
0 0.1 0.2 0.3 0.4 0.5
Axial Position (m)

COUNTER-FLOW

Fig. 12.19 Temperature profiles for a pipe length half that of the base case
468 12 Heat Exchanger Fundamentals

INPUT PARAMETERS
L= 1 m pipe length
rhoA= 1000 kg/m^3 density of A
cpA = 4200 J/kg/K specific heat of A
rhoB= 1000 kg/m^3 density of B
cpB= 4200 J/kg/K specific heat of B
Dinner = 0.02 m inner diameter of inner pipe
t= 0.005 m pipe thickness
D= 0.2 m inner diameter of outer pipe
k= 40 W/m/K thermal conductivity of pipe
AV,A= 8.91E-06 m^3/s volumetric flow rate of A
AV,B= 8.91E-05 m^3/s volumetric flow rate of B
Tain= 100 oC inlet temperature of A
Tbin= 0 oC inlet temperature of B
hinner = 1000 W/m2/K inner convection coefficient
houter = 1000 W/m2/K outer convection coefficient

DERIVED PARAMETERS
dx= 0.1 m length of numerical segment (for 10 section grid)
Rinner = 0.159155 oC/W thermal resistance between A and inner wall
Rwall= 0.001941 oC/W thermal resistance across pipe wall
Router= 0.106103 oC/W thermal resistance between B and outer wall
Uinner= 595.6416 W/m2/K overall heat transfer coefficient
mdotA= 8.91E-03 kg/s mass flow rate of A
mdotB = 8.91E-02 kg/s mass flow rate of B
(mdot*cp)A= 3.74E+01 oC/W Flow capacitance of A
(mdot*cp)B= 3.74E+02 W/K Flow capacitance of B
LthermalA= 0.999798 m thermal length of A
LthermalB= 9.997978 m thermal length of B
A= 1.000202 L/LthermalA, dimensionless length of A
B= 0.10002 L/LthermalB, dimensionless length of B
qmaxA = 3741.77 W maximum possible heat transfer rate of A
qmaxB = 37417.7 W maximum possible heat transfer rate of B
qmax 3741.77 W maximum possible heat transfer rate

RESULTS (change = 0.00E+00 0.00E+00)

2-Node co-flow counter-flow
TAout= 52.4 41.1 40.0 oC exit temperature of A
TBout= 4.8 5.9 6.0 oC exit temperature of B
qA= -1782.0 -2203.9 -2244.0 W heat transfer rate to A
qB= 1782.0 2203.9 2244.0 W heat transfer rate to B
effectiveness= 0.476 0.589 0.600 heat exchanger effectiveness

NTU= UA/Cmin 1.00 "Number of Transfer Units" = Longer of the 2 thermal lengths
Cmin/Cmax= 0.10 ratio of thermal lengths
AV,A= 5.35E-01 LPM volumetric flow rate of A
AV,B= 5.35E+00 LPM volumetric flow rate of B
Lt,A= 0.999798 m Thermal length of A
Lt,B= 9.997978 m Thermal length of B
AREA= 0.062832 m^2 Inner Surface Area

Fig. 12.20 Parameter values with fluid B having ten times the flow rate of the base case
Workshop 12.1. Double-Pipe Heat Exchangers 469

120
TA
100 Tinner
Touter
TB
80
Temperature (oC)

60

40

20

0
0 0.2 0.4 0.6 0.8 1
Axial Position (m)

CO-FLOW

120

100 TA
Tinner
80 Touter
Temperature (oC)

TB
60

40

20

0
0 0.2 0.4 0.6 0.8 1
Axial Position (m)

COUNTER-FLOW

Fig. 12.21 Temperature profiles with fluid B having ten times the flow rate as the base case
470 12 Heat Exchanger Fundamentals

INPUT PARAMETERS
L= 1 m pipe length
rhoA= 1000 kg/m^3 density of A
cpA = 4200 J/kg/K specific heat of A
rhoB= 1000 kg/m^3 density of B
cpB= 4200 J/kg/K specific heat of B
Dinner = 0.02 m inner diameter of inner pipe
t= 0.005 m pipe thickness
D= 0.2 m inner diameter of outer pipe
k= 40 W/m/K thermal conductivity of pipe
AV,A= 8.91E-06 m^3/s volumetric flow rate of A
AV,B= 8.91E-07 m^3/s volumetric flow rate of B
Tain= 100 oC inlet temperature of A
Tbin= 0 oC inlet temperature of B
hinner = 1000 W/m2/K inner convection coefficient
houter = 1000 W/m2/K outer convection coefficient

DERIVED PARAMETERS
dx= 0.1 m length of numerical segment (for 10 section grid)
Rinner = 0.159155 oC/W thermal resistance between A and inner wall
Rwall= 0.001941 oC/W thermal resistance across pipe wall
Router= 0.106103 oC/W thermal resistance between B and outer wall
Uinner= 595.6416 W/m2/K overall heat transfer coefficient
mdotA= 8.91E-03 kg/s mass flow rate of A
mdotB = 8.91E-04 kg/s mass flow rate of B
(mdot*cp)A= 3.74E+01 oC/W Flow capacitance of A
(mdot*cp)B= 3.74E+00 W/K Flow capacitance of B
LthermalA= 0.999798 m thermal length of A
LthermalB= 0.09998 m thermal length of B
A= 1.000202 L/LthermalA, dimensionless length of A
B= 10.00202 L/LthermalB, dimensionless length of B
qmaxA = 3741.77 W maximum possible heat transfer rate of A
qmaxB = 374.177 W maximum possible heat transfer rate of B
qmax 374.177 W maximum possible heat transfer rate

RESULTS (change = 1.42E-14 0.00E+00)

2-Node co-flow counter-flow
TAout= 91.7 90.9 90.0 oC exit temperature of A
TBout= 83.3 90.9 99.8 oC exit temperature of B
qA= -311.8 -340.0 -373.3 W heat transfer rate to A
qB= 311.8 340.0 373.3 W heat transfer rate to B
effectiveness= 0.833 0.909 0.998 heat exchanger effectiveness

NTU= UA/Cmin 10.00 "Number of Transfer Units" = Longer of the 2 thermal lengths
Cmin/Cmax= 0.10 ratio of thermal lengths
AV,A= 5.35E-01 LPM volumetric flow rate of A
AV,B= 5.35E-02 LPM volumetric flow rate of B
Lt,A= 0.999798 m Thermal length of A
Lt,B= 0.09998 m Thermal length of B
AREA= 0.062832 m^2 Inner Surface Area

Fig. 12.22 Parameter values with the flow rate of B ten times lower than the base case
Workshop 12.1. Double-Pipe Heat Exchangers 471

120

100

80
Temperature (oC)

TA
60
Tinner
Touter
40 TB

20

0
0 0.2 0.4 0.6 0.8 1
Axial Position (m)

CO-FLOW

120

100

80
Temperature (oC)

TA
60
Tinner
Touter
40 TB

20

0
0 0.2 0.4 0.6 0.8 1
Axial Position (m)

COUNTER-FLOW

Fig. 12.23 Temperature profiles with the flow rate of B ten times lower than the base case
472 12 Heat Exchanger Fundamentals

INPUT PARAMETERS
L= 1 m pipe length
rhoA= 0.946278 kg/m^3 density of A
cpA = 1005 J/kg/K specific heat of A
rhoB= 1000 kg/m^3 density of B
cpB= 4200 J/kg/K specific heat of B
Dinner = 0.02 m inner diameter of inner pipe
t= 0.005 m pipe thickness
D= 0.2 m inner diameter of outer pipe
k= 40 W/m/K thermal conductivity of pipe
AV,A= 3.19E-03 m^3/s volumetric flow rate of A
AV,B= 7.24E-07 m^3/s volumetric flow rate of B
Tain= 100 oC inlet temperature of A
Tbin= 0 oC inlet temperature of B
hinner = 50 W/m2/K inner convection coefficient
houter = 1000 W/m2/K outer convection coefficient

DERIVED PARAMETERS
dx= 0.1 m length of numerical segment (for 10 section grid)
Rinner = 3.183099 oC/W thermal resistance between A and inner wall
Rwall= 0.001941 oC/W thermal resistance across pipe wall
Router= 0.106103 oC/W thermal resistance between B and outer wall
Uinner= 48.35856 W/m2/K overall heat transfer coefficient
mdotA= 3.02E-03 kg/s mass flow rate of A
mdotB = 7.24E-04 kg/s mass flow rate of B
(mdot*cp)A= 3.04E+00 oC/W Flow capacitance of A
(mdot*cp)B= 3.04E+00 W/K Flow capacitance of B
LthermalA= 0.999798 m thermal length of A
LthermalB= 1.000492 m thermal length of B
A= 1.000202 L/LthermalA, dimensionless length of A
B= 0.999508 L/LthermalB, dimensionless length of B
qmaxA = 303.7844 W maximum possible heat transfer rate of A
qmaxB = 303.9953 W maximum possible heat transfer rate of B
qmax 303.7844 W maximum possible heat transfer rate

RESULTS (change = 0.00E+00 0.00E+00)

2-Node co-flow counter-flow
TAout= 66.7 58.1 52.4 oC exit temperature of A
TBout= 33.3 41.9 47.6 oC exit temperature of B
qA= -101.3 -127.4 -144.7 W heat transfer rate to A
qB= 101.3 127.4 144.7 W heat transfer rate to B
effectiveness= 0.333 0.419 0.476 heat exchanger effectiveness

NTU= UA/Cmin 1.00 "Number of Transfer Units" = Longer of the 2 thermal lengths
Cmin/Cmax= 1.00 ratio of thermal lengths
AV,A= 1.92E+02 LPM volumetric flow rate of A
AV,B= 4.34E-02 LPM volumetric flow rate of B
Lt,A= 0.999798 m Thermal length of A
Lt,B= 1.000492 m Thermal length of B
AREA= 0.062832 m^2 Inner Surface Area

Fig. 12.24 Parameters with air as fluid A, and the flow rate adjusted to achieve a thermal length
of unity in both streams
Workshop 12.1. Double-Pipe Heat Exchangers 473

120
TA
100 Tinner
Touter
TB
80
Temperature (oC)

60

40

20

0
0 0.2 0.4 0.6 0.8 1
Axial Position (m)

CO-FLOW

120
TA
Tinner
100
Touter
TB
80
Temperature (oC)

60

40

20

0
0 0.2 0.4 0.6 0.8 1
Axial Position (m)
COUNTER-FLOW

Fig. 12.25 Temperature profiles with air as fluid A and both flow rates adjusted to achieve a
thermal length of unity
474 12 Heat Exchanger Fundamentals

Workshop 12.2. Serving Coffee at the Right Temperature

Motivation. A real revolution in the coffee bar experience would be if a customer

could specify the serving temperature of coffee. Coffee must be brewed at close to
the boiling point of water to obtain the desired chemical extraction, but that
temperature is generally too hot to drink (except for my mom). So, as a capstone
workshop to this textbook, a preliminary heat exchanger design is outlined whose
function is to cool coffee from its brewing temperature (defined as 95  C) to a
consumer-selected outlet temperature (ranging from, say 60–80  C). However,
rather than dissipating the excess heat to the environment, heat is transferred
from the coffee to feedwater which offsets the heat required in the brewing process.
The problem is set up to be a continuous process, which would be difficult to
execute in practice, as coffee flowing through a tap is generally intermittent. The
spirit of the exercise is to practice heat exchanger calculations while creating food
(rather, drink) for thought.
Problem statement. A double-pipe heat exchanger is used to cool coffee to a desired
serving temperature while preheating feedwater. Coffee at TAin ¼ 95  C steadily
enters the center pipe (3/800 K-type copper tubing, with inner diameter of 0.402 in.
and outer diameter 0.500 in.). Feedwater at TBin ¼ 15  C enters the annular region
(1/200 K-type copper tubing with inner diameter of 0.527 in. and outer diameter
0.625 in.) at a 20 % higher flow rate than the coffee (to account for evaporation and
water retention in the coffee beans). The outer pipe is insulated to minimize heat
loss. A valve with discrete settings controls the flow rate and thereby exit tempera-
ture of the coffee. The system is designed to deliver a 16 fluid ounce coffee at
TAout ¼ 80  C in 2 seconds considered to be the base case. Coffee at lower
temperatures (75, 70, 65, 60  C) is delivered by a valve setting that lowers the
flow rate to a specific corresponding value. The flow rate of the feedwater is 20 %
higher than the coffee flow rate (to account for evaporation and water absorbed in
coffee beans). Determine:

• The inner and outer convection coefficients for the base case.
• The outlet temperature of the feedwater for the base case.
• The length of the heat exchanger using the well-mixed model and the average
temperature model. Use the latter going forward.
• The filling times (for a 16 oz. coffee) and feedwater exit temperatures required
for the lower coffee serving temperatures.
• Be an entrepreneur and bring this system to market!

Bonus. To tie in concepts from earlier in the course, analyze what happens when the
flow rate is stopped, and the heat exchanger is full of coffee and feedwater. What is
the final equilibrium temperature, and how long does it take to achieve it. Note that
the volume of the annular region (feedwater) was deliberately chosen to be small
compared to the inner region (coffee) so that the coffee would not be TOO cold.
This bonus question should raise a lot of issues with the reliability of the practical
application of this idea.
Evaporation and Mass Transfer
Fundamentals 13

This chapter introduces fundamental mass transfer concepts through the modeling
of the evaporation process that takes place at the free surface of a mug of coffee. An
evaporative heat transfer coefficient, hevap, is defined that can be included in any of
the transient numerical simulations from earlier chapters (i.e., lumped 1-node, few
node, multi-node). Evaporation is driven by concentration gradients, not a temper-
ature difference, and defining an evaporation coefficient this way allows direct
comparison of the relative effects of evaporation, convection, and radiation on the
cooling of the free surface. As will become apparent, this approach makes sense
only for cases when the liquid surface temperature differs significantly from
ambient. The case where the coffee reaches thermal equilibrium with the environ-
ment and the slower evaporation continues is treated differently, and it will be seen
that the coffee attains a temperature below ambient, a form of evaporative cooling.
Specifically, coffee (treated as water and in a quasi-steady state) is at an average
temperature Tcoffee in a cylindrical container of inner diameter D, with a free
surface a distance L below the top rim. Ambient air is at a temperature T1 with a
relative humidity rh. The goal is to estimate the rate of energy transport by the
evaporation process in a Newton’s Law of Cooling form, that is:

qevap ¼ hevap SðT s  T 1 Þ

where S is the surface area, and Ts is the surface temperature of the water. Compare,
as a function of coffee temperature, the magnitude of this mechanism to the
convection and radiation processes that act in parallel.

13.1 Background

The free surface of the coffee is the only fluid/fluid interface in the cooling mug of
coffee scenario. Something different happens there than at fluid/solid interfaces,
namely, evaporation. Technically, evaporation is not heat transfer; it is mass transfer.

# Springer International Publishing Switzerland 2015 475

G. Sidebotham, Heat Transfer Modeling: An Inductive Approach,
DOI 10.1007/978-3-319-14514-3_13
476 13 Evaporation and Mass Transfer Fundamentals

Heat transfer is energy that crosses a boundary due to a temperature difference.

Mass transfer is mass that crosses a boundary due to molecular diffusion driven by a
concentration gradient. That mass has energy associated with it, and therefore,
thermal consequences. It is a fundamentally different mechanism than conduction
or radiation. It is also a fundamentally different mode of energy transfer than
advection, in which case energy flows into a control volume with the bulk mass.
Empirically, the rate of evaporation appears to be much slower than the rate of
cooling. The coffee reaches room temperature without any noticeable change in
coffee volume (barring sipping, of course). However, there is phase change
involved, and a small amount of mass in the liquid phase that exits at a much
higher energy level as vapor is more important from an energy perspective than
from a mass perspective.
The evaporation process is influenced by humidity, the familiar weather term.
The absolute humidity, not reported in weather reports, is generally defined as the
mass of water vapor in a given volume of humid air. In weather reports, humidity is
given as a percentage, called the relative humidity. Figure 13.1 shows a schematic
snapshot of a sample of humid air, a term which means that there is air (small dots)
with a few water molecules dispersed in it (big dots). If a snapshot were taken a
short time after this one, the picture would look the same, but every molecule would
be in a different place. All the molecules migrate, performing a random walk, but
there is no net drift. At any one time, if a molecule moves up, another is moving
down. In reality, the water molecules are a little smaller than air molecules, but they
are shown as big dots because they are the main focus of the present discussion.
Air, with all water vapor removed, is a mixture of fixed component species
consisting of 78.0840 mol% N2, 20.9460 % O2, 0.9340 % Argon, 0.0390 % CO2

Fig. 13.1 Schematic

representation of a snapshot
of humid air, with air
molecules represented as
small dots, and water
molecules represented as
large dots
13.1 Background 477

(and rising due to fossil fuel combustion) and many other components in decreasing
concentration.1 These components are considered to be noncondensable, except
under sufficiently low temperatures and/or high pressures, and can be treated
together (in the absence of chemical reactions), with an equivalent molecular
weight of 28.8 kg/kmol.
On the other hand, water vapor is condensable. If humid air is sufficiently cooled
(isobarically), it will condense (at its dew point temperature). The concentration
(expressed as a number per unit volume, mole fraction, or mass fraction) of water
vapor in humid air is variable and limited by the liquid/vapor phase change
properties of water. Dalton’s Law defines partial pressures for gas mixtures, in
which each component of the mixture exerts a pressure on the walls of the vessel as
if the other components did not exist. The total pressure is the sum of the partial
pressures. The partial pressure exerted by water vapor in humid air is shown as a
function of temperature in Fig. 13.2 for several values of the relative humidity. The
partial pressure exerted by water vapor for 100 % relative humidity air corresponds
to the saturation pressure, that is, the pressure at which pure water would boil for a
given temperature. Relative humidity curves are simple fractions of that curve.
For nominal room temperature air (say 25  C at 1 atm total pressure), the partial
pressure exerted by water vapor in 100 % humid air is only 0.025 atm. If a drop of

1
0.9
0.8
Partial Pressure of HOv (atm)

0.7
0.6
0.5 Relative Humidity = 100%
80%
0.4 60%
0.3 40%
20%
0.2
0.1
0
0 10 20 30 40 50 60 70 80 90 100
Temperature (oC)

Fig. 13.2 Saturation (vapor) pressure of water as a function of temperature between the normal
freezing and boiling points (0 and 100  C) for various values of relative humidity. The saturation
curve (100 % relative humidity) is formed from a fit of the form P ¼ AeB/T (consistent with the
Clausius–Clapeyron equation) using tabulated points at 25 and 100  C

1
http://en.wikipedia.org/wiki/Atmosphere_of_Earth, accessed 19 November 2010.
478 13 Evaporation and Mass Transfer Fundamentals

liquid water were added, that drop will simply fall to the floor of the container, and
none of it would evaporate. However, if that initial 100 % relative humidity air at
25  C were heated at constant total pressure to, say, 70  C, it would exert the same
partial pressure, but would now correspond to a relative humidity of less than 20 %.
At the higher temperature, the water vapor is capable of exerting a much higher
partial pressure, but the amount of water vapor is fixed by what was initially
present. If a drop of liquid water were now added, that liquid would evaporate,
raising the relative humidity. More liquid water could be added, and it would
continue to evaporate until the partial pressure of water was 0.3 atm, at which
point the air would be saturated.
Consider the scenario in Fig. 13.3 in which hot coffee is poured into a container
and a lid is suddenly placed on it, trapping room air. The sketch on the left shows
the situation immediately after placing the lid. The sketch on the right shows the
situation after some period of time long enough that the air becomes saturated
(right), but short enough that the coffee is still hot. There is a vent hole that allows
the total pressure to remain atmospheric and therefore room air can flow in, or the
vapor/air mixture inside can flow out, depending on the circumstances. The trapped
air starts with the same humidity as the ambient air, which is less than or equal to
100 %. Inside the mug, there are two regions in different phases, namely, a liquid
(coffee, shown as solid black), and a gas above it, which is a mixture of water
molecules (big dots) and air molecules (small dots). The air will naturally be
warmed to a value between the coffee temperature and ambient. A good exercise
would be to estimate its temperature using a resistance network. If the container and

Fig. 13.3 Schematic representation of a container of liquid water with trapped room air immedi-
ately after a lid is placed (left) and after enough time for equilibrium to be established (right)
13.1 Background 479

Fig. 13.4 Schematic representation of the dynamic quasi-equilibrium at a water/humid air

interface. Gas phase water molecules that strike the surface become absorbed in the surface as
some liquid molecules near the surface can break free

lid are well insulated, the air temperature will be close to the coffee temperature.
Given time, presumably short compared to the cooling time, enough liquid will
evaporate until the now warm humid air will become saturated.
Figure 13.4 is a close-up snapshot of the dynamic equilibrium associated with
the coffee-free surface, with two water molecules shown with arrows to indicate
their motion. The molecules in the liquid phase are all vibrating, but are held
together by intermolecular forces. Liquid molecules near the free surface, however,
can occasionally vibrate so hard that they break free from their neighbors, and enter
into the gas phase. The higher the temperature of the liquid, the harder they vibrate,
and the more leave in a given time frame. These molecules are evaporating (going
from liquid to gas phase), and the number of them evaporating in a given period of
time increases strongly with liquid temperature.
The water molecules in the gas phase, by contrast, are not heavily influenced by
intermolecular forces, until they experience a collision. They travel in straight lines
until they encounter either an air molecule, another water molecule, the free surface
or the walls of the container. During collisions between molecules, the intermolec-
ular forces are very high, short duration, and repulsive. The collision serves to
redirect both collision partners. Collisions with the free surface of the liquid are
different, however. A water molecule that collides with the free surface can be
absorbed into the liquid phase. Those molecules condense (going from gas to liquid
phase). The number of molecules that condense in a given period of time, then,
depends directly on the number of collisions with the free surface that take place in
that time. It is generally assumed (and justifiable) that a quasi-equilibrium is
quickly established in the near vicinity of the liquid surface in which the rate of
molecules evaporating equals the rate of molecules condensing. Therefore, the
partial pressure of the water vapor molecules is generally assumed to be equal to
the saturation pressure just on the gas side of the free surface.
480 13 Evaporation and Mass Transfer Fundamentals

Fig. 13.5 Schematic representation after the lid is first removed (left) and the subsequent quasi-
steady scenario, with the migratory path of a target molecule shown in red (right)

Figure 13.5 shows the mug just after the lid is removed. Those vapor molecules
that would otherwise have hit the lid and bounced back inside now escape to the
room. The concentration (number density) therefore decreases near the open top of
the container. Those escaping molecules are replaced by molecules from below,
some of which then escape, and a form of “concentration wave” gradually works its
way down toward the free surface of the coffee. At the free surface, molecules from
the liquid continue to evaporate into the gas phase, while gas phase water molecules
that hit the liquid are reabsorbed. The concentration of vapor molecules just above
the coffee surface is therefore maintained at some level near the saturation condi-
tion associated with the coffee surface.
A quasi-steady state situation is achieved (in short order, in practice, compared
to the cooling rate) in which vapor molecules escape at a steady rate. The right
sketch shows the path of an individual vapor molecule, painted red. It starts out near
the free surface and is intermittently redirected at each collision. The average length
of each straight line, called the “mean free path,” is the average distance of travel
between collisions. Eventually, the molecule works its way up and out. The speed
of the molecules is close to the speed of sound (i.e., fast), but the speed that it moves
in the vertical direction is much slower than that. This migration from regions of
high concentration to low concentration is called diffusion and is the fundamental
mechanism required for evaporation to occur.
13.2 Model Development 481

With this backdrop of the basic physics of vapor pressures and diffusion, some
simplified models aimed at quantifying the evaporation rate, and its thermal
implications, are developed.

13.2 Model Development

A series of models to quantify fundamental diffusion effects is developed. First, the

energy balance on the lumped capacity model is modified to include the effects of
evaporation. Then a model of pure diffusion, called Model 1 across the air cylinder
above the coffee surface is derived heuristically. Model 2 recognizes that along
with the diffusion of water vapor through the air cylinder, there is a net drift, or bulk
flow, that was ignored in Model 1. Model 3 considers a control volume outside the
mug and yields a working model more suitable when the mug is nearly full. Model
4 combines Models 2 and 3. These models are intended to predict the effect of
evaporation during the cooling phase of the coffee, and the change in mass of the
coffee/mug system is neglected. A separate final analysis focuses on the change in
mass for a mug of coffee left indefinitely on a table. How long does it take to
completely evaporate?

13.2.1 Lumped Model Revisited

Consider the original system used to analyze the cooling mug of coffee. The
boundary was placed on the free surface of the mug, and the mass that crossed it,
hence the evaporation, was neglected. That is, it was treated as a closed system, to
which the conservation laws of mass and energy apply. To address the phenomenon
of evaporation, the system must be treated as an open system, that is, one in which
mass crosses the boundary.
The First Law of Thermodynamics for this particular open system is not derived
in detail here. The result, which adds a simple term to account for evaporation to the
lumped capacity model, is:

dT coffee
ðMcÞCoffeeþMug ¼ Q_ FreeSurface þ Q_ sides þ Q_ floor  m_ evap 
hfg
dt
The heat transfer rate into the system from the free surface, sides, and floor are
appropriate convection and radiation modes of heat transfer. The free surface term,
however, represents the heat transfer between the free surface and ambient, which
acts in parallel to the evaporation process, and can be considered a convective/
radiative mode. The last term, the evaporative cooling term, is the focus of this
chapter. The rate of evaporation, ṁevap, is the rate of evaporation, in kg/s, and 
hfg is
the enthalpy of evaporation of water. Water that evaporates enters the free surface
as a liquid, and exits as a vapor, with a higher specific energy content. The energy
required to change the phase has the effect of reducing the energy stored in the
482 13 Evaporation and Mass Transfer Fundamentals

CV. There is nothing to stop a definition of an evaporative “convection-like”

coefficient:

m_ evap 
hfg  hevap SðT 1  T S Þ

where S ¼ πD2/4 is the area of the free surface of the coffee. Rearranging to define
the evaporative heat transfer coefficient:

m_ evap 
hfg
hevap ¼
Sð T 1  T S Þ

This evaporative heat transfer coefficient model really makes the most sense when a
hot liquid is cooling, and the surface temperature of the liquid is substantially
different than ambient. When the coffee temperature equals ambient, evaporation
is not zero, and the coefficient is undefined (divided by zero).
A traditional way to express the mass transfer process, analogous to Newton’s
Law of Cooling, is to define an evaporative coefficient, γevap, that is the
proportionality factor (not necessarily constant) for evaporation driven by a differ-
ence in concentration of the evaporating species, expressible as the mass fraction, Y:

m_ evap ¼ γ evap SðY S  Y 1 Þ

Here, YS is the mass fraction of the evaporating component (water vapor) just
above the surface, and Y1 is that in the free stream (which is where humidity of the
air comes into play). This latter model is more appropriate for a pure evaporation
process, not one in which cooling of the bulk liquid is taking place simultaneously
with the evaporation. That is, the evaporation process after the coffee cools to
ambient (or very close to it). In practice, correlations similar to Nusselt number
correlations are used to determine mass transfer coefficients.
Something surprising happens at steady-state, namely, the coffee temperature
does not equal ambient temperature:

Q_ FreeSurface þ Q_ sides þ Q_ floor ¼ m_ evap 

hfg ðat steady stateÞ

The energy required for evaporation is supplied by heat transfer from the environ-
ment to the coffee. Using a lumped capacity resistance network, this heat transfer
rate can be expressed in terms of an overall heat transfer coefficient, U, based on the
surface area of the mug of coffee:


Q_ FreeSurface þ Q_ sides þ Q_ floor ¼ UA T 1  T coffee, SS ¼ m_ evap 
hfg

Solving for the steady-state coffee temperature:

m_ evap 
hfg
T coffee, SS ¼ T 1 
UA
13.2 Model Development 483

The only way heat can be transferred to the coffee is if the coffee is at a lower
temperature than ambient. This phenomenon is difficult to measure, but is real, and
is why we sweat.

13.2.2 Model 1: Pure Diffusion Model and Fick’s Law

To model the rate of evaporation, consider a control volume that consists of the
humid air column between the free surface and the exit plane below the rim, shown
in Fig. 13.6. On the bottom of the control volume, at y ¼ 0, the air is assumed to be
at the temperature of the liquid at the surface (not necessarily equal to the bulk
coffee temperature), and with a mass fraction Yv0. An upper limit to Yv0 would be
to assume the air is saturated there. At the top of the control volume, at y ¼ L, the air
has a mass fraction YvL, which has a lower limit defined by the humidity of ambient
room air. Even if the liquid is at ambient temperature, the concentration of water
vapor at the surface is higher than that in the ambient, and a concentration gradient
exists, unless the room is at 100 % relative humidity and the same temperature as
the liquid. The actual concentration of water vapor at the exit plane is generally
higher than that of the ambient air, but air currents from the room tend to make it
equal, in the limit, especially for large values of L. For liquids colder than ambient,
the concentration is lower and water vapor diffuses in and a net condensation
occurs.

Fig. 13.6 Control volume of

the humid air column between
the coffee surface (at y ¼ 0)
and the exit plane (at y ¼ L),
shown as dashed line
484 13 Evaporation and Mass Transfer Fundamentals

Since evaporation is driven by a difference in concentration, it is postulated that

the rate of evaporation, expressed as a mass flow rate of water vapor through the
CV, is proportional to the difference in concentration, expressed as a mass
fraction, Y, between the top and bottom of the control volume:

m_ evap  ðY v0  Y vL Þ

It is further postulated that the evaporation rate varies inversely with the distance
between the two surfaces, L:
1
m_ evap 
L
These two effects together make up a gradient, a change in mass fraction over a
distance.
Furthermore, if the surface area (S ¼ πD2/4) is doubled, the evaporation rate is
expected to double, meaning the evaporation rate is proportional to S:

m_ evap  S

Now, the more dense the humid air, the more molecules are packed in a given
volume, all of which are diffusing out:

m_ evap  ρ

Putting these four effects together:

ρSðY v0  Y vL Þ
m_ evap 
L
The proportionality factor includes all other effects, most importantly, how easy it
is for molecules to migrate from the concentrated area to the less concentrated area.
In keeping with convention, a diffusion coefficient Dv is introduced:

ρDv S
m_ evap, MODEL 1 ¼ ðY v0  Y vL Þ
L

The SI units of Dv are m2/s. A literature value for the diffusion coefficient of water
vapor into air (at nominal room conditions) is:

Dv ¼ 25:6ð10Þ6 m2 =s

Fick’s Law of Diffusion has just been derived heuristically. It is analogous to

Fourier’s Law of Conduction. This model neglects the fact that there is a bulk
13.2 Model Development 485

flow leaving the mug, which complicates matters and is addressed in the next
model. A more general expression in terms of a local concentration gradient is:

∂Y v
m_ diffusion ¼ ρDv S
∂y

The negative sign forces the direction of diffusion to be from high to low
concentration.
In Model 1, the gradient of mass fraction is taken as being the overall change in
mass fraction divided by the overall length. Consider the limits. The mass fraction
of vapor just above the free surface cannot exceed the saturation condition:

Y v0  Y SAT , surface

At the exit plane, the mass fraction cannot be lower than ambient:

Y vL  Y 1

Both of these limits work in the same direction, namely, to reduce the value of the
concentration difference from the free surface to the exit plane, and therefore:

Y v0  Y vL  Y SAT , surface  Y 1

For Model 1, the limiting case is considered, and the inequality is treated as an
equality.
The Appendix details the calculation of mass fractions for this case. The mass
fractions are directly related to the partial pressure exerted by the vapor, and
therefore are very sensitive to temperature.

Model 1 Results
Results of model predictions are presented for a base case, with parameter values
shown in Fig. 13.7. The mug has an 8.93 cm inner diameter (D) and it is filled with
coffee at various temperatures in a room with air at 25  C and a 70 % relative
humidity. The amount of coffee added is also varied, and therefore the distance
from the free surface to the exit plane (the air cylinder height, L, which is the
distance water must diffuse to exit the mug).
The semi-logy plot shows the evaporation rate as a function of instantaneous
coffee temperature for several values of the air cylinder length, L. At a fixed value
of L, the evaporation rate is zero at a temperature somewhat below ambient
(approximately 19  C). This temperature is the dew point temperature and
corresponds to the condition in which the mass fraction of vapor on the free surface
is equal to the mass fraction in the ambient. The evaporation rate increases sharply
as the temperature moves away from the dew point on both sides. At lower
temperatures, however, the evaporation rate is negative (which cannot be plotted
on semi-log axes so the absolute value is shown), indicating that water vapor is
diffusing into the mug and condensing. At ambient temperature, the evaporation
rate is not zero because the room air has a relative humidity less than 100 %.
486 13 Evaporation and Mass Transfer Fundamentals

INPUT PARAMETERS
D 0.0893 m mug inner diameter
L VARY m distance from free surface to exit plane
Pinf 1.013E+05 N/m2 atmospheric pressure
Tinf 25 oC ambient temperature
rh 0.7 relative humidity of ambient
PHYSICAL CONSTANTS
Dv 2.56E-05 m2/s Diffusion coefficient (water vapor into air)
DA 2.06E-05 Diffusion coefficient (air into itself)
hfg 2.33E+06 J/kg Heat of Vaporization of water
A 9.1496E+10 N/m2 CLAUSIUS-CLAPYRON CONSTANTS FOR WATER
B -5.1152E+03 K B
(T + 273)
DERIVED PARAMETERS Psat (T ) = Ae
S 0.00626315 m^2 Free Surface Area
Xinf 0.0222 mol fraction of ambient
Yinf 0.0138 mass fraction of ambient
rho 1.1857 kg/m3 dry air density
X0,sat 0.0317 saturated surface mol fraction at ambient
Y0,sat 0.0199 saturated surface mass fraction at ambient
Dew Point 18.93 oC Dew Point temperature

1.E-04
L = 0.001 m
ambient

1.E-05
0.02
ABS(Evaporation rate) (kg/s)

0.04
1.E-06
0.06
0.08
1.E-07 0.10

1.E-08

1.E-09

1.E-10
0 20 40 60 80 100
Coffee Temperature (oC)

Fig. 13.7 Evaporation rate for Model 1 (pure diffusion) as a function of coffee temperature for
various lengths of the air column above the free surface. Below 18.93  C (the dew point), the
evaporation rate is negative
13.2 Model Development 487

As temperature increases above ambient at a fixed L, the evaporation rate

increases steadily to the boiling point, changing by over two orders of magnitude.
This exponential growth is a direct result of the vapor pressure above the liquid
surface. At a fixed temperature, the evaporation rate decreases with L, the distance
across which vapor must diffuse. The case for a very nearly full mug (L ¼ 0.001 m)
is shown to demonstrate that the evaporation becomes very sensitive to L as L
approaches zero, at least as predicted by Model 1. This behavior will be shown to
represent a breakdown of the model, which imposed an unrealistic constraint that
the vapor mass fraction at the exit plane is equal to the ambient value. The lower the
value of L, the worse that assumption is expected to be. This limitation of the pure
diffusion model will be addressed in Model 3.
A more subtle limitation to the pure diffusion model is that with evaporation,
there is a net mass flow upward. This effect is one of advection, and is addressed
in Model 2.

13.2.3 Model 2: Advection/Diffusion Model

There are two parallel means by which water vapor is transported through the air
column. There is diffusion across the concentration gradient, and simultaneously,
there is a bulk flow that carries it. Model 1 does not account for the bulk flow effect.
To do so, it is insightful (and convenient) to view things from the perspective of the
air. In this two-component gas system (air and water vapor), the local mass
fractions of air and water must everywhere account for all species:

Y v þ Y air ¼ 1

Taking the derivative:

dY v dY air
þ ¼0
dy dy

This relation demonstrates that if there is a concentration gradient in the water

vapor, there must be a gradient in the air of equal magnitude and opposite sign. In
the humid air column, air diffuses in the opposite direction as the vapor, toward the
coffee surface. However, air does not experience a net flow downward, as it is
blocked by the free surface of the coffee (which is considered to be fixed, although
it is receding slowly). So as air diffuses downward, it is equally swept upward by
the bulk flow. To express these two effects mathematically, with DA being a
diffusion coefficient for air into humid air, the mass flow rate of air crossing any
plane inside the humid air column is:

∂Y air
m_ air ¼ 0 ¼ ρDA S þ m_ evap Y air
∂y
488 13 Evaporation and Mass Transfer Fundamentals

The mole fractions are assumed to be uniform in the horizontal direction, and the
partial derivative can be replaced by a total derivative. Separating variables:
 
dY air m_ evap
¼ dy
Y air ρDA S

Integrating across the entire control volume (from the free surface to the exit plane),
recognizing that the mass flow rate through the column is the evaporation rate and is
independent of position:

YðAL ðL  
dY air m_ evap
¼ dy
Y air ρDA S
Y A0 y¼0
   
Y AL m_ evap
ln ¼ L
Y A0 ρDA S

Expressing the air mass fraction in terms of the vapor mass fractions (using
YA + YV ¼ 1), and rearranging for the evaporation rate:
 
ρDA S 1  Y vL
m_ evap ¼ ln
L 1  Y v0

In keeping with tradition, it is useful to define a mass transfer number, B0L, not to be
confused with a Biot number, as follows:
 
1  Y vL
 1 þ B0L
1  Y v0

The two subscripts refer to the two boundaries (coffee surface and exit plane).
Performing some algebra:
     
1  Y vL 1  Y v0 Y v0  Y vL
B0L ¼  ¼
1  Y v0 1  Y v0 1  Y v0

The final result for the evaporation rate is:

ρDA S
m_ evap ¼ lnð1 þ B0L Þ
L
where

Y v0  Y vL
B0L ¼
1  Y v0
13.2 Model Development 489

Fig. 13.8 Demonstration of 1000

low Biot number limit. The
functions B and ln(1+B) are B
plotted versus B ln(1+B) 100

10

1
0.001 0.01 0.1 1 10 100 1000
0.1 B, Mass Transfer Number

0.01

0.001

The mass transfer number is useful since ln(1) ¼ 0, and the evaporation rate is zero
for B ¼ 0, is positive for B > 0, corresponding to evaporation, and negative for
B < 0, corresponding to condensation. Furthermore, there is a useful limit for low
values of B, namely:

lim lnð1 þ BÞ ¼ B
B!0

As shown in Fig. 13.8, this limit is valid for B < 0.1, approximately.
To make quantitative predictions, the mass fractions must be estimated. Consid-
ering the limits, the mass fraction of vapor just above the free surface cannot exceed
the saturation condition:

Y v0  Y SAT , surface

At the exit plane, the mass fraction cannot be lower than ambient:

Y vL  Y 1

Both of these limits work in the same direction, namely, to reduce the value of the
transfer number, so it can be stated that:

Y SAT , surface  Y 1
B0L 
1  Y SAT , surface

A diffusion coefficient for air into humid air is different from the diffusion coeffi-
cient of water vapor into humid air, namely:

DA ¼ 20:6ð10Þ6 m2 =s
490 13 Evaporation and Mass Transfer Fundamentals

100

10

B=0
B<0 B>0
Abs(B0inf) Number

Ambient
0.1

0.01

0.001
0 10 20 30 40 50 60 70 80 90 100
Coffee Temperature (°C)

Fig. 13.9 Mass transfer number (B) as a function of coffee temperature, assuming saturation
conditions at the coffee surface, and ambient conditions at the exit plane

This difference is due fundamentally to the different size of the molecule.

Model 1 was derived from the point of view of water diffusing out, while Model
2 is based on air diffusing in. These approaches are not inconsistent, just different
points of view.

Model 2 Results
The results of Model 2 presented assume the mass fraction limits hold in evaluating
the B number. The accuracy of that assumption is expected to improve as the
distance between the free surface and exit plane (L) increases. Figure 13.9 shows
the absolute value of the mass transfer number (B) as a function of coffee tempera-
ture on a semi-logy scale. At approximately 18.5  C, the B number is zero (as is
ln(1 + B)), corresponding to the mass fraction on the free surface taking on the
ambient value. This temperature corresponds to the dew point temperature of the
ambient air, that is, the temperature at which water would begin to condense
on isobaric cooling. The B number (and therefore ln(1 + B)) is negative below the
dew point, and it is positive above. Negative evaporation rates correspond to
condensation.
The B number increases by over two orders of magnitude between 30 and 90  C.
Above 90  C, it turns sharply upward as the mass fraction at the surface (YSAT,0)
approaches unity, and the denominator goes to zero. In this calculation, the surface
temperature is set at a particular value. The evaporation process has a cooling effect
on the surface of the coffee, and it becomes more difficult to maintain the surface
temperature the closer to the boiling point. To demonstrate that effect, consider the
13.2 Model Development 491

boundary condition at the coffee surface, where the heat transfer to the surface from
the adjacent fluid is modeled with heat transfer coefficients:


hs, coffee S T coffee  T S þ m_ evap 
hf ¼ hs, air SðT S  T ambient Þ
þ m_ evap hg ðsurface energy balanceÞ

Solving for the surface temperature:

hs, sir ST ambient þ hs, coffee ST coffee  m_ evap 

hfg
TS ¼
hs, air S þ hs, coffee S

Assuming the heat transfer coefficient between the surface and coffee is much
greater than that between the surface and ambient air (hs,coffee  hs,air):

m_ evap 
hfg
T S ¼ T coffee 
hs, coffee S

The source of the energy required for the evaporation process is heat transfer from
the adjacent fluids, and therefore the surface temperature is below what it would be
in the absence of evaporation. As the boiling point is approached, the evaporation
rate increases to very large values (as the mass fraction of vapor approaches unity).
Figure 13.10 compares the evaporation rates for the pure diffusion model with the
advection/diffusion model for the base case (with L ¼ 0.05 m), with the evaporation

1.E-04

1.E-05
ABS(Evaporation Rate) (kg/s)

Advection/Diffusion Model

1.E-06

1.E-07
Pure Diffusion Model

1.E-08

Model 2
1.E-09
Model 1

1.E-10
0 20 40 60 80 100
Coffee Temperature (°C)

Fig. 13.10 Comparison of the evaporation rate from the pure diffusion and advection/diffusion
models
492 13 Evaporation and Mass Transfer Fundamentals

rate plotted against coffee temperature. The comparison is qualitatively similar for
all non-zero values of L (and are indeterminate for L ¼ 0, where the evaporation rate
is infinite). The models agree almost exactly for temperatures below approximately
70  C. Above that, the advection/diffusion model predicts an increasingly higher
evaporation rate than the pure diffusion model. Physically, there are two parallel
mechanisms for the transport of vapor across the air cylinder in Model 2.
Both Models 1 and 2 assume that the condition at the exit plane is that of the
ambient air. This artificial constraint becomes less realistic as L decreases.
It is apparent that the effects of advection are only important as the temperature
approaches the boiling point of the liquid. Returning to the original problem of the
cooling coffee cup, where the coffee starts at a temperature close to its boiling point
and gradually cools, the advection/diffusion model represents a significant
improvement over the pure evaporation model. It has not been demonstrated (yet)
whether the evaporation process has a significant impact on the cooling time and
that will be addressed after the final model has been developed.

13.2.4 Model 3: Diffusion from a Point Source

The models considered thus far have focused on a control volume about the air
column inside the mug. Evaluation of the evaporation rate requires an assumption
about the value of the water mass fraction at both the free surface (with an upper
limit at the saturation condition) and at the exit plane of the mug (with a lower limit
of the ambient absolute humidity). For the condition at the free surface, it can be
argued that setting the mass fraction (Yv0) to the saturation limit (Ysat,0) is a good
approximation, except perhaps as the boiling point is approached, and the evapora-
tion rate is highest.
On the other hand, to say that the mass fraction of water vapor at the exit plane is
equal to the ambient value seems reasonable for a large value of L, but not when the
mug is nearly full. Visual inspection of the vapors leaving a hot cup of coffee
suggests that there is a significantly higher vapor concentration at the exit plane
than exists in the free stream. Furthermore, for full mugs (L ! 0), the predicted
evaporation rate increases dramatically, making a difference between the exit plane
and ambient even more likely.
In order for water vapor to leave the free surface, it must pass through the air
column to the exit plane, and then pass from the exit plane to the far field away from
the mug. To get a handle on that outer world, a simplified quasi-steady
one-dimensional analysis can be developed. Figure 13.11 shows a schematic that
models the exit plane, at a fixed mass fraction, to be a hemispherical shell of radius
R0. The area of the shell is the same as the area of the exit plane:

πD2 1

¼ 4πR20
4 2
13.2 Model Development 493

Fig. 13.11 Radial

coordinate system outside the
mug, with the surface as a
hemisphere whose surface
area equals that of the
coffee-free surface area

D
R0 ¼ pffiffiffi
8
The use of the hemisphere is conceptually accurate if the exit plane is at the same
level as the floor, and not a height H above it. This estimate therefore is an
underestimate of the ability for water vapor to diffuse away. The effects of the
global natural convection have a much larger effect in practice. Therefore, the result
of this model provides a lower limit on the ability for the environment to maintain
the exit plane at ambient conditions.
At an arbitrary radial position from the center line, there is no net mass flux
of air:
 
dY A 4πr 2
m_ air ¼ 0 ¼ m_ evap Y a þ ρDA
dr 2

Separating variables:
 
dY A m_ evap dr
¼
YA 2πρDA r 2

Integrating from the exit plane (at R0, where YA ¼ YAL) to “infinity” (where
YA ¼ Y1):

Yð1   1
ð
dY A m_ evap dr
¼
YA 2πρDA r2
Y A ¼Y aL r¼R0
    1
Y1 m_ evap 1 
ln ¼
Y AL 2πρDA r r¼R0

Evaluating the limits, expressing mass fractions in terms of water (1 ¼ Yv + YA)

and rearranging for the evaporation rate, with the point source radius (R0)
expressed in terms of the mug diameter (D, not to be confused with the diffusion
coefficient DA):
494 13 Evaporation and Mass Transfer Fundamentals

 
D 1  Y v1
m_ evap ¼ 2π pffiffiffi ρDA ln
8 1  Y vL
π
m_ evap ¼ pffiffiffi DρDA lnð1 þ BL1 Þ
2
where the transfer number applies between the exit plane (L) and ambient (1):

Y vL  Y v1
BL1 ¼
1  Y vL
The major assumption of the point source model is that the only air movement
outside the coffee cup is that due to the evaporation process. In reality, especially
when the coffee is hot, natural convection currents will create a vertical flow that
will bring ambient air closer to the exit plane, and ease the ability for vapor to
diffuse from the exit plane to ambient. Therefore, the point source represents a
conservative prediction for the evaporation rate.

Model 3 Results
Figure 13.12 shows the evaporation rate as a function of coffee temperature on
semi-logy axes for the point source model (Model 3, dotted line) and the advection/
diffusion model (Model 2) for several values of air column lengths (L). The point
source model prediction is independent of the length of the air column (L), so only

1.E-04
L = 0.001 m
ambient

1.E-05
0.02
Point Source Model
Evaporation rate (kg/s)

0.04
1.E-06
0.06
0.08
1.E-07 0.10

1.E-08

1.E-09

1.E-10
0 20 40 60 80 100
Coffee Temperature (oC)

Fig. 13.12 Model 2 (air cylinder with advection/diffusion) and Model 3 (point source) evapora-
tion rates
13.2 Model Development 495

one curve is shown. The shape of the curve is similar for both models because both
are being driven by the same transfer number, B0L.
The evaporation rate predicted by the air column model (Model 2) is greater than
that predicted by the point source model for air column lengths less than approxi-
mately 3 cm. Conceptually, the exit plane is shared by both models, but it
corresponds to the exit plane for the air column and is the inlet plane for the
point source. The point source curve (dotted) line represents the maximum rate at
which vapor can diffuse away from the exit plane into the room (subject to the
major assumptions of the model). For L < 0.03, then, the mass delivered to exit
plane by the air column exceeds the ability for the point source to remove it.
One way around this contradiction in implementing the model would be to
calculate the evaporation rate from both models, and choose the lower of the two
as being controlling. Another, more sophisticated, way around it is to relax the
imposition of the mass fraction at the exit plane to be at one extreme or the other
(ambient in the air column model and surface saturation in the point source model).
That leads to Model 4.

13.2.5 Model 4: Matching the Air Column with the Point Source

This model matches Model 3 (point source) with Model 2. In order to escape from
the free surface to the room, water vapor must first diffuse, with the help of a
convective push, across the air cylinder to the exit plane. From there, it must diffuse
away to ambient. Models 2 and 3 result in simple expressions for the evaporation
rate, but the transfer number depends on the mass fractions at the boundaries of
each model. The two mass transfer regions (inside the air cylinder and the spherical
shell outside) are in series. Expressing the evaporation rate in terms of mass thermal
resistances, first across the air column, and then for the point source outside
the mug:

Y v0  Y vL Y vL  Y v1
qevap ¼ m_ evap 
hfg ¼ ¼
Rm, 0L Rm, L1

Rearranging for the intermediate mass fraction (YvL) and dropping the “m” from
the subscripts:

RL1
Y vL ¼ Y v1 þ ðY v0  Y v1 Þ
R0L þ RL1
The resistances cannot be evaluated completely from input variables. They depend
on YvL, making the equation nonlinear, and difficult, if not impossible, to solve
analytically. An iterative numerical method is viable, however. Forming a resis-
tance ratio:
496 13 Evaporation and Mass Transfer Fundamentals

1
Y vL ¼ Y 1 þ R0L ðY v0  Y 1 Þ
=RL1 þ 1

Entering the appropriate Models 2 and 3 results and expressing S in terms of D:

LðY v0 Y vL Þ    
R0L ρDahfg Slnð1þB0L Þ pffiffiffi L lnð1 þ BL1 Þ Y v0  Y vL
¼ pffiffi ¼ 8
RL1 2ðY vL Y v1 Þ D lnð1 þ B0L Þ Y vL  Y v1
πDρDahfg lnð1þBL1 Þ

An implicit expression for the intermediate mass fraction (i.e., at the exit plane) is:

ðY v0  Y 1 Þ
Y vL ¼ Y 1 þ pffiffiffi
 
lnð1þBL1 Þ Y vL
8 D lnð1þB0L Þ YYvLv0Y
L
v1
þ1

The transfer numbers both depend on YvL, making this equation nonlinear. The
equation in this form is a suitable recursion relation for application of the method of
successive substitution, however, in which case a value for YvL is guessed, followed
by evaluation of the right hand side, whose result is used as the next guess. An
iterative loop is run until the method either converges, or fails. If it fails, another
method can be attempted.

Model 4 Results
Implementation of Model 4, the matched model, requires an iterative solution. It is
cumbersome, though possible, to solve the transient coffee problem using Excel
(with macros and/or user-defined functions) because an iterative loop is required for
each time step. It is easier (but still requiring a macro) to write an excel file that
merely calculates everything as a function of instantaneous coffee temperature.
A macro-enabled excel file was written to do so. A macro was written that performs
the copy and “paste as values” required to conduct an iteration. A test for conver-
gence is conducted. However, the user is required to check to see if the solution is
converged, and run the macro repeatedly until it does. Every time a change is made
to an input variable (i.e., L), the iteration macro must be run.
Figure 13.13 shows the evaporation rate as a function of coffee temperature for
several values of air column length (L) for the matched model (Model 4) and the
point source model (Model 3, which is independent of L). This plot shows the same
cases as for Fig. 13.12, where the air column model was compared with the point
source model. In that latter case, the air column model predicted higher evaporation
rates for nearly full mugs (L < 0.03 m). In the matched model, the point source
model places an upper limit on the air cylinder. The problem from the straight air
column model, where the evaporation increases dramatically L approaches zero is
eliminated.
Model 4 is considered to be the most comprehensive model, yet it must be
emphasized that it is based on a one-dimensional model, and the effects of natural
13.2 Model Development 497

1.E-04

1.E-05
Evaporation Rate (kg/s)

1.E-06

Point Source
1.E-07
0.001
0.02
1.E-08
0.04
0.06
1.E-09
0.08
0.1
1.E-10
0 20 40 60 80 100
Coffee Temperature (°C)

Fig. 13.13 Model 4 evaporation rate predictions for different air column lengths (in m)

convection currents would render that approximation questionable at best. Model

4 is therefore considered to somewhat underestimate the actual evaporation rates.

13.2.6 Comparison of Heat Transfer Coefficients

To this point, only the evaporation rate has been investigated from a purely mass
transfer point of view. It is clear that the evaporation depends strongly on the
instantaneous temperature of the liquid, driven by the highly nonlinear saturation
pressure. What is not yet clear is whether the evaporation has a significant effect on
the original problem of the cooling of a cup of coffee. From before, an evaporative
heat transfer coefficient can be defined as:

m_ evap 
hfg
hevap ¼
Sð T 1  T S Þ

Inserting the evaporation rate expression:

 
ρDv 
hfg lnð1 þ B0L Þ
hevap ¼
L ðT S  T 1 Þ

where, again,

Y v0  Y vL
B0L ¼
1  Y v0
498 13 Evaporation and Mass Transfer Fundamentals

100

Evaporative Heat Transfer Coefficient

(W/m2/oC) 10

Point Source
1
0.001
0.02
0.04
0.1 0.06
0.08
0.1
0.01
0 20 40 60 80 100
Coffee Temperature (oC)

Fig. 13.14 Evaporative heat transfer coefficients predicted by Model 3 (point source) and
4 (matched) for various values of L (distance between coffee surface and exit plane)

Figure 13.14 shows the evaporative heat transfer coefficients plotted on semi-logy
axes against coffee temperature for several values of the air column length (L) using
the matched model (Model 4) and the point source model (Model 3). The evapora-
tive heat transfer coefficient is very sensitive to temperature and exhibits some
dramatic behavior when the coffee is close to ambient, to be discussed shortly.
Three parallel mechanisms for energy transfer from the free surface to ambient
(radiation, convection, and evaporation) are compared in Fig. 13.15 for the base
case (L ¼ 0.05 m, a half-mug of coffee). The radiation model used assumes the free
surface is completely surrounded visually by ambient. In reality, the inner walls of
the mug will be somewhat higher, and the effective radiative heat loss will be lower
than predicted by this model, so the radiation curve is an upper estimate. The
convection model shown is based on published Nusselt number correlation for a
heated horizontal plate. That is the model that would seem most applicable if the
free surface were treated as a non-evaporating surface, and the mug were filled to
the brim. The actual heat transfer rate by conduction to the surface would be
lessened somewhat by the evaporative flow (especially near the boiling point),
and also by the free surface being below the rim. So the convection curve is also an
upper limit. On the other hand, the evaporation curve is an underestimate because
use of the point source model as controlling the diffusion away from the mug is the
slowest possible case, particularly when the coffee is hot and generating a strong
vertical natural convective flow.
When the coffee is first poured, the evaporative heat loss is very high, and
dominates the heat loss from the free surface. However, not modeled here, this
effect would tend to lower the free surface temperature below the average coffee
temperature, reducing the effects of evaporation. As the coffee continues to cool,
evaporation effects drop dramatically and becomes a small contributor. The natural
13.3 Evaporation of a Forgotten Cup of Coffee 499

20

Heat Transfer Coefficient (W/m2/K) 15

Evaporative
Convective
Radiative
10

0
0 10 20 30 40 50 60 70 80 90 100
Coffee Temperature (oC)

Fig. 13.15 Comparison of the evaporative, natural convective and radiative heat transfer
coefficients associated with the coffee-free surface as a function of coffee temperature

convection and radiation models are equally important, except as the coffee
approaches ambient, where the natural convection drops below to zero if the coffee
temperature equals ambient.
Therefore, for the overall cooling of the coffee, evaporation will contribute to a
high heat loss rate at the very beginning, at the same time the cold mug is being
heated, and subsequently is noticeable, but not dominant until the coffee is nearly at
room temperature.
Mathematically, the evaporative heat transfer coefficient is infinite when the
coffee temperature equals ambient, but the heat flux is finite. Use of an evaporative
heat transfer coefficient is reasonable when the coffee is cooling (or heating), and
when the coffee is near ambient. The energy transfer (not heat) associated with
evaporation is driven by diffusion, not by a temperature difference.
When the coffee is at the dew point temperature, the evaporation rate is iden-
tically zero (no concentration difference) and the evaporative heat transfer coeffi-
cient is zero. On the other hand, when the coffee temperature is precisely equal to
ambient, the evaporation rate is finite, but the temperature difference is zero.

13.3 Evaporation of a Forgotten Cup of Coffee

After the coffee cools to room temperature, in several hours, the event is over,
right? Wrong. The cooling event is over, but the coffee continues to evaporate.
Unless the room is at 100 % relative humidity, the concentration of vapor just above
the free surface will be higher than that in the room, and water vapor will continue
500 13 Evaporation and Mass Transfer Fundamentals

Fig. 13.16 Photo of

evaporation experiment.
A ceramic cup is initially
filled to the brim with water
and placed on a weighing
scale and left in a room,
maintained at nominal room
temperature. The mass and
distance from the free surface
are measured as functions
of time

to diffuse. Left undisturbed, the liquid will all eventually evaporate into the room.
The goal of this section is to predict the evaporation time and compare it to an
experiment.

13.3.1 Experiment

Figure 13.16 shows a photo of the evaporation experiment. A ceramic mug with an
inner diameter of approximately 6.2 cm and inner height of 10 cm is placed on a
food scale. Empty, the mug weighs 0.210 kg. Two hundred and fifty milliliters of
nominal room temperature water is added, filling the mug to the brim. The mug was
placed in a corner of a room (maintained nominally at 22  C, during winter, when
the inside relative humidity is expected to be low) and forgotten about, except for
occasionally, when a data point (mass and distance) was collected. The ambient
conditions were not well controlled.
The experimental results are shown in Fig. 13.17. The length was measured
periodically for some 63 days. The data is plotted and a power law fit applied. The
fit is extrapolated until the length equals the depth of the cup (0.1 m), which shows
105 days to complete evaporation. Unfortunately, the experimentalist forgot to take
data after day 63. . ..
13.3 Evaporation of a Forgotten Cup of Coffee 501

0.1

L(m), distance from free surface to exit plane

0.09
0.08
0.07
0.707229
L = 0.003698(time)
0.06
0.05
0.04
0.03
0.02
0.01
0
0 20 40 60 80 100 120
Time (days)

Fig. 13.17 Measured distance vs. time. A power law fit is applied, and the dotted line
extrapolates until complete evaporation (L ¼ H ¼ 0.1 m). A total evaporation time of 105 days is
predicted

INPUT PARAMETERS
Pinf 1.01E+05 N/m2 atmospheric pressure
Tinf 22 oC ambient temperature
H 0.1 m inner height of mug
D 0.056 m inner diameter of mug
rh 0.3 -- relative humidity
PHYSICAL CONSTANTS
rholiq 1000 kg/m^3 liquid density
Dv 2.56E-05 m^2/sec mass diffusivity of vapor
hfg 2.33E+06 J/kg heat of vaporization
A 9.1496E+10 N/m2 CLAUSIUS-CLAPYRON CONSTANTS FOR WATER
B
B -5.1152E+03 K
DERIVED PARAMETERS Psat(T ) = Ae (T+273)
rhogas 1.20 kg/m^3 gas density
Xvinf 0.0080 molV/mol mol fraction of vapor in ambient
Yvinf 0.0051 massV/mass mass fraction of vapor in ambient
X0 0.0266 molV/mol mol fraction of vapor on free surface
Y0 0.0167 massV/mass mass fraction of vapor on free surface
B0inf 1.17E-02 Transfer number, surface to ambient

Fig. 13.18 Input and derived parameter values for evaporation experiment

13.3.2 Theory

Three different models will be developed and a case study run based on the input
parameters shown in Fig. 13.18, which correspond loosely to the experimental
conditions. For derived parameters, the gas density is obtained from the ideal gas
502 13 Evaporation and Mass Transfer Fundamentals

law. The mole fraction of vapor in the ambient room is given by the relative
humidity times the saturation pressure (from the Clausius–Clapeyron fit) divided
by the total pressure. The mass fraction is obtained by the conversion derived in the
Appendix. The free surface is assumed to be at ambient temperature for this case.
A calculation will show that the free surface is theoretically colder than ambient
(because of evaporative cooling), but the heat transfer response time is so much
faster than the evaporation time, that an isothermal situation is an excellent approx-
imation. This approximation would have to be revisited if the ambient temperature
were close to the boiling point. The mole and mass fractions of vapor at the free
surface, then, represent the saturation state at ambient temperature. The mass
fraction at the free surface is 0.0167, showing that the partial pressure of water is
only about 1.7 % of the total, the balance being air. This behavior is referred to as
the “dilute gas approximation,” where the partial pressure exerted by the vapor is
much less than the total pressure. This behavior holds when the ambient tempera-
ture is sufficiently far from the boiling point, as in this case.
The overall transfer number, B01, is defined by:

Y0  Y1
B01 ¼
1  Y0
This transfer number represents the maximum it could possibly be and has a value
of 0.0117, which shows that the low B limit model applies, that is, for the transfer
number defined between generic surfaces A and B:

Y0  Y1
lnð1 þ B01 Þ B01 ¼
1  Y0
In addition, the air is always dilute in the vapor (Y
1), and therefore the
denominator is approximately equal to unity. Therefore, the low B number, dilute
system approximation is:

B01 Y 0  Y 1

Different evaporation models are developed shortly that all result in a B number,
but with different assumptions about the mass fraction values.

13.3.3 System Mass Balance

The goal of this study is to model the receding surface level, and therefore the
distance between the free surface and the exit plane is not a constant. A quasi-
equilibrium approximation is used, however, in which the rate of evaporation is
determined at any time as if that length were constant.
13.3 Evaporation of a Forgotten Cup of Coffee 503

Application of the conservation of mass principle to the system defined

(as before) to include the mug and the coffee, as a word statement, is:
     
Rate of Change Mass Flow Mass Flow
¼ 
of Mass Stored Rate In Rate Out

Introducing appropriate symbols, recognizing that there are two subcomponents,

mug and coffee, no mass inflow, and the exiting mass flow is the evaporation rate:


d Mcoffee þ Mmug
¼ 0  m_ evap
dt
Since the mug is not evaporating, the rate of change of the mass of the mug is zero,
and the mass balance becomes:

dMcoffee
¼ m_ evap
dt
Expressing the coffee mass as the liquid density (ρLiquid) times the instantaneous
coffee volume (in terms of the length of the air column, L):


d ρLiquid SðH  LÞ
¼ m_ evap
dt
Simplifying and rearranging:

dL m_ evap
¼
dt ρLiquid S

The evaporation rate depends on the model used. The rate of change of the coffee
surface is related to the rate of evaporation and depends inversely on the density of
the liquid and the free surface area (S). This expression will be revisited after
developing models for the evaporation rate. From experience, this height should not
change very much during the initial cooling process. It does change if left unat-
tended a few days.

Early Time Model: L

D
Starting with a completely full mug, the free surface “sees” a nominal hemispherical
air environment around it, with little movement. Modeling the evaporation as a
diffusion process from a hemispherical source, with a surface area equal to that of the
coffee-free surface, the differential equation for L is:

pπffiffiDρ Dv lnð1 þ BL1 Þ

dL m_ evap gas
¼ ¼ 2

dt ρLiquid S ρLiquid πD 4
2
504 13 Evaporation and Mass Transfer Fundamentals

Rearranging:
pffiffiffi
dL 8ρgas Dv lnð1 þ BL1 Þ
¼
dt ρLiquid D

The transfer number for this case is:

YL  Y1
BL1 ¼
1  YL
YL represents the mass fraction of vapor at the exit plane. When the mug is nearly
full, the free surface is near the exit plane, and the mass fraction at the exit plane is
assumed equal to the mass fraction on the free surface in this model limit:

YL Y0

The transfer number becomes equal to the transfer number applied to the system,
from the free surface to ambient:

Y0  Y1
BL1 B01 ¼
1  Y0
Furthermore, the low B number limit for dilute gas systems is applied and the
differential equation for air column length becomes:
pffiffiffi
dL 8ρgas Dv ðY 0  Y 1 Þ
¼
dt ρLiquid D

The velocity of the receding free surface (dL/dt) is a constant in this model, and the
equation can be integrated directly:
"pffiffiffi #
8ρgas Dv ðY 0  Y 1 Þ
L¼ t þ L0
ρLiquid D

L0 is the length of the air column at the start, which will be taken to be zero (a full
mug). The total evaporation time (tevap) can be calculated by setting the length of
the air column equal to the inner height of the mug (H):

ρLiquid DðH  L0 Þ
tevap, LowL ¼ pffiffiffi
8ρgas Dv ðY 0  Y 1 Þ
13.3 Evaporation of a Forgotten Cup of Coffee 505

Inserting values for the case study considered:


1, 000mkg3 ð0:563 mÞð0:1  0Þm
tevap, LowL ¼ pffiffiffi

8 1:22mkg3 2:56ð10Þ5 ms ð0:0266  0:0167Þ
2

   
h day
tevap, LowL ¼ 5:46ð10Þ6 s  
3, 600 s 24 h

tevap, LowL ¼ 63 days

This model predicts the complete evaporation of water from the mug will be about
2 months. It is about 40 days shorter than measured.

Late Time Model: L  D

Once the free surface recedes below the exit plane of the mug, the diffusion of water
vapor through the air column between the free surface and the exit plane is assumed
to become the basic means of evaporation. Inserting the evaporation model for this
diffusion process into the derivative for L:

ρgas Dv Slnð1þB0L Þ
dL m_ evap
¼ ¼ L
dt ρLiquid S ρLiquid S

Rearranging:
" #
dL ρgas Dv lnð1 þ B0L Þ 1
¼
dt ρLiquid L

The transfer number for this case is:

Y0  YL
B0L ¼
1  Y0
YL represents the mass fraction of vapor at the exit plane. When the mug is nearly
empty, the free surface is far from the exit plane, and the mass fraction at the exit
plane is assumed equal to the mass fraction of ambient air in this model limit:

YL Y1

The transfer number becomes equal to the transfer number applied to the system,
from the free surface to ambient:

Y0  Y1
B0L B01 ¼
1  Y0
506 13 Evaporation and Mass Transfer Fundamentals

Furthermore, the low B number limit for dilute gas systems is applied, and the
differential equation for air column length becomes:
" #
dL ρgas Dv ðY 0  Y 1 Þ 1
¼
dt ρLiquid L

In this model, the speed of the receding free surface (dL/dt) depends inversely on
the instantaneous value of the length of the air column (L). As the surface slowly
recedes, the rate of diffusion from the free surface to the exit plane decreases due to
the reduced concentration gradient that drives the mass transfer.
Separating variables (with the factor 2 applied for later convenience) and
applying limits of integration:

ðL ðt " #
2ρgas Dv ðY 0  Y 1 Þ
2LdL ¼ dt
ρLiquid
L¼L0 t¼0
" #
2ρgas Dv ðY 0  Y 1 Þ
L 
2
L20 ¼ t
ρLiquid

This model results in the so-called L2 Law, which has an important counterpart in the
evaporation of liquid droplets. L0 is the length of the air column at the start, which will
be taken to be zero (a full mug). The total evaporation time (tevap) can be calculated by
setting the length of the air column equal to the inner height of the mug (H):


ρLiquid H 2  L20
tevap, Large L ¼
2ρgas Dv ðY 0  Y 1 Þ

Inserting values for the case study considered:



1, 000mkg3 0:12  0 m2
tevap, Large L ¼
 
2 1:22mkg3 2:56ð10Þ5 ms ð0:0266  0:0167Þ
2


  
h
7 day
tevap, Large L ¼ 1:37ð10Þ s  
3, 600 s 24 h

tevap, Large ¼ 158 days

This model predicts the complete evaporation of water from the mug will be about
5 months, more than twice that predicted from the point source model.

Matching Model
In both the early and late evaporation models, the mass fraction of water vapor at
the exit plane was assumed to be at one extreme (free surface) or the other
13.3 Evaporation of a Forgotten Cup of Coffee 507

(ambient). In this final model, these assumptions will be relaxed, and an expression
for the evaporation rate that applies for the entire evaporation event is developed.
To do so, the evaporation rate is expressed in both forms:

π ρgas Dv Slnð1 þ B0L Þ

m_ evap ¼ pffiffiffi Dρgas Dv lnð1 þ BL1 Þ ¼
2 L

Simplifying and rearranging:

 
pffiffiffi L
8 lnð1 þ BL1 Þ ¼ lnð1 þ B0L Þ
D

Applying the low B number limit for dilute gas systems:

 
pffiffiffi L
8 ðY L  Y 1 Þ ¼ ðY 0 þ Y L Þ
D

Solving for YL:

pffiffiffi
L 
Y0 þ 8 Y1
YL ¼ pffiffiffi
DL 
1þ 8 D

Notice that in the limit L/D ! 0, the mass fraction at the exit plane is the free
surface value (Y0), and in the L/D ! 1, the mass fraction at the exit plane is the
ambient value (Y1). The evaporation rate, using the air cylinder model with the
low B, dilute gas approximations is:

ρgas Dv Slnð1 þ B0L Þ ρgas Dv SðY 0  Y L Þ

m_ evap ¼
L L
Substituting for YL:
  pffiffi 
Y 0 þ 8ðDL ÞY 1
ρgas Dv S Y 0  pffiffi
‵1þ 8ðDL Þ
m_ evap
L
pffiffiffi
8ρgas Dv SðY 0  Y 1 Þ ρgas Dv SðY 0  Y 1 Þ
m_ evap ¼
pffiffiffi  ¼
D
D 1 þ 8DL =p8ffi þ LÞ

Now, to enter this expression into the mass balance equation:

ρgas Dv SðY 0 Y 1 Þ

D 
dL m_ evap =pffi8 þL
¼ ¼
dt ρLiquid S ρLiquid S
508 13 Evaporation and Mass Transfer Fundamentals

Separating variables, and applying integration limits, setting L0 to zero (to simplify
subsequent algebra):

ðL ðt " #

D  ρgas Dv ðY 0  Y 1 Þ
=pffi8 þ L dL ¼ dt
ρLiquid
L¼0 t¼0

Performing the integrations results in a quadratic equation for the air cylinder
length (L) as a function of time:
" #
1 2 D ρgas Dv ðY 0  Y 1 Þ
L þ pffiffiffiL  t¼0
2 8 ρLiquid

Solving for L:
vffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
u 2 " #
D uD 2ρgas Dv ðY 0  Y 1 Þ
L ¼ pffiffiffi þ t þ t
8 8 ρLiquid

The total evaporation time is obtained by setting L to H:


ρLiquid H2 pffiffi
þ DH
2 8
tevap, Matched ¼
ρgas Dv ðY 0  Y 1 Þ

Rearranging:

ρLiquid H 2 1 þ HDpffiffi2
tevap, Matched ¼
2ρgas Dv ðY 0  Y 1 Þ

Notice that this result reduces to the point source model (L

D) when D  H and
to the air cylinder model (L  D) when H  D. Evaluating for the case study:


1, 000mkg3 ð0:1 mÞð0:1 m þ 0:056 mÞ
tevap, Matched ¼

2 1:22mkg3 2:56ð10Þ5 ms ð0:0266  0:0167Þ
2

   
h day
tevap, Matched ¼ 1:95ð10Þ7 s  
3, 600 s 24 h

tevap, Matched ¼ 225 days

13.3 Evaporation of a Forgotten Cup of Coffee 509

Point Source Model Air Cylinder Model

0.1
0.09
Length of Air column, L (m) 0.08
Matching Model
0.07
0.06
0.05
0.04
0.03
0.02
0.01
0
0 50 100 150 200 250
Time (days)

Fig. 13.19 Air column length vs. time. Model predictions and experimental data

13.3.4 Results and Discussion

Figure 13.19 shows the length L as a function of time for the three models, and with
the experimental data. For the point source model, the air cylinder length increases
linearly, while the mass (not shown) decreases linearly. When all of the water is
evaporated (L ¼ H), the event is over. The air diffusion cylinder model predicts an
initial vertical slope and subsequently the slope decreases with time as the concen-
tration gradient weakens. If the mug in question were twice as deep as the one used,
the difference between the point source and air diffusion cylinder would be much
more dramatic. The final model matches the air cylinder to the point source. The
simulation follows the point source model initially, and as time proceeds, gradually
takes the shape of the air cylinder model, shifted in time to account for the initial
period where the model breaks down.
The experimental points appear to follow the air cylinder model most closely.
However, the functional dependence of length with time is different. Figure 13.20
shows the three theoretical curves plotted, and then a power law fit to them. The
point source and air cylinder models exhibit the linear and square root dependencies
as expected. However, the matched model, whose solution does not conform to a
simple power law fit, exhibits a power law exponent of 0.69, which agrees well with
the experimental value of 0.707 (from Fig. 13.17). Therefore, it is concluded that
the more involved matched model exhibits a similar functional dependency as the
experiment, but an overall slower evaporation rate.
Figure 13.21 shows the mass fraction at the exit plane (YL) as a function of time.
The point source model forces the mass fraction to be the saturated surface value.
The air cylinder forces it to be the ambient value. For the matching model, the exit
510 13 Evaporation and Mass Transfer Fundamentals

0.12 Point Source Model Air Cylinder Model

y = 0.0016x y = 0.0079x0.5
0.1
Length of Air column, L (m)
Matched Model
0.08
y = 0.0025x0.6907

0.06

0.04

0.02

0
0 50 100 150 200 250
Time (days)

Fig. 13.20 Model predictions, with power law fits. For each model, the dashed or dotted line is
the model prediction, the light line is a power law fit, with the equation shown

0.018 Point Source Model

0.016
Mass Fraction at Exit Plane, YL

0.014

0.012

0.01
Matching Model
0.008

0.006 Air Cylinder Model

0.004

0.002

0
0 50 100 150 200 250
Time (days)

Fig. 13.21 Predicted mass fraction at the exit plane for the three evaporation models as a function
of time

plane is intermediate, and the mass fraction there gradually varies as the instanta-
neous length to diameter ratio changes. The mass fraction drops rapidly and very
gradually approaches the ambient value. Note that this particular mug has a modest
L/D ratio of 1.8.
The experiment shows a more rapid evaporation process than predicted by the
matching model for the assumed experimental conditions. However, the theoretical
13.3 Evaporation of a Forgotten Cup of Coffee 511

400

350
Evaporation Time (days) 0.8
300

250 0.6
0.4
200 0.2
rh = 0
150

100

50

0
10 15 20 25 30 35 40
Ambient Temperature (oC)

Fig. 13.22 Evaporation time as a function of ambient temperature and relative humidity for the
matched model

prediction is very sensitive to the assumed ambient temperature and humidity, as

shown in Fig. 13.22, which shows the predicted evaporation time as a function of
ambient temperature for various relative humidity values.
Perhaps a bigger deviation from the model would be due to air currents in the
room, both induced by the natural uncontrolled movements in the room, as well as
those created by the mug of water itself. Figure 13.23 shows a suggested mecha-
nism. Since the molecular weight of water is less than that of air, the density of the
humid air inside the mug is less than that of the room. The effect is small, from the
point of view of generating an updraft, but any updraft would have a profound
effect on the concentration field, and thus evaporation rates. The point source model
is one which would be appropriate for a completely still room, in the absence of
gravity. So it is reasonable to expect somewhat faster evaporation than predicted by
these models, whose main goal is to demonstrate the fundamental processes
associated with mass transfer.

13.3.5 Surface Temperature

The final resistance network of this textbook is one that shows that evaporation
causes the surface temperature of the evaporating mug of coffee, left unattended in
a room, to be below the ambient temperature, in apparent violation of the Second
Law. Figure 13.24 shows a resistance network to demonstrate the basic heat and
mass transfer channels. There are no capacitances shown, as this situation is after
512 13 Evaporation and Mass Transfer Fundamentals

Fig. 13.23 Suggested

mechanism for enhanced
evaporation rates compared to
the one-dimensional models

Fig. 13.24 Resistance

network for estimating the
coffee and surface
temperature of an evaporating
mug of coffee

the cooling period is over, and any transient effects are considered to be unimpor-
tant, an assertion that could be justified with a full transient analysis.
The free surface of the coffee experiences three heat transfer channels: convec-
tive (Rh) and radiative (Rr) parallel channels directly to ambient and a convective
channel to the bulk of the coffee (Rc,s). The coffee in turn is in thermal contact with
ambient (Rc,1 which is a complex channel through the wall and to ambient). The
evaporation process is shown as a sink, since it is mass transfer driven (not
temperature). It is this evaporation process that acts as a cooling mechanism for
the surface, which must be below ambient in order for heat to flow into it.
A nodal balance on the free surface yields the temperature of the free surface,
noting that the series channel from ambient through the coffee is used:
Workshop 13.1. Evaporation Experiments 513


T1 1
Rrþ R1h þ Rc, 11þRs, c  m_ e 
hfg
Ts ¼ 
Rr þ Rh þ Rc, 1 þRs, c
1 1 1

This equation is implicit, since the evaporation rate is a function of the surface
temperature. Once thermal resistances are approximated, this equation is in a form
suitable for the Method of Successive Substitutions.
The coffee temperature, being a voltage divider between the surface and ambient
can be obtained next, as:

TS
R S, C þ RTC1, 1
T coffee ¼ 1
R S, C þ RC1, 1

The effect can be enhanced by using an insulated mug, in which case Rc,1 can be
made large compared to all other thermal resistances. The limit as Rc,1 goes to
infinity is:

T s ¼ T 1  Req, s1 m_ e 
hfg

T coffee ¼ T s

where the equivalent resistance between the surface and ambient is:

1 1 1
¼ þ
Req, s1 Rr Rh

Numerical determination of this effect is expected to show a small, perhaps

immeasurable temperature drop, and is reserved for the curious reader to tie
together in a Workshop.

Workshop 13.1. Evaporation Experiments

During the first week of class, procure samples of clear tubing. Choose samples
with a much larger length to diameter ratio than the value of 1.8 investigated in this
chapter. Mount them vertically with an open end up top, and sealed at the bottom.
Mark them with a scale. Find locations for them where they will not be disturbed,
fill them to the top with water (eliminate all air bubbles trapped), and start recording
L vs. t. Consider using different liquids, and note that if liquid mixtures are used
with components of different volatility (i.e., isopropyl alcohol/water mixtures), that
the composition will change with time. Modeling this effect requires some basic
understanding of distillation-type processes. Compare results to theoretical
predictions by the end of the semester.
514 13 Evaporation and Mass Transfer Fundamentals

Workshop 13.2. Surface Temperatures During Evaporation

Calculate the surface and coffee temperatures for an evaporating mug of coffee as a
function of ambient temperature (assume zero relative humidity). Demonstrate that
the effect becomes very large as the temperature approaches the boiling point of the
liquid. Use the point source model as giving the largest effect modeled here.
Alternatively, use published correlations for the mass transfer coefficients.

Workshop 13.3. A Cooling Mug of Coffee with Evaporation

Effects

Include a lumped capacitance to the coffee node of Fig. 13.24 and include the
effects of evaporation onto the basic cooling response of the coffee. Use a well-
insulated mug and treat the thermal resistance between coffee and ambient (Rc,1)
as a constant. Use either an evaporative heat transfer coefficient (hevap) or directly
include the evaporation as a sink. Implement an implicit Euler method in which the
surface temperature, evaporation rates, radiation, and natural convection are all
calculated (iteratively) at each time step. Start with the coffee at its boiling point,
and notice how much lower the surface temperature is than the average coffee
temperature early on. Plot the dimensionless temperature vs. time on semi-logy
axes to observe deviations from first-order behavior.

Appendix. Evaluating Dew Points and Mass Fractions

An important detail in evaluating evaporation rates is to relate the mass fractions in

terms of input variables, which requires a little more background into modeling gas
mixtures. The gas phase is modeled as being a mixture of two ideal gases, air and
water vapor.
For humid air at temperature T, the partial pressure of water is given by:

pV ¼ r h psat ðT Þ

where rh is the relative humidity.

The dew point temperature (TDewPoint) of air with a relative humidity (rH) at an
ambient temperature (T1) is defined by the temperature at which the partial
pressure of saturated vapor equals the partial pressure of vapor in the humid (but
unsaturated) air. That is:

psat ðT DewPoint Þ ¼ r h psat ðT 1 Þ

Appendix. Evaluating Dew Points and Mass Fractions 515

The relationship between the saturation pressure and the temperature can be
modeled using a curve fit of the form given by the Clausius–Clapeyron relation,
namely:
B
psat ðT Þ ¼ AeT

where A and B are constants for a given species, and T is the absolute temperature.
The values of A and B can be calculated by fitting two points taken from saturation
tables in the vicinity of the problem in question. This curve for water was plotted in
Fig. 13.2.
The partial pressure of water vapor is related to the concentration of water vapor.
For nv moles of water vapor at absolute temperature T occupying a volume V, the
ideal gas law is:

pv V ¼ nv R^ T
J
R^ ¼ 8, 314
kmol K
is the universal gas constant. Multiplying and dividing by the total number of moles
and rearranging:

n nR^ T 
v
pv ¼ ¼ Xv P
n V

where P is the total pressure (the sum of all partial pressures). The mole fraction of
vapor, Xv is the number of moles of vapor divided by the total number of moles in
the gas mixture. The mass fraction and mole fractions are related to each other
through the molecular weights of the mixture and vapor by the following word
string:
     
Mass of Vapor Moles of Vapor Moles of Mixture Mass of Vapor
¼
Mass of Mixture Moles of Mixture Mass of Mixture Molesof Vapor
 
1

Y v ¼ ðX v Þ MWvapor
MWmix

The molecular weight of the mixture is obtained from the following word string:
   
Mass of Mixture Mass of Vapor þ Mass of Air
¼
Moles of Mixture Moles of Mixture
516 13 Evaporation and Mass Transfer Fundamentals

    
Mass of Mixture Moles of Vapor Mass of Vapor
¼
Moles of Mixture Moles of Mixture Moles of Vapor
  
Moles of Air Mass of Air
þ
Moles of Mixture Moles of Air

MW mix ¼ Xvapor MW vapor þ Xair MW air

 
MWvapor
Y v ¼ ðX v Þ
Xv MWvapor þ ð1  Xv ÞMW air

1
Yv ¼  
1þ 1Xv
Xv
MW air
MW vapor