ONE DIMENSIONAL TRIANGULAR FIN EXPERIMENT

Current Technical Advisor: Dr. Kelly O. Homan

Original Technical Advisor: Dr. D.C. Look

(Updated on 26th August 2009, Original version: 3rd November 2000)

1 GENERAL OBJECTIVES

a) To understand a one-dimensional experimental approximation.

b) To understand the art of experimental measurement; in particular, the judicious

use of data.

c) To learn a practical method of measuring a convective heat transfer coefficient for

a triangular fin using a statistical analysis. Particularly, the MATLAB API

(application programmer’s interface) and the Levenberg–Marquardt algorithm to

solve non-linear least squares problems.

2 INTRODUCTION

While fins are used everyday, it is tempting to believe that the simplified

analysis presented in basic textbooks is not a “real world” description or that it

is an unrealistic approximation of what might be observed in the laboratory.

This experiment is designed to demonstrate that the fin concept is

straightforward and accurate.

3 THEORY

Consider a section of the one-dimensional triangular fin as in Figure 1. Note that

the root is at a known temperature, TW, and that the ambient is at a known

temperature, T∞. Heat is lost from the fin by convection, the rate of which is

proportional to the heat transfer coefficient, h, of the slanted surface. Presuming

conduction with-in the fin to be primarily considered one-dimensional (i.e., the z

dimension is effectively infinite and the perimeter can be approximated as 2z

since l <<z), an energy balance for a differential slice of the fin takes the form (1)
 d  dT  ds
 kA( x)  = hp ( x)(T − T∞ ) (1)
dx  dx  dx

where T(x) is the temperature of the fin at a particular dimensional position x (refer “Heat
Transfer from Extended Surfaces” section in [1]).

In order to simplify the expression, non-dimensionalize by introducing the following definitions:

x
θ = T − T∞ , x∗ =
L
Taking k and h to be constants, the expression simplifies to
1

d  ∗ dθ  hl  L 
2
  l 2  2 (2)
x =   1 +    θ
dx∗  dx∗  k  l    L  

hl  L    l   2
2 2

Further defining B =   1 +    , expression (2) simplifies to

k  l    L  

d  ∗ dθ 
x  = Bθ (3)
dx∗  dx∗ 

on expanding:
d 2θ dθ
∗
x + − Bθ = 0
dx∗2 dx∗

which may be recognized as a second-order ordinary differential equation with variable

coefficients subject to the following boundary conditions:

θ ( x∗ = 0) = finite (4)
θ ( x∗ = 1) = θ0 (5)

where θ0 ≡ Tw − T∞ and the solution is

θ I0 2 Bx
=
( ∗
) (6)
θ0 I0 2 B ( )
on the right hand side, I0 is the modified Bessel function of zeroth order.

At this point, if all of the dimensional parameters (h, l , k, L, TW and T∞ ) were known, the

axial temperature profile for the fin would be known as a function of x. Of the parameters
listed, h, the convection heat transfer coefficient is the most difficult to determine,

especially since its value depends entirely on the flow field characteristics. The present

experiment seeks to estimate h from measured quantities.

This method is based upon a statistical analysis of dimensionless position, x∗ , and

dimensionless temperature, θ data using equation (6). The selected method is referred to

as the Method of Non-Linear Least Squares (NLLS) [4,5], a form of non-linear

regression. Based upon the assumption that the errors in the experimental measurements

follow a Gaussian distribution, the NLLS produces a unique value for the determined

constants. That is, the magnitudes of the constants that are determined give the “most

probable” form of the given equation that fits the data. The NLLS is based upon the

differences between the independent values of collected experimental data and the

expected values as provided by the theoretical result.

For convenience, let θi = θ / θ 0 , then from the theory at each position xi,

(
θi = DI 0 2 Bxi∗ ) (7)

where (
D = 1/ I0 2 B ) is a constant to be determined so as to fit that experimental data.

For simplicity, let (

I i = I 0 2 Bxi ∗ ) (8)

In this case, the residual for each data point is

ρi = θi − DI i (9)

Note that the dimensionless temperature θi , and the dimensionless position xi ∗ (or the

modified Bessel function at a specific position, I i ) are the input data. Then the sum of the

residuals squared is S = ∑ ρi 2 (10)

The value of h that would give the least value of residuals squared sum S is found by
writing a MATLAB program using the Levenberg–Marquardt algorithm.

Refer to Appendix A at the end for MATLAB API (application programmer’s interface) on
the Levenberg–Marquardt algorithm.
4 EXPERIMENTAL SETUP

The configuration and schematic of the experimental setup (1-D triangular

fin) is presented in Figs. 1 and 2 respectively. The sides and the bottom are all

well insulated. This setup represents a physical approximation to the upper half

of a symmetric, infinitely long fin. Figure 3 presents the actual experimental fin

information.

The temperature indicators are thermocouples embedded within the

stainless steel fin as indicated in Figure 4.

5 DATA ACQUISITION PROCEDURE

a. Familiarize yourself completely with the experimental setup. Do NOT,

under any circumstances, touch the stainless steel surfaces because the

setup will be turned on before class. Figure 3 is included so that you may

view the setup without taking it apart and subjecting yourself to harm.

b. Confirm that the temperature profile is at steady state (not a function of

time). Do not adjust the heater temperature setting. You should have data

to prove that steady state exists.

c. Take the data as per the included data sheet, including T∞. Repeat the

data collection process to confirm the steadiness of the temperatures.

6 DATA REDUCTION PROCEDURE

a. First plot T (or θ) versus x in order to check the data for the appropriate

shape as well as to check its one-dimensional character. Note that Tw is

the arithmetic average of the temperature readings at the fin root.

 b. If the shape of the temperature variation is appropriate, enter the data sets

(Ti and xi) and T∞ into the MATLAB program to solve for ‘h’ using

Levenberg–Marquardt algorithm (non-linear least squares method).

c. Run the MATLAB program and read the convection coefficient, h.

7 POINTS OF INTEREST

In your technical memorandum you should include a discussion of the following:

1) The physical significance of the boundary conditions listed by Equations

(4) and (5).

2) Explain clearly how insulation on the two sides simulates a fin which is

infinitely long in the z direction.

3) Why is it important for the fin to be at steady state?

4) Plot the resulting model (Ti from θi using equation (7) with h value from

MATLAB program and calculated values of D and B using h) and the

measured steady state temperature versus x . Discuss the accuracy of

the model.

5) As an appendix, consider a differential element of the length dx of the

experimental fin and obtain the governing equation for the fin. Apply

the boundary conditions and hence obtain Equation 6. State the

assumptions you made in the process of obtaining Equation 6. Are they

all justified for the experimental setup used? Explain. (refer [1])
REFERENCES

1. Incropera, F.P., Fundamentals of Heat and Mass Transfer, John Wiley & Sons, Inc.,

New Jersey, 2007, pp. 137-162.

2. Chapman, A.J., Heat Transfer, Fourth ed., Macmillan Publishing Company, New

York, Collier Macmillan Publishers, London, 1984, pp. 67-89.

3. Hildebrand, F.B., Advanced Calculus for Applications, second ed., Prentice-Hall,

Inc., Englewood Cliffs, New Jersey, 1976, pp. 141-153.

4. Bain, L.J., and M.Engelhardt, Introduction to Probability and Mathematical Statistics,

PWS Publishers, 1987, pp. 300-307.

5. Lipson, C., and N.J. Sheth, Statistical Design of Engineering Experiments, McGraw-

Hill Book Company, 1973, pp. 372-387.

6. Levenberg–Marquardt algorithm, www.wikipedia.org

NOMENCLATURE

h = surface convection coefficient for a triangular fin (W/m2 0C)

k = thermal conductivity (W/m 0C)

l = one half fin thickness at the root (m)

L = fin length (m)

s = slant length (m)

T = fin temperature (0C)

TW = fin root temperature (0C)

T∞ = ambient temperature (0C)

x = dimensional axial coordinate along fin length (m)

x∗ = x /L, dimensionless axial coordinate

y = coordinate along fin height

z = coordinate along fin width

A( x ) = cross-sectional area parallel to the wall and enclosed by the perimeter

(i.e. 2zy=2zx /L )

p(x) = perimeter of the cross sectional area of the triangular fin (i.e. 2(z+2y) = 2z, as

z>>2y from a thermodynamic point of view)

ds = ((dx) 2 + (dy) 2 )1/2

θ = (T - T∞)

θ0 = T W - T ∞
 One Dimensional Triangular Fin Experimental Data Sheet

Date: TW=

Name: T∞ =

TC# 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16

Depth from Fin

0 1.2 0 0.1 0.5 0.8 0.3 0 1.0 0.4 0.6 0.2 1.5 0 0.9 0.4
Bottom (in)

Distance from root

0 1.25 2.5 4.5 0 1.25 2.5 3.75 0 1.25 2.5 3.75 0 1.25 2.5 3.75
(in)

Temperature (F)

TC#
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16
time
 z y
T∞ h
Tw ds
l dy

x dx
x=L x=0
 protective cover box
(heater inside)

heater on back
surface

control panel used to

turn on the experiment

thermocouple selector
switch allows from
only one thermocouple
cable between selector
to be displayed
device and display device

temperature
display device
 APPENDIX A

MATLAB API (application programmer’s interface) for the Levenberg–Marquardt

algorithm to solve non-linear least squares problems.

Example: Rosenbrock function - in mathematical optimization, the Rosenbrock function is

a non-convex function used as a test problem for optimization algorithms. The global
minimum is inside a long, narrow, parabolic shaped flat valley. To find the valley is trivial,
however to converge to the global minimum is difficult. The Rosenbrock function is
defined as

f ( x, y ) = (1 − x) 2 + 100( y − x 2 ) 2

To test if Levenberg–Marquardt algorithm is able to find the global minimum (1,1):

create a new file in MATLAB with the following commands (make sure the new
MATLAB file and LMFsolve.m, Levenberg–Marquardt algorithm are in the same working
directory). In this case m=[x;y], m(1)=x and m(2)=y.

rosenbrock= @(m) [1-m(1); 10*(m(2)-m(1)^2)];

[m,residuals_squares_sum,iterations]=LMFsolve(rosenbrock,[-1.2,3],'Display',1,'MaxIter',15000)

“1-m(1)” is the first function and “10*(m(2)-m(1)^2)” is the second function

with Levenberg–Marquardt algorithm we are trying to find the minimum of the squares-
sum of these two functions.

[-1.2,3] is the initial guess value for m

“'Display',1,'” is for displaying iteration information after every iteration

“'MaxIter',15000” is to set the maximum number of iterations to 15000

Running the program will return “m = [1; 1]” i.e, m(1) = x = 1 and m(2) = y = 1 (in
approximately 4000 iterations) which is the global minimum for the Rosenbrock function.

Fin Experiment: Develop a new MATLAB program for this experiment with the help of
the example as shown above.

Hint: The important commands are-

residuals= @(h) [ρa;ρb;ρc;ρd;ρe];

[h,residuals_squares_sum,iterations]=LMFsolve(residuals,[100],'Display',1,'MaxIter',15000)

where ρ is the error (equation 9) at different x values as a function of h

“100” is the initial guess value for h