Formulation of mathematical

models

Francisco Trujillo
CEIC3000 Process modelling and
Analysis
 What is mathematical modelling
• Is a form of imitating nature in mathematical terms.
• Modelling is an attempt to reproduce the details of
reality.
• If the model is correct, it can predict the behaviour of
reality at conditions that have not been experimentally
tested.
• The pragmatic reality is that mathematics is a very
effective method to know and understand what we see
around us.
• Of course a mathematical model must be validated to
assure that is correct.
 How to formulate a mathematical
model
1. Have clarity on you mind of what the problem is about
– To know as much as you can about the problem beforehand.
– To draw a picture of the system to be studied.
2. Bring:
– physical and chemical information,
– assumptions
– Conditions, parameters, properties
– conservation laws
– Rates of expressions

At this point is important to convert mental images to symbols (Mathematical

symbols)

Clarify the real scope and deep of the model

3. Setting down of finite or differential volume elements and to write

conservation laws.
 Example 1
heat transfer - plug flow pipeline
Step 1: Have clarity on you mind of what the problem is about?
 What is plug flow? It means that the fluid velocity has the shape of a
plug – it is uniform at all radial positions.
 This almost always means turbulent flow – fluid elements are well
mixed in the radial direction so velocity and temperature are uniform
in a plane normal to the flow field.
 Example 1
heat transfer - plug flow pipeline
Step 2: Bring physical and chemical information assumptions etc:
1. The velocity profile has a Plug shape (velocity is constant and does not
change in the z direction)
2. The fluid is well mixed (highly turbulent), so the temperature is uniform
in the radial direction.
3. The system is under steady state
4. Physical properties are constant (temperature independent)
5. The wall temperature Tw is constant and uniform ( does not change in z
or r direction)
6. The inlet temperature To is constant and uniform (does not change in r
direction)
7. Thermal conduction of heat along the axis is small relative to advection.
Hence, the convective movement of heat on the fluid is due only to
advection.

Scope:
calculate the temperature distribution along z for a plug flow in a tube according the
assumptions above
 Example 1 – plug flow
Step 3: Setting down of finite or differential volume elements and to write
conservation laws.
Write the conservation of energy law on the control
volume (sketch on the left)

Heat in – heat out + rate generation = rate of

accumulation (1.1)

Because steady state, then accumulations is zero.

Heat generation is also absent.

The heat moves trough the tube by convection (advection

only) .

The only way heat can be exchanged with the outside is

through the external area of the tube (on the element)
( 2 Rz ). That heat can be expressed as:

Q   2 Rz  h T  z   Tw  (1.2)

 Example 1 – plug flow
h is the heat transfer coefficient.
The bar over T represents the average between T(z) and T(z+Δz)
T  z   T  z  z 
T  z  (1.3)
2
In the limit as Δz ͢ 0 then
lim T  z   T  z  (1.4)
z  0

Along the axis, heat can enter and leave the element only by advection so
equation 1.1 can be rewritten as:

v0 A C p T  z   Tr   v0 A C p T  z  z   Tr    2 Rz  h T  Tw   0 (1.5)

Rate of heat Rate of heat Rate of heat loss

Flow in Flow out Through the wall
 Example 1 – plug flow
Tr, the reference temperature can be cancelled from eq. (1.5).

v0 A C pT  z   v0 A C pT  z  z    2 Rz  h T  Tw   0 (1.6)

Equation 1.6 is rearranged as:

T  z  z   T  z 
v0 A C p   2 Rh  T  Tw   0 (1.7)
z
Taking the limit as Δz ͢ 0 yields:
dT
v0 A C p  2 Rh T  Tw   0 (1.8)
dz
As you can see the minus signs are cancelled.
Before solving the equation is good practice to group parameters:
dT
  T  Tw   0 (1.9)
dz
Where: 2 Rh
 (1.10)
v0 A C p
 Example 1 – plug flow
dT
  T  Tw   0
dz (1.9)
Equation 1.9 is a linear, inhomogeous equation of first order, which as seen in
next presentation can be solved using the Integration-Factor method.
However, after some inspection we can easily find a solution to this equation.
Defining
  T  Tw (1.11)
Equation 1.9 reads
d
   0 (1.12)
dz
This can be integrated by separation of variables:
d d
  dz     dz ln     z  c   e   z ec   Ke   z (1.13)
 
Where K  ec is the integration constant, which can de determined with a
boundary condition.
 Example 1 – plug flow
Boundary condition:
T  To (1.14)
z 0

Hence,
 z 0  To  Tw   o (1.15)
So from equation 1.13 the constant K is:
 o  Ke   z z o
 K  o
 (1.16)

So equation 1.13 takes the form of

  o e  z (1.17)
Or
 2 Rhz 
T  Tw 
 v0 A C p


e  
(1.18)
T0  Tw
 Example 1 – plug flow
 2 Rhz 
T  Tw 
 v0 A C p


e  
(1.18)
T0  Tw
So the dimensionless temperature can be defined as:
T  Tw
 (1.19)
T0  Tw
And the dimensionless length scale
2 Rhz (1.20)
 
v0 A C p
So a problem with 6 parameters, two external conditions (T0, Tw), and two
variables, one independent (z) and one dependent (T), can be reduced to only
two dimensionless variables related as follows:

  e   (1.21)
 Example 2 – parabolic velocity
Step 1: Have clarity on you mind of what the problem is about?
 In previous example “plug flow” implied turbulent flow. In this case
the flow is laminar (Re < 2100), hence the velocity profile is
parabolic:
vz  2vo 1   r R  
2
  (1.22)

Where vo is the average velocity, r is the radial position, R is the

radios of the tube, and vz denotes local velocity at position r.
(Equation 1.22 will be derived in Advanced Transport Phenomena)
 Example 2 – parabolic velocity
Step 2: Bring physical and chemical information assumptions etc.:
1. The velocity is a function of r (eq. 1.22) following a parabolic shape.
2. The fluid is NOT well mixed in the radial direction, so radial heat
conduction must be accounted.
3. The system is under steady state
4. Physical properties are constant (temperature independent)
5. The wall temperature Tw is constant and uniform ( does not change in z
or r direction)
6. The inlet temperature To is constant and uniform (does not change in r
direction)
7. Velocity is slow (Laminar) therefore advection is slow. In this case heat
conduction may also be important and needs to be accounted.
 Example 2 – parabolic velocity
Step 3: Setting down of finite or differential volume elements and to write
conservation laws.
Under this new conditions the “control
volume” take the shape of an annulus .

As seen in the bottom figure, heat now

crosses two surfaces, the annular area
(normal to fluid flow), and the area along the
perimeter of the annulus (parallel to the fluid
flow).

Both heat fluxes (heat per unit of area) can be

represented as
qr  r , z   Radial heat flux (1.23)
qz  r , z   Axial heat flux (1.24)
 qr  r , z   Radial heat flux (1.23)
qz  r , z   Axial heat flux
(1.24)

Example 2 – parabolic velocity

Step 3: Setting down of finite or differential volume elements and to write
conservation laws.
The net rate of heat flux is the heat flux
(equations 1.23 – 1.24) times the normal area of
the control volume.

Write the conservation of energy law on the

control volume (sketch on the left)

Heat in – heat out + rate generation = rate of

accumulation (1.25)

Because steady state, then accumulations is zero.

Heat generation is also absent.

 qr  r , z   Radial heat flux (1.23)
qz  r , z   Axial heat flux
(1.24)

Example 2 – parabolic velocity

Step 3: Setting down of finite or differential volume elements and to write
conservation laws. (1.26)
Heat in – heat out = 0

Been more specific, heat move by convection

(advection + conduction) in the direction parallel
to the flow only.

But heat also moves by conduction on the radial

direction. So equation 1.26 becomes

Heat in advection + Heat in conduction (radial) +

Heat in conduction (axial) –
{ heat out advection + Heat out conduction
(radial) + Heat out conduction (axial) }
=0
(1.27)
 Example 2 – parabolic velocity
Step 3: Setting down of finite or differential volume elements and to write
conservation laws.
Area perpendicular to the flow (annular region).
Let’s call that axial area (Az) because is the area
perpendicular to the axial movement of the flow
and heat
Az    r  r    r 2
2

Az    r 2  2r r  r 2    r 2
In the limit when Δr ͢ 0, Δr2 becomes insignificant
Az  2 r r (1.28)
Area parallel to the flow. Let’s call that area radial
(Ar) because is the area perpendicular to the radial
flow of heat
Ar  2 r z (1.29)
The mass flow rate (kg/s) passing the annular
region, which is the flow in the axial direction is
mz  vz Az  (1.30)
Where ρ is density
 Az  2 r r (1.28)
mz  vz Az  (1.30)

Example 2 – parabolic velocity

Step 3: Setting down of finite or differential volume elements and to write
conservation laws.
If your remember from thermodynamics, the heat
content of a mass “m” is
(1.31)
Q  mC p T  Tr 
Where Q is the total heat in Jules, Cp is the specific
heat capacity, m is the mass in kilograms, T is
temperature and Tr is the reference temperature.
If the mass is not static but is moving at certain
velocity, then it will move heat as well, that is
advection (many times just called convective heat )
Q  mC p T  Tr 
Where Qcv is the advective heat in (J/s) and m is (1.32)
mass flow rate (kg/s).
Replacing eq. 1.30 and 1.28 into eq. 1.32 the heat
by advection in the axial direction is:

Q  v z Az  C p T  Tr   v z  2 rr   C p T  Tr  (1.28)
 Example 2 – parabolic velocity
Step 3: Setting down of finite or differential volume elements and to write
conservation laws.
There is no advection in the radial direction
because the flow moves axially only.
As explained before, the net heat by conduction
is the rate of heat flux times the normal area.
Hence, the net radial heat flow is
qr  r , z   qr Ar   2 r z  qr (1.29)
The net axial heat flow is
qz  r , z   qz Az   2 r r  qz (1.30)
Calling equation 1.27
Heat in advection + Heat in conduction (radial) +
Heat in conduction (axial) –
{ heat out advection + Heat out conduction
(radial) + Heat out conduction (axial) }
=0
(1.27)
 Example 2 – parabolic velocity
Step 3: Setting down of finite or differential volume elements and to write
conservation laws.
Heat in advection + Heat in conduction (radial) +
Heat in conduction (axial) –
{ heat out advection + Heat out conduction (radial)
+ Heat in conduction (axial) } (1.27)
=0
where each term is as follows:
Heat in advection : from equation 1.28
Q  v z  2 rr   C p T z  Tr  (1.31)
Where T z is the temperature at z position.
heat out advection :
Q  v z  2 rr   C p T z z  Tr  (1.32)
Where T z z is the temperature at (z + ∆z)
Heat in conduction (radial): From eq. 1-29
qr  r , z   qr Ar   2 r zqr  r (1.33)
 Example 2 – parabolic velocity

Step 3: Setting down of finite or differential volume elements and to write

conservation laws.

Heat out conduction (radial) :

qr  r  r , z    2 r zqr  r  (1.34)
r

Heat in conduction (axial) : From eq. 1-30

(1.35)
qz  r , z    2 r rqz  z
Heat out conduction (axial) :
qz  r , z  z    2 r rqz  z  (1.36)
z

Putting equations 1.31 to 1.36 into 1.27

Heat in advection + Heat in conduction (radial) + Heat in conduction (axial) –
{ heat out advection + Heat out conduction (radial) + Heat out conduction
(axial) } =0
(1.27)
 Example 2 – parabolic velocity

Heat in advection - heat out advection +

Heat in conduction (radial) - Heat out conduction (radial) +
(1.37)
Heat in conduction (axial) – Heat out conduction (axial)
=0
Gives
vz  2 r r   C p T z  Tr   z  2 r r   C p T z z
 Tr  
 2 r zqr  r   2 r zqr  r  r


 2 r rqz  z   2 r rqz  z  z
0 (1.38)

Terms with Tr cancel each other then

vz  2 r r   C p T z  z  2 r r   C p T z z

 2 r zqr  r   2 r zqr  r  r


 2 r rqz  z   2 r rqz  z  z
0 (1.39)
 Example 2 – parabolic velocity

Dividing each term of equation 1.39 by 2r z

T z T  rqr  r   rqr  r   rqz  z   rqz  z  (1.40)

vz  C p r z z
 r
 z
0
z r z
Notice that you cannot divide by r
because the terms rqr change
Multiplying by (-1) with r. So you must keep r
T T  rqr  r    rqr  r  rqz  z    rqz  z together with qr.
vz  C p r z z z
 r
 z
0 (1.41)
z r z

Taking the limit when Δz ͢ 0 and Δr ͢ 0

T   rqr    rqz  T   rqz    rqr 
vz  C p r   0 vz  C p r   (1.42)
z r z ; or z z r
Because r is independent of z, it can came out of the qz term
T   qz    rqr    qz    rqr  T
vz  C p r  r  r   vz  C p r (1.43)
z z r z r z
 Example 2 – parabolic velocity

After dividing by r
  qz  1   rqr  T
   vz  C p (1.44)
z r r z
At this stage equation 1.44 is insoluble as it is only one equation with 3
unknowns (T, qr , qz ). To solve this, we can introduce the Fourier’s equation for
heat conduction
q   k T (1.45)
Which on radial coordinates has two components
T
qr  k
r (1.46)
T
qz   k (1.47)
z
Replacing equations 1.46 and 1.47 into equation 1.44 gives
 T   T 
   r 
k 
z  k 1  r   v  C T
z r z
z p
r (1.48)
 Example 2 – parabolic velocity

 T   T 
   r 
k 
z  k 1  r   v  C T
(1.48)
z r z
z p
r
 T 
 
z  k T
2
k 
z z 2  T 
r vz  2vo 1   r R  
2
  
1  r   k 1   r T 
k  
r r r r  r 

Putting all the terms together gives

 2T 1   T  2 T
k 2 k r   2v  C 
p 1   r R  
 z (1.49)
z r r  r
o
 or
 2T   2T 1 T  2 T
k 2 k 2    2v  C 
p 1    
r R (1.50)
z  r r r  z
o

This is a second order partial differential equation that needs to be solved with
numerical methods unless we simplify it somehow
 Boundary conditions
• For most problems occurring in chemical engineering, there are three
types of boundary conditions.

• The three classical types of boundary conditions at a point xo are:

• Type (i) (Dirichlet boundary) y  x  ; @ xx (1.51a)
o

• Type (ii) (Neumann) dy

 ; @ x  xo (1.52a)
dx

• Type (iii) (Robin) y

dy
 ; @ x  xo
dx (1.53a)

• The boundary values above are not homogenous.

 Boundary conditions

• For most problems occurring in chemical engineering, boundary conditions

or initial conditions are (or can be made to be) of the homogenous type.

• The three classical homogenous types of boundary conditions at a point xo

are:
• Type (i) (Dirichlet boundary) y  x   0; @ x  xo (1.51)

dy
• Type (ii) (Neumann)  0; @ x  xo (1.52)
dx

• Type (iii) (Robin) dy

y  0; @ x  xo (1.53)
dx

• Boundary values for derived models are frequently non homogenous, but
they can be made so with a change of variable.
 Boundary conditions
For instance in example 1 and 2, the wall temperature was constant

T  r , z   Tw ; @ rR for all z (1.54)

As it stands it is not homogenous, however by defining the variable
  T  r , z   Tw (1.55)
With this change of variable, the boundary condition is homogenous type (i)
  r, z   0 @ r  R , for all z (1.56)
With this change of variable the differential equation 1.50 becomes
 2   2 1   2 
k 2 k 2    2v  C 
p 1   r R  
 z (1.57)
z  r r r 
o
 Boundary conditions
It often occurs that instead of having a constant wall temperature, there is
convective heat transfer at the wall (Newton’s law of cooling)

dT
qr  k  h T  Ta  ; @ r  R ; for all z (1.58)
dr
Where h is the heat transfer coefficient and Ta is the ambient temperature
Care must be taken with the signs. This has been identified as a cooling
problem. If that is the case, heat moves as in the directions identified in the
right hand side figure at the top.
In that case the temperature is higher in the centre of the tube and reduces
as r increases. This is because heat moves opposite to the gradient of
temperature.
dT
qr  0,
dr
0 then qr  h T  Ta  must be positive as Ta< T
 Boundary conditions
The boundary condition 1.58 is not homogenous
dT (1.58)
qr  k  h T  Ta  ; @ r  R ; for all z
dr

But with a change of variable

  T  Ta

Then the boundary condition is transformed as

dT
qr  k  h T  Ta  ; @ rR (1.59)
dr
d
qr  k  h ; @ rR (1.60)
dr

Finally, rearranging 1.60

h d (1.61)
  0; @ rR
k dr
Which is a type (iii) homogenous boundary condition
 Boundary conditions
Finally if the wall tube is well insulated

dT
qr  k  0; @ r  R; for all z (1.62)
dr

This conditions is a homogenous type (ii)

 Summary of the model building
process
The main three steps of how models on examples 1 and 2 were
set up are:

 Step 1: Have clarity on you mind of what the problem is

about?

 Step 2: Bring physical and chemical information assumptions

etc.

 Step 3: Setting down of finite or differential volume elements

and to write conservation laws.
 More detailed step by step procedure
1. Draw a sketch of the system to be modelled and label/define the various geometric,
physical and chemical quantities.
2. Carefully select the important dependent (response) variables.
3. Select the possible independent variables (e.g., z, t) changes in which must necessarily
affect the dependent variables.
4. List the parameters (physical constants, physical size, and shape) that are expected to
be important; also note the possibility of non constant parameters[e.g., viscosity
changing with temperature].
5. Draw a sketch of the expected behaviour of the dependent variable(s),such as the
"expected" temperature profile to help setting the signs correctly.
6. Establish a "control volume" for a differential or finite element of the system to be
modelled; sketch the element and indicate all inflow-outflow paths.
7. Write the "conservation law" for the volume element: Express flux and reaction rate
terms using general symbols, which are taken as positive quantities, so that signs are
introduced only as terms are inserted according to the rules of the conservation law.
8. After rearrangement to the proper differential format, re arrange the equation to take
the limits to produce a differential equation.
9. Introduce specific forms of flux (e.g., Jr = —DdC/dx) and rate (RA = kCA); note, the
opposite of generation is depletion, so when a species is depleted, then it must be
entered with the appropriate sign in the conservation law equation.
10. Write out all possibilities for boundary values of the dependent variables; the choice
among these will be made in conjunction with the solution method selected for the
defining (differential) equation.
11. Search out solution methods, and consider possible approximations for: (i) the
defining equation, (ii) the boundary conditions, and (iii) an acceptable final solution.
 Model hierarchy

 As explained in step 2 of “how to formulate a model “ (slide 3), you must

“clarify the real scope and deep of the model”.
 So it is possible to have a hierarchy of models from a very simple one
(almost like a black box”, to the highest level of complexity when you can
account for all possible transfer phenomena, thermodynamics, variation
of properties with temperature and composition, as well as other physical
associated phenomena such as electromagnetic fields, structural analysis,
sound, radiation etc.
 As the level of sophistication increases, the mathematical complexity
increases.
 The higher the complexity of the model the more detailed information
about the physical system can be obtained.
 Models of lower complexity can be solved mathematically, but models of
very high complexity needs to be solved via numerical methods and/or
computational fluid dynamics.
 However, many times, high level of complexity models are not needed. So
the engineer must decide the scope and deep of the model, depending of
what she/he wants to achieve from the model
 Example 1 vs. example 2

Example 1
1. The velocity profile has a Plug shape Example 2
(velocity is constant and does not change in 1. The velocity is a function of r (eq. 1.22)
the z direction) following a parabolic shape.
2. The fluid is well mixed (highly turbulent), so 2. The fluid is NOT well mixed in the radial
the temperature is uniform in the radial direction, so radial heat conduction must be
direction. accounted.
3. The system is under steady state 3. The system is under steady state
4. physical properties are constant 4. physical properties are constant (temperature
(temperature independent) independent)
5. The wall temperature Tw is constant and 5. The wall temperature Tw is constant and
uniform ( does not change in z or r uniform ( does not change in z or r direction)
direction)
6. The inlet temperature To is constant and
6. The inlet temperature To is constant and uniform (does not change in r direction)
uniform (does not change in r direction)
7. Velocity is slow (Laminar) therefore advection is
7. Thermal conduction of heat along the axis is slow. In this case the convective heat is due to
small relative to conduction. Hence, the advection + conduction; conduction may be
convective movement of heat on the fluid is important and needs to be accounted.
due only to advection.
 Higher complexity beyond example 2
Is it possible to extend the complexity of the model example 2?
Yes 8. Coupling other physical phenomena
such as electric fields, sound, radiation,
1. Modelling the heat conduction structural mechanics etc.
Through the walls of the tube 7.Instead of using Newton’s law
Of cooling, modelling the heat
2. Modelling the details of how and fluid transfer outside the
the fluids develops before tube
entering the tube
6. Modelling the boundary
Layer

3. Modelling the dependency of 5. Modelling the turbulence

physical properties with Of the fluid flow
temperature 4. Modelling the system as
Unsteady state

Low complexity models such as example 1 can be solved mathematically, but higher
Complexity models needs to be solved numerically, either programing numerical methods, or
Using computational fluid dynamics packages such as ANSYS-FLUENT or multi-physics packages
Such as COMSOL
 Example 3 – Hierarchy levels 1-4
Heat removal from a bath by immersing a steel rod into the bath and allowing
to heat dissipate from the solvent to rod, and then from the rod to the
environment.
 Example 3 – level 1
Level 1
In this level, let us assume that:
1. The rod temperature is uniform, that is, from the bath to the
atmosphere.
2. Ignore heat transfer at the two flat ends of the rod.
3. Overall heat transfer coefficients are known and constant.
4. No solvent evaporates from the solvent air interface.

Let T0 and T1 be the atmosphere and solvent

temperatures.
Because there is no accumulation of heat by the road
(steady state assumption) the heat entering via
convection from the solvent to the rod on section L1,
must be equal to the heat leaving the rod to the
atmosphere on section L2.
Q  hL  2 RL1 T1  T   hG  2 RL2 T  To 
(1.63)
 Example 3 – level 1

Q  hL  2 RL1 T1  T   hG  2 RL2 T  To  (1.63)

T can be solved from equation 1.63 algebraically

 hL 2 RL1  hG 2 RL2  T   hG 2 RL2To  hL 2 RL1T1 
 hL 2 RL1  hG 2 RL2  T   hG 2 RL2To  hL 2 RL1T1 
 hL L1 
 o
T  T1 
 hG 2 RL2To  hL 2 RL1T1   hG L2To  hL L1T1  T  
h L
T T
G 2

 hL 2 RL1  hG 2 RL2   hL L1  hG L2   hL L1 hG L2 
  
 G 2
h L hG 2 
L

T
To   T1  (1.64)
1   
hL L1

hG L2 (1.65)
 Example 3 – level 1

T
To   T1  (1.64)
1   
hL L1
 (1.65)
hG L2

Equation 1.64 gives an easy estimate of the rod temperature and how it
varies with exposure length.
For example, if a is much greater than unity (i.e. long L1 section and high
liquid heat transfer coefficient compared to gas coefficient), the rod
temperature is then very near T1.
Taking the rod temperature to be represented by Eq. 1.64, the rate of heat
transfer is readily calculated from Eq. 1.63 by replacing T (Eq. 1.64)
Q  hL  2 RL1 T1  T   T   T1  
Q  hL  2 RL1   T1  o 
 1    
hL 2 RL1
Q T1 1     To  T1   Q
hL 2 RL1
T1  To 
1    1   
(1.66)
 Example 3 – level 1
hL 2 RL1 hL L1
Q T1  To   2 R T1  To  (1.67)
1    

hL L1 
 1 
 h G 2 
L
1
Q 2 R T1  To 
 1 1  (1.68)
  
 hL L1 hG L2 
When α = hLL1/hGL2 is very large (hLL1 is large and 1/(hLL1) approach zero), the
rate of heat transfer becomes simply
Q  2 RhG L2 T1  To  (1.69)
Thus, the heat transfer is controlled by the segment of the rod exposed to the
atmosphere.
It is interesting to note that when the heat transfer coefficient contacting the
solvent is very high (i.e., α » 1), it does not really matter how much of the rod
is immersed in the solvent.
Thus for a given temperature difference and a constant rod diameter, the rate
of heat transfer can be enhanced by either increasing the exposure length L2
or by increasing the gas heat transfer rate by stirring the solvent.
 Example 3 – level 1
However, these conclusions are tied to the assumption of constant rod
temperature, which becomes tenuous as atmospheric exposure is increased.
To account for effects of temperature gradients in the rod, we must move to
the next level in the model hierarchy, which is to say that a differential
volume must be considered.
 Example 3 – level 2
Level 1
1. The rod temperature is uniform, that is, from the bath to the
atmosphere.
2. Ignore heat transfer at the two flat ends of the rod.
3. Overall heat transfer coefficients are known and constant.
4. No solvent evaporates from the solvent air interface.

Level 2
Let us relax part of the assumption (1) of the level 1 by assuming only that the
rod temperament below the solvent liquid surface is uniform at a value T1.
This is a reasonable proposition, since the liquid has a much higher thermal
conductivity than air.
The remaining three assumptions of the level 1 model are retained.
 Example 3 – level 2
Next, choose an upward pointing coordinate x with the origin at the solvent-
air surface.
The figure below shows the coordinate system and the elementary control
volume.

Applying a heat balance around a thin shell segment with thickness ∆x gives

 R 2 qx   R 2 qx x  2 RxhG T  To   0 (1.70)

 Example 3 – level 2

 R 2 qx   R 2 qx x  2 RxhG T  To   0 (1.70)

by writing this, temperature gradients are likely to exist in the part of the
rod exposed to air.
Dividing Eq. 1.70 by πR2 ∆x and taking the limit as ∆x -> 0

 R 2 qx   R 2 qx x  2 RxhG T  To   0
qx  qx x 2hG
 T  To   0
x R
qx x  qx 2hG
 T  To   0
x R
yields the following first order differential equation for the heat flux, q:

dqx 2hG
 T  To   0 (1.71)
dx R
 Example 3 – level 2
dqx 2hG
 T  To   0 (1.71)
dx R

Assuming the rod is homogeneous, i.e., the thermal conductivity is uniform,

the heat flux along the axis is related to the temperature according to
Fourier's law of heat conduction
dT (1.72)
qx   k
dx
Substitution of eq. 1.72 into 1.71 yields
d 2T 2hG (1.73)
k 2  T  To 
dx R
Equation 1.73 is a second order ordinary differential equation, and to solve
this, two conditions must be imposed.
T  T1 ; @ x0 (1.74)
 Example 3 – level 2
at x = L2 , at the end of the rod, one can assume Newton's law of cooling
prevails, but since the rod length is usually longer than the diameter, most of
the heat loss occurs at the rod's lateral surface, and the flux from the top
surface is small, so write approximately:
dT
 0; @ x  L2 (1.75)
dx
Equation 1.73 is subjected to the two boundary conditions to yield the
solution
cosh  m  L2  x  
T  To  T1  To  (1.76)
cosh  mL2 

2hG
m (1.77)
Rk

The solution method will be explained in subsequent lectures.

 cosh  m  L2  x  
T  To  T1  To  (1.76)
cosh  mL2 

Example 3 – level 2 m
2hG
Rk

Once we know the temperature distribution of the rod above the solvent-air
interface, then the rate of heat loss can be calculated either of two ways:
• In the first, we know that the heat flow through area πR2 at x = 0 must be
equal to the heat released into the atmosphere, that is,
dT
Q   R 2 k (1.78)
dx x 0

Applying equation 1.76 to 1.78 gives

Q  2 RhG L2 T1  To  (1.79)

where
tanh  mL2 
 (1.80)
mL2
This dimensionless group (called effectiveness factor) represents the ratio of
actual heat loss to the (maximum) loss rate when gradients are absent.
 Example 3 – level 2
The following figure below shows the log-log plot of η versus the
dimensionless group mL2. We note that the effectiveness factor approaches
unity when mL2 is much less than unity and it behaves like 1/ mL2 as mL2 is
very large.

In the limit for small mL2, we can write

tanh  mL2 
 1 (1.81)
mL2
 2hG
Example 3 – level 2 m
Rk

tanh  mL2 
 1 (1.81)
mL2

which is the most effective heat transfer condition. This is physically achieved
when
I. The rod thermal conductivity is large.

For such a case, we can write the elementary result

Q  2 RhG L2 T1  To  (1.82)

which is identical to the model level 1 (Eq. 1.69). Thus, we have learned that
the first model is valid only when mL2 << 1.
 cosh  m  L2  x  
T  To  T1  To 
Example 3 – level 2 cosh  mL2 
(1.76)

Another way of calculating the heat transfer rate is carrying the integration of
local heat transfer rate along the rod
x  L2 x  L2

Q  dq   hG T  To  2 Rdx  (1.83)
x 0 x 0

where T is given in Eq. 1.76 and the differential transfer area is 2πRdx.
Substituting T of Eq. 1.76 into Eq. 1.83 yields the same solution for Q as given
in Eq. 1.79.
Levels 1 and 2 solutions have one assumption in common: The rod
temperature below the solvent surface was taken to be uniform.
The validity of this modelling assumption will not be known until we move up
one more level in the model hierarchy.
 Example 3 – level 3
Level 1
1. The rod temperature is uniform, that is, from the bath to the
atmosphere.
2. Ignore heat transfer at the two flat ends of the rod.
3. Overall heat transfer coefficients are known and constant.
4. No solvent evaporates from the solvent air interface.

Level 2
relax part of the assumption (1) of the level 1 by assuming only that the rod
temperament below the solvent liquid surface is uniform at a value T1.

Level 3
Again we relax the assumption (1) allowing for temperature gradients in the
rod for segments above and below the solvent-air interface.
 Example 3 – level 3

Let the temperature below the solvent-air interface be TI and that above
the interface be TII. Carrying out the one-dimensional heat balances for the
two segments of the rod, we obtain
d 2T I 2hL I (1.84)
dx 2

kR
T  T1 
d 2T II 2hG II (1.85)
dx 2

kR
 T  To 

We shall still maintain the condition of zero flux

at the flat ends of the rod. This means of course
that
dT I
 0; @ x   L1 (1.86)
dx
dT II
0 ; @ x  L2 (1.87)
dx
 Example 3 – level 3
Two other conditions arise from the continuity of temperature and flux at the
x = 0 position

T I  T II ; @ x0 (1.88)

dT I dT II
 ; @ x0 (1.89)
dx dx
Equations 1.84 and 1.85, subject to conditions 1.86 and 1.87,
can be solved analytically via methods that will be explained in
subsequent lectures. The solution is
T I  T1  A cosh  n  x  L1   (1.90)

T II  To  B cosh  m  L1  x   (1.91)
Where m is as defined on level 2 (Eq.1.77)
2hG
m (1.92)
Rk

and n is:
2hL
n (1.93)
Rk
 Example 3 – level 3
The constants of integration, A and B, can be found by substituting Eqs. 1.90
and 1.91 into the continuity equations 1.88 and 1.89
B
T1  To 
 m sinh  mL2   (1.94)
 cosh  2
mL  cosh  1 
nL
 n sinh  nL 1  

A
T1  To 
 n sinh  nL1  
 cosh  1
nL  cosh  2 
mL
 m sinh  mL 2  
(1.95)

The rate of heat transfer can be obtained by using either of the two ways
mentioned earlier, that is, using flux at x = O, or by integrating around the lateral
surface. In either case we obtain
dT I dT II
Q  Ax q x 0    R  k
2
   R  k
2
(1.96)
dx x 0
dx x 0

Q  2 RhG L2
T1  To 
 m tanh  mL2   (1.97)
1  
 n tanh  nL 
1 
 2hG
m (1.77)
Example 3 – level 3 Rk

Q  2 RhG L2
T1  To 
 m tanh  mL2   (1.97)
1  
 n tanh  nL 
1 

where the effectiveness factor η was in Eq. 1.80 on model level 2

tanh  mL2 
 (1.80)
mL2

What is the difference between the solution obtained by the level 2 and level 3?
the rate of heat transfer calculated at level 3 is less than that calculated by the level two
model where the rod temperature was taken to be uniform at T1 below the liquid surface.
Level 2 Level 3
Q  2 RhG L2 T1  To  (1.79) Q  2 RhG L2
T1  To 
(1.97)
 m tanh  mL2  
1  
 n tanh  nL 
1 
This implies from Eq. 1.77 that the heat resistance in the submerged region is negligible
compared to that above the surface only when
m tanh  mL2 
1 (1.98)
n tanh  nL1 
 2hG
m (1.77)
Example 3 – level 3 Rk

m tanh  mL2 
1 (1.98)
n tanh  nL1 

When the criterion (1.98) is satisfied, the rate of heat transfer given by
Model 2 is valid. This is controlled mainly by the ratio m/n = (hG/hL)1/2,
which is always less than unity. In other words, when hL is much higher
than hG, which is expected.
What we have seen in this exercise is that higher complexity levels of modelling yield more
information about the system and hence provide needed criteria to validate the model
one level lower.
In our example, the level 3 model provides the criterion (1.98) to indicate when the
resistance to heat flow underneath the solvent bath can be ignored compared to that
above the surface.
Level 2 model provides the criterion (1.81) to indicate when there is negligible conduction-
resistance in the steel rod.

tanh  mL2 
 1 (1.81)
mL2
The next level of modelling is by now obvious: At what point and under what conditions
do radial gradients become significant? This moves the modelling exercise into the domain
of partial differential equations
 Example 3 – level 4
Level 1
1. The rod temperature is uniform, that is, from the bath to the
atmosphere.
2. Ignore heat transfer at the two flat ends of the rod.
3. Overall heat transfer coefficients are known and constant.
4. No solvent evaporates from the solvent air interface.
Level 2
relax part of the assumption (1) of the level 1 by assuming only that the rod
temperament below the solvent liquid surface is uniform at a value T1.
Level 3
Again we relax the assumption (1) allowing for temperature gradients in the
rod for segments above and below the solvent-air interface.
Level 4
We relax the assumption (1) allowing for temperature gradients in the radial
direction of the rod.
 Example 3 – level 4
we include radial heat conduction. This is important if the rod diameter is
large relative to length.
Let us assume in this model that there is no resistance to heat flow
underneath the solvent interface, so as before, take temperature T = T1 when
x < 0.
This then leaves only the portion above the solvent surface to study.
 Example 3 – level 4
Setting up the annular shell shown in the figure below and carrying a heat
balance in the radial and axial directions, we obtain the following heat
conduction equation

 2 r xqr  r   2 r xqr  r    2 r rqx  x   2 r rqx  x 

r x
0 (1.99)

Dividing this equation by 2πΔrΔx and taking limits, we obtain

  rqr    qx  (1.100)
 r 0
r x
Next, insert the two forms of Fourier's laws
T
qr  k (1.101)
r
T
qx   k
x (1.102)

To finally get
 1   T   T
2
k r  2   0 (1.103)
 r r  r  x 
 Example 3 – level 4
 1   T   T
2
k r  2   0 (1.103)
 r r  r  x 

What are the assumptions of the model regarding thermal conductivity k?

• is constant (does not change with temperature),
• Isotropic (uniform in both x and r directions)
Equation 1.103 is an elliptic partial differential equation. The physical
boundary conditions to give a suitable solution are the following
T
 0; @ r 0 (1.104)
r
T
qr  k  hG T  To  ; @ rR (1.105)
rR
r
T  T1 ; @ x0 (1.106)
T
 0; @ x  L2 (1.107)
x
 Example 3 – level 4
T
 0; @ r 0 (1.104)
r
Equation 1.104 implies symmetry at the centre of the rod.
T
qr  k h  T  T  ; @ r  R (1.105)
rR
r
G o

Equation 1.105 implies that the heat by conduction, at the curved outer surface of
the rod, must be equal to the convective heat. This is because the system is steady
state and there is not accumulation of heat.
T  T1 ; @ x0 (1.106)
Equation 1.106 says that the temperature at x = 0 is T1
T
 0; @ x  L2 (1.107)
x
Equation 1.107 states that there is no heat flow across the flat end
of the rod. This is tantamount to saying that either
• the flat end is insulated, or
• the flat end area is so small compared to the curved surface of
the rod that heat loss there is negligible.
 Example 3 – level 4
When dealing with simple equations (as in the previous three models), the
dimensional equations are solved without recourse to the process of
nondimensionalisation.
Now, we must deal with partial differential equations, and to simplify the notation
during the analysis and also to deduce the proper dimensionless parameters, it is
better to reduce the equations to nondimensional form.
To achieve this, we introduce the following nondimensional variables and parameters:
T  To
u Dimensionless temperature – The dependant variable
T1  To (1.108)
x
  Dimensionless x position – First independent variable
L2 (1.109)
r
 Dimensionless radius– Second independent variable (1.110)
R
R
 First dimensionless parameter (1.111)
L2
hG R
Bi  Biot number – second dimensionless parameter (1.112)
k
By doing this, the system has been simplified to 1 dependent variable, 2 independent
variables and only 2 parameters
 Example 3 – level 4
The dimensionless relations now become:
1   u   2u
  2  0 (1.113)
     
Subjected to the following boundary conditions:
u
 0; @  0 (1.114)

u (1.115)
  Bi  u; @  1

u  1; @  0 (1.116)
u
 0; @  1 (1.117)

The solution of u is:
 
cosh  n 1    
T  To 
1, K n  
u  K n   (1.118)
T1  To n 1 Kn , Kn  
cosh  n 

 Example 3 – level 4
 
cosh  n 1    
T  To 
1, K n  
u  K n   (1.118)
T1  To n 1 Kn , Kn  
cosh  n 

Where the functions are defined as:
K n    J o   n  (1.119)

and the many characteristic values (eigenvalues) are obtained by trial-and-

error from
 n J1  Bi  J o   n  (1.120)
The other functional groups are defined as
J  
1, K n  1 n (1.121)
n
J12   n     n  
2

Kn , Kn  1     (1.122)
2   Bi  
where J0 (βn) and J1 (βn) are tabulated relations called Bessel functions, which
will be discussed later
 Example 3 – level 4
The rate of heat transfer can be calculated using the heat flux entering at
position x = 0, but in order to do that, we must account for radial variation of
temperature so that the elemental area is 2πrdr; thus integrating over the
whole base gives
r R r R
 T  (1.123)
Q 
r 0
qx x 0
 dAx  r 0 k x  x0   2 rdr 
Putting this equation in dimensionless form:
rR  1
2 R 2 k  u 
Q   qx x 0  dAx  T1  To      d (1.124)
r 0
L2   0      0
Inserting dimensionless temperature from Eq. 1.118, we obtain the following
result for heat transfer rate
 n 1, K n
2
2 R 2 k 
  n  
Q T1  To   tanh     (1.125)
L2  n 1 Kn , Kn   
This illustrates how complexity grows quickly as simplifications are relaxed.
For small Bi << 1, it is not difficult to show from the transcendental equation
1.120 that the smallest eigenvalue is
(1.126)
1   2Bi 
1/ 2
 Example 3 – level 4
1   2Bi 
1/ 2
(1.126)

Substituting this into Eq. 1.125

 1, K n
2
2 R 2 k 
  (1.125)
Q T1  To   n tanh   n    2 RhG L2 T1  To 
L2  n 1 Kn , Kn   

we will obtain Eq. 1.79, which is the solution for level 2.

Thus, the level 4 model shows that the radial heat conduction inside the rod
is unimportant when
Bi 1 (1.126)
In summary, we have illustrated how proper model hierarchy sets limits on
the lower levels. In particular, one can derive criteria (like Eq. 1.126) to show
when the simpler models are valid.
 Heat transfer rate comparison levels 1-4
Level 1

. Q  hL  2 RL1 T1  T   hG  2 RL2 T  To  (1.63)

Level 2 Level 3
Q  2 RhG L2 T1  To  (1.79) Q  2 RhG L2
T1  To 
(1.97)
 m tanh  mL2  
1  
 n tanh  nL 
1 

This implies from Eq. 1.77 that the heat resistance in the submerged region is negligible
compared to that above the surface only when
m tanh  mL2 
1 (1.98)
n tanh  nL1 

 n 1, K n
2
2 R 2 k 
 
Level 4 Q T1  To   tanh   n    2 RhG L2 T1  To  (1.125)
L2  n 1 Kn , Kn   
level 4 also says that that the radial heat conduction inside the rod is unimportant when
Bi<<1.
 Example 3 – levels 1- 4
Level 1
1. The rod temperature is uniform, that is, from the bath to the atmosphere.
2. Ignore heat transfer at the two flat ends of the rod.
3. Overall heat transfer coefficients are known and constant.
4. No solvent evaporates from the solvent air interface.
Level 2
relax part of the assumption (1) of the level 1 by assuming only that the rod
temperament below the solvent liquid surface is uniform at a value T1.
Level 3
Again we relax the assumption (1) allowing for temperature gradients in the rod for
segments above and below the solvent-air interface.
Level 4
We relax the assumption (1) allowing for temperature gradients in the radial direction
of the rod.

So far we had only relaxed assumption one.

Can we relax assumptions 2,3 and 4 to increase the level of complexity?
 When is a model of a process good enough?
 This is not a trivial question, and it can only be answered fully
when the detailed economics of design and practicality are
taken into account.
 Here, we have simply illustrated the hierarchy of one simple
process, and how to find the limits of validity of each more
complicated model in the hierarchy.
 In the final analysis, the user must decide when tractability is
more important than precision.
 Scope of this first part of CEIC3000
 to be able to formulate mathematical models
 to understand and been able to justify the assumptions in a model as well
as the assumptions and conditions under which the model is valid.
 To apply mathematical solutions to solve models of lower level of
complexity.
 To apply numerical methods to solve models that cannot be solved
analytically.
 References
• Chapter 1 - Rice, R. G. and Do, D.D., Applied
Mathematics and Modelling for Chemical
Engineers, John Wiley, 1995.