You are on page 1of 21

Modeling and Simulation in Building Materials Technology

2. The Thermal problem of symmetrical one dimensional heat transfer


2.1 Problem definition

We still investigate the thermal shock of a plate –s/2  x  s/2. In case of constant thermal
conductivity l the following partial differential equation governs the thermal problem:

T  2T
=a 2
t x

T (x ,0) = T0 , −
s s
initial condition: x
2 2
boundary condition:

T   s 
left boundary −l =   T − T  − ,t  
x x =−
s
  2 
2

𝜕𝑇 𝑠
right boundary 𝜆 ቤ
𝜕𝑥 𝑥=𝑠
= 𝛼 𝑇∞ − 𝑇 , 𝑡
2
2

H. Harmuth 1 53
Modeling and Simulation in Building Materials Technology

It is useful to apply transformations to dimensionless quantities. The following


transformations are applied here:

i) Dimensionless temperature:

T − T
=
T0 − T

Obviously Tt = t  (T0 − T ) and T =   (T − T ) , therefore the governing differential equation


xx xx 0 

transforms to the intermediate result: t = a  xx

x 2x
ii) Dimensionless length coordinate:  = =
s2 s
s s
This transforms the interval − x to −1    1
2 2
a t a t
iii) Dimensionless time: = =4 2
(s 2) 2
s

This ratio is called Fourier number.

H. Harmuth 2 53
Modeling and Simulation in Building Materials Technology

Now the partial differential equation is transformed:

2
 x =   x = 
s
4 t = a  xx →  = 
 xx = 
s2
4a
t =   t = 
s2

We now transform the boundary and initial conditions.

Left boundary condition:


T   s 
−l =   T − T  − ,t   : (T0 − T )
x x=−
s
  2 
2

2
− l  x = −  x = 
s
2
l  =  
s
 s
 = 
2l
 s
 = Bi   , Bi =
2l

H. Harmuth 3 53
Modeling and Simulation in Building Materials Technology

Similarily the right boundary condition reads:

 = − Bi  

Initial condition:

T − T
T = T0 → = =1
T0 − T T =T0

The problem definition in the transformed state is summarized here:

 =  −1    1 initial condition: ( ,0) = 1


boundary condition:
 s
left boundary  = Bi   , Bi =
2l
right boundary  = − Bi  

The transformation rules are:

T − T 2x  =4
a t
= , = ,
T0 − T s s2

H. Harmuth 4 53
Modeling and Simulation in Building Materials Technology

Exercise: The Fourier number and the Biot number usually are defined as:
a t X
Fo = ; Bi =
X2 l
Here X is a characteristical geometrical dimension. Above X=s/2 has been chosen. Considering an
infinite plate of thickness s, in which case would you recommend the choice X=s, in order to apply
the same dimesionless solution?

Which advantages has the transformation to dimesionless variables?


i) Reduction of the number of independent variables

Number
original problem x s t T T0 T∞ a l  9
reduced problem x s t T-T∞ T0-T∞ a l  8
units [m] [K] [s] [K] [K] [K] [m²/s] [W/mK] [W/m²K]
units necessary 4 (m,s,K,J)
transformed    Bi 4
problem

The reduction of Variables 8-4=4 satisfies the P-Theorem according to Buckingham:


A relation between n dimensionful quantities in r elementary units may be reduced to n-r
dimensionless quantities.

H. Harmuth 5 53
Modeling and Simulation in Building Materials Technology

ii) Increase of generality

A solution of the transformed problem (,) can be applied for


-all values for T0, T∞ resulting in calculated  values and
-all values of the other parameters for which , and Bi have the same values.
The solution may easier be depicted as a family of curves as  only depends on 3 parameters , and Bi.

2.2 A numerical solution of the dimensionless problem

Different numerical methods may be applied, the most prominent are the finite element and the finite
differences method. The latter one will be applied in this section.
Finite differences are applied to approximate the derivatives. The forward difference quotient may be
applied to approximate the first derivative with respect to :

 ( , +  ) − ( , )

 

Evidence about the error of this approximation follows from the taylor expansion of (,+):

 1  2
 ( , +  ) =  ( , ) +  +  2 + ....
  , 2   ,
2

  
0 ( 2 )

H. Harmuth 6 53
Modeling and Simulation in Building Materials Technology

Recalculating the difference approximation from the taylor expansion yields:

 ( , +  ) − ( , )
= + O( )
 

i.e. the error is of the order of  and is expected to approach zero for →0.

We now apply a discretisation of the / space with equidistant points in each direction:

 i =  0 + i  ;  j =  0 + j 
( i , j ) =  ij

With this notation we can write:

 ij +1 − ij

 

For the second derivative we apply a backward difference from two forward differences:

 
1

(
  −   − 
) = 1   i +1 j


−  ij

 ij −  i −1 j  i +1 j − 2 ij + i −1 j

 =
  2

H. Harmuth 7 53
Modeling and Simulation in Building Materials Technology

If we substitute both approximations into the dimensionless heat transfer equation we receive:

 = 
 ij +1 −  ij  i +1 j − 2 ij +  i −1 j
=
  2

If j represents the initial condition or any already calculated temperature distribution, the above
difference equation contains only one unknown variable which is ij+1 and can be explicitely
calculated from:


ij +1 = (i+1 j − 2ij + i−1 j ) + ij = 2 i+1 j + 1 − 2 2 ij + 2 i−1 j
 2
    

In a similar way we generate difference approximations for the boundary conditions:


Left Boundary

 = Bi   ,
 2 , j − 1, j
= Bi  1 j

From this we can calculate the temperature at the boundary:
2, j
1, j =
1 + Bi  

H. Harmuth 8 53
Modeling and Simulation in Building Materials Technology

For the right boundary we apply a backward difference and receive a similar result:
 = − Bi   ,
 n , j −  n −1, j
= − Bi   nj

 n −1 j
 nj =
(1 + Bi   )

Now we consider the order of the discretization error for the boundary condition. For  at the
location =0 and =s/2 a first order approximation is used, while for the second derivative and
constant  a second order approximation is applied. Therefore the boundary condition as quoted
above somewhat restricts the accuracy of the total procedure and a second order approximation is
more favourable. Application of a second order approximation for the derivative  results in:

4 2, j − 3, j 4 n −1, j −  n − 2, j
1, j = , n, j = .
3 + 2  Bi 3 + 2  Bi

For the derivation of these results, see Exercise 2.2.

H. Harmuth 9 53
Modeling and Simulation in Building Materials Technology

It is reasonable to apply symmetry and to solve only one symmetrical halve of the problem:

From the symmetry follows:

(−  , ) = ( , )

This helps do determine the difference


approximation for 1j+1:

    
(0 , +  ) = ( , ) + 1 − 2 2 (0, ) + (−  , )
 2
    2
   
=2 ( , ) + 1 − 2 2 (0, )
 2
  
   
1 j +1 = 2  2 j + 1 − 2 2 1 j
 2
  

H. Harmuth 10 53
Modeling and Simulation in Building Materials Technology

A restriction for the choice of the local and time increments:


The difference approximation

    
ij +1 = i +1 j + 1 − 2 2 ij + i −1 j
 2
    2

contains the coefficient

  
1 − 2 2 
  

which has to be larger than zero:


1− 2 0
 2
 1
or 
 2
2

This is a so called stability condition. In case it is not satisfied roundoff errors will propagate in an
unfavourable manner and anticipate a valid result.

H. Harmuth 11 53
Modeling and Simulation in Building Materials Technology

It may be useful to change the interval length of the local discretisation. A smaller interval
length is applied for a steeper slope of the temperature.

i-1 i i+1
i-1 i

 i
=
1
 i −1
(  
i
−   i −1
) = 1  ( i +1 ) − ( i ) ( i ) − ( i −1 ) 

i −1    i

  i −1


or in simplified notation:

 i =
1
 i −1
(  i −  i −1
) = 1   i +1 − i  i − i −1 

 

 

i −1  i i −1 

We now obtain the difference equation:

   i +1 j −  ij  ij −  i −1 j 
 ij +1 =  ij +  − 
 i −1   i  i −1 
  
 ij +1 =  ij +  −   1
+
1  ij +   i +1 j
 2
i −1 j
    2   i −1 i
i −1  i −1 i i −1 

And for the boundaries using a first order approximation we receive:

H. Harmuth 12 53
Modeling and Simulation in Building Materials Technology

    
left: 1 j +1 = 1 j + 2 ( −  ) = 2  + 

1 − 2 
2  1j
 12  12 
2j 1j 2j
 1 

right:  = − Bi   ,
 n , j −  n −1, j
= − Bi   nj
 n −1
 n −1 j
 nj =
(1 + Bi   n−1 )

Often – imagine the case of rapid cooling – the temperature curve will be steeper near the surface, and a
geometrical partition of the interval 0    1 may be applied.
Let
(1 = 0, 2 ) be the first interval with length 1 and

(n−1 ,n = 1) be the last (= n-1 th) interval with length n-1

Then a geometrical partition yields the following values for the i :
 n − 2 =  n −1  k
 n −3 =  n − 2  k =  n −1  k 2
.......
 n −l =  n −1  k l −1

H. Harmuth 13 53
Modeling and Simulation in Building Materials Technology

The sum of all i equals unity:


n −1 n −1 n −1 n−2
1 =  i =   n −l =  n −1  k l −1
=  n −1  k l
i =1 l =1 l =1 l =0

We apply the sum of a finite geometrical series:


m
k m +1 − 1

i =0
k =
i

k −1

This yields:

k n −1 − 1
1 =  n −1
k −1
 n −1k − k + 1 −  n −1 = 0
n −1

A possible solution may e.g. be obtained by the Newton-Raphson method:

f (k ) =  n −1k n −1 − k + 1 −  n −1
f (ki )  k n −1 − k + 1 −  n −1
ki +1 = ki − = ki − n −1
f ´ (ki )  n −1 (n − 1)k n − 2 − 1

For e.g. 30 intervals (n=31) and n-1=0,001 follows k= 1,191685429.

H. Harmuth 14 53
Modeling and Simulation in Building Materials Technology

3. The numerical solution for the thermomechanical problem of a plate

The following diagram shows m-a where m is the mean and a the outer surface temperature in
dependence of  and with Bi as a second parameter calculated by the explicite difference procedure
quoted above:

H. Harmuth 15 53
Modeling and Simulation in Building Materials Technology

As a rule maximum temperature differences and further maximum stresses are of interest. The
following diagram shows the maximum values of m-a and m-i (i is the center temperature) in
dependence of Bi calculated by the explicite difference procedure quoted above:

H. Harmuth 16 53
Modeling and Simulation in Building Materials Technology

Now we calculate the maximum occuring stresses from this solution. For this we recall:

E T − T
= (T − T ) and the transformation rule: =
(1 − ) m T0 − T

From this follows:

Tm − T = (m − ) (T0 − T ) E


and = (T − T ) ( − )
(1 − ) 0  m

 max,Bi =
Here max, Bi= has the meaning of the maximum thermal stress occuring in case of a thermal
shock with T=T0-T and Bi=; it will occur immediately after exposure to thermal shock at the
surface and yield tension in case of cooling and compression in case of heating.

 max,Bi =  max,Bi =  ( m − )

We can now define a dimensionless stress variable :

 max,Bi
= = m − 
 max,Bi =

H. Harmuth 17 53
Modeling and Simulation in Building Materials Technology

Note: The dimensionless stress equals the dimensionless temperature difference m-.

Now we investigate the dimensionless time  at which the maximum temperature difference occurs:

H. Harmuth 18 53
Modeling and Simulation in Building Materials Technology

The dependence of (m-a) and (m-i) from Bi shows the following properties:
− (m-a) is smaller than (m-i) because for the latter one heat transfer takes more
time to proceed in the center of the plate
− (m-a) and (m-i) depend linearly on log Bi for Bi app. <1
- For Bi→∞ both temperature differences approach a constant value which is 0 for (m-
a) and app. 0,117 for (m-i)

H. Harmuth 19 53
Modeling and Simulation in Building Materials Technology

Exercise 2.1
Determine the order n of the Error O(hn) of the difference approximation for the second derivative
shown in the section before.
Exercise 2.2
Derive the discrete boundary conditions with the error O(2) shown in the above section. Hint:
Assume constant . Write a Taylor expansion for 2 and 3 around 1. From these two equations,
generate one where the second order term is eliminated. From this determine the first derivative. Do
the same for n. Then substitute the results in the respective boundary conditions.
Exercise 2.3
Write an EXCEL-Sheet which solves the dimensionless problem of the thermal shock of a plate in
dependence of the Biot number and depict the result with the diagrams shown in the section before.
For the boundary condition, use a second order approximation.
Exercise 2.4
Proceed as in Exercise 2.3, but use a first order approximation for the boundary condition. Compare
the results with those of Exercise 2.3.
Exercise 2.5
A plate of thickness s=100mm is cooled with a Biot number of 2,78 from 1200°C in air with a
temperature of 20°C. Using the diagrams of the last section calculate the maximum temperature
difference and the maximum stress following from it for the materials of exercise 1.1. Would you
expect crack formation?
Exercise 2.6
Solve first exercise 2.4. Then for the problem of exercise 2.5 determine the time of crack initiation.
H. Harmuth 20 53
Modeling and Simulation in Building Materials Technology

Exercise 2.7
Assume a magnesia plate with the properties of Exercise 1.1 and s=100mm at a
temperature of 1200°C. It is put out of the furnace and cooled in air of 20°C. For the heat
transfer use the following relation:
Convection: Radiation:

 T  4  Tair  4 
 c = 1,854(T − Tair )
0 ,25
 r = 4,616    −   (T − Tair )
 100   100  

In case of radiation T has to be the absolute temperature in K.


Use a finite difference procedure to calculate the temperature distribution and the stresses in
dependece of time. Sove the dimensionless problem at first and transform the results
afterwards.

During preparation of this exercise you will face the problem that the difference equation of the
boundary condition is not explicite in the temperature any more. How can you overcome this?

H. Harmuth 21 53

You might also like