Professional Documents
Culture Documents
0307-904X/84/03169-10/$03.00
0 1984 Butterworth & Co. (Publishers) Ltd Appl. Math. Modelling, 1984, Vol. 8, June 169
Determination of thermal diffusivity: V. I. lordanov and A. P. Steward
(14)
@(a) = 5 [Ti(x*) - T(x* >Ti> all’ (1) Because the solution given by equation (14) is not always
i=l uniformly convergent we will present another form of (14).
where pi, i = 1, . . . , N are the time values and X = X* at P3,
First the integral in equation (14) is evaluated by multiple
To determine the theoretical function T(X, r, o) for a integration by parts. We obtain:
given value of (Ywe consider a slab so that the distance m 3
between PI and PZ is L and 0 < X <L. The problem of e(x, t) = ,gl i sin(inx) F,(t)
heat conduction is given as:
a TV, r) a* T(X, T)
p=Iy (2)
a7 ax2 x izI i sin(i7rx)Fi(O) exp [-(irr>‘ot]
O<XfL r>o
The next slowly convergent series in equation (15) is a In the appendix equation (27) is written using different
solution of the following system: boundary functionsf,(t) andf2(t), which are used in the
program. There are also given the first four polynomials
P&(x, t) 1
+-[.fl(~)(l-X)+fz(~)4=O (21) R,(x, t), n = 1) 2, . ) 4.
a.2 Q To determine the unknown thermal diffusivity (Yin
Rz(O,t) = 0 equation (1) we use the following:
(22)
R*(l) t) = 0 (23) ~= Al - 2 aA(x~~~ (y) [8j(X*)-e(X*, Ti, Cw)]
which has a direct solution:
= 0
R
2
(x
>
Q =’ fiCr) l
([~+p~)-f;(r))] x
(28)
01 where x* = X*/L.
Equation (28) is solved by the bisection method to find
--x2
f;(t) -~(f;(r)
-f:(r))X3) (24) the thermal diffusivity.
2
Consequently:
Program characteristics
2 The method presented above has been implemented in a
R,(x,t)= f - sin(inx) [f;(t) - (- l)‘fi(t)]
j=i (In)3 (Y FORTRAN-66 program. The program has been written as
an interactive program; the user responds to a series of
questions, so defining the values of the parameters of the
model. Having obtained some preliminary results a variety
of possible changes can be made. The main properties of
L(f)
--x2-$(l)f&))Xl)
2
(25)
this program are:
(1) It considers one-dimensional heat-conduction in
Notice that every slowly convergent series in equation (15) homogeneous and isotropic solids.
is a solution of a similar system like equations (21)-(23). (2) It uses a least-squares approximation to form the func-
The authors have found common systems whose solutions tions representing the boundary condition.
are polynomials, with forms given by: (3) There are two types of function: exponential and poly-
nomial (parabolic, cubic and quartic). When the user
n+lX2n-j-l ~(n-l)(~)
l C-1) chooses the exponential type of function he has to input
MX> f> =A c (IL i j=O (2n -j-
I
l)! the first coefficient. It is possible to change the first coeffi-
cient or type of function, if the user is not satisfied with
+-l (-lyiA._, ix2(n-i)+l the result. The user can choose either both exponential or
+c both polynomial or any mixture of exponential and poly-
j=2 2(n--i)[2(n-_j)f 11 nomial functions.
1 (-1)” lJ/!“-l’(l) (4) It uses the bisection method to solve the transcendental
equation which gives the thermal diffusivity.
+c j=O (2n -i- l)!
[ (5) The program provides helpful instructions at all stages
(26)
of the user input.
II
+nci (-l)n-i+lA,_l,i (6) The program provides tabular information at three
j=z 2(n--j)[2(f7--j)+ l] x stages, fitting, minimization and temperature .listribution,
so that the user can change his input to obtain satisfactory
where: results. The program has been run on a Harris S125 under
Go(r) =f;(t) -f‘1(t) J/1(t) =f1(4
the Vulcan operating system.
The main components of the program are described in
Thus we can write equation (15) in the following form: Figure 2. A detailed flow chart showing the operation of.
the described model is shown in Figure 3. For an example
(3(x, t) = 2 Rj(X, t) of the dialogue see Appendix 2.
j=l
1 tempera-
tures
tempera-
tures
error
/ \ 12.00
14.00
43.570
47.470
43.567
47.474
0.003
-0.004
0.624379E
-0.855836E
- 02
- 02
Minimization
16.00 51.020 51.016 0.004 0.720150E - 02
Minimise functional @(a)to find thermal diffusivity 18.00 54.200 54.198 0.002 0.436259E - 02
use bisection method to solve transcendental equation 20.00 57.030 57.034 -0.004 -0.651931E-02
6@(a) I o
ti
sivity with values obtained from the thermal diffusivity
when different boundary conditions are imposed. Taking
Temperature distribution \ a given value for 01we first use the temperature distribution
Input-thermal diffusivity option of the program to compute a solution for the
compute temperature distribution distribution with this value. This distribution is calculated
Output-temperature distribution for four different boundary conditions and for the non-
dimensional distance variable in the range x = O.O(O.l)l .O.
Figure 2 The time variable is in the range t = 0(2)20.
The four types of boundary conditions that were chosen
are constant, linear, exponential and parabolic. They were
chosen to give a varied sample of different heating regimes.
In all cases the two surfaces are assumed to have the same
boundary conditions producing a symmetrical problem.
The solutions in each of the four cases were recorded at
x = 0.2 and x = 0.5 and these temperature histories were
used in the thermal diffusivity option of the program. The
estimates of the thermal diffusivities thus obtained are
listed in Table 2.
For each of the boundary conditions the fitting func-
tions exponential, parabolic, cubic and quartic were used.
The exact values of the thermal diffusivity were: (Y=
0.1125 x 1Oe3 in the constant case and 01= 0.8667 x 10e4
in the other cases.
In Figures 4-7 the thick continuous line represents the
boundary conditions whereas the thinner continuous lines
represent the exact solution at x = 0.2 and x = 0.5.
The symbols:
* exponential n cubic
0 parabolic 0 quartic
represent the values obtained using these different fitting
functions. Where a symbol is not included then the fitting
is very close to the exact solution.
The third exercise shows how errors in the input
data values can influence the accuracy of determining
the thermal diffusivity. Exact data values were entered and
the subroutine RANDOM produced a random variation
from these figures. For each of the fitting functions
exponential, parabolic, cubic and quartic the subroutine
produced five sets of random data. The thermal diffusivity
was then calculated for each of the four functions and each
of the five random sets of data. Figures 8 and Y show a
typical set of random errors from an exact set of data
values. The cases cover both symmetrical and non-sym-
metrical boundary conditions. In Table 3 is shown the values
Figure 3 of the thermal diffusivity determined with different maxi-
Fitting function
Exponential O.116832X1O-3 0.78505 X 1 0m4 0.847309 x 1 O-” 0.794935 x 1o-4 9.42
Parabolic 0.117787X10~3 0.865388 X 1 O-4 0.862894 X 1 O-4 0.867707 X 1 O-4 4.70
Cubic O.1133O6X1O-3 0.865408 X 1 O-4 0.863919 X IO-’ 0.865148 X1O-4 0.72
Quartic O.111675X1O-3 0.8645 X 1 0m4 0.862656 X 1O-4 0.864873 X IO-” 0.47
Exact value O.1125X1O-3 0.8667 X 1 0m4 0.8667 X IO-“ 0.8667 X IO-”
Maximum error (%) 4.7 9.42 2.23 8.34
X=Oorl
x
I I I I I I I I I I
2 4 6 8 10 12 14 16 16 2C
2 4 6 8 10 12 14 16 18 20
f, time (5)
t, time (s)
Figure 4 Figure 6
60-
I I I
.p- I
12 ’ I I I I I I I I Y
0 2 4 6 8 10 12 14 16 18 20 L 1 I I I I I I I I
t, time (5) 0 2 4 6 8 10 12 14 16 18 2
t, time (5)
Figure 5
Figure 7
mum errors; l”C, 2”C, 3°C and 4°C in the data. The maxi-
mum percentage error in the thermal diffusivity for each The authors do not discuss the accuracy of the solution
fitting function and each size of temperature error is shown of the inverse heat conduction problem. There are many
in each box. papers presented recently considering the comparison of
methods of solution of the inverse conduction problem.5-6
In Table 3 the authors show the influence of errors in
Conclusions the input on the accuracy of the computed thermal diffu-
A method and a program for determining the thermal sivity. It is clear that accuracy in determining the coeffi-
diffusivity of a solid, is provided in this paper. The method cients of the boundary functions will influence the accu-
is based on the solution of the inverse heat transfer prob- racy of determining the thermal diffusivity.
lem. Before using the program, temperatures must be taken The accuracy of the method depends also on how the
in three interior points in a solid. solid is heated and where the values of the temperature
References
Carslaw, H. S., Jaeger, J. S. ‘Conduction of heat in solids’ (2nd
edn), Oxford University Press, Oxford, 1959
Luimov, A. V. (cd.), ‘Methods for determining thermal con-
ductivity and thermal diffusivity’, Energy , Moscow, 1973 (in
Russian)
Nikitenko, N. I., Kolodnyi, Ju. M. ‘Numerical solution of the in-
verse heat-conduction problem for determining thermal con-
stants’, .7. Eng~g P~zJ~s. 1978, 33 (6), 1461
Beck, J. V. ‘Transient determination of thermal properties’,
Nucl. Eng. Design 1966, 3, 313
Beck, J. V. ‘Criteria for comparison of methods of solution
of the inverse heat-conduction problem’, Nucl. k,‘ng. Design
0 2 4 6 8 10 12 14 16 18 20
1979,53,11
t, time (s)
Kolp, Ya., Lcbedcv, V. V. ‘Comparison of solutions of the
Figure 8 inverse heat-conduction problem by the methods of Tikhonov
and Sparrow’, High Temp. (USSR) 1973, 11, 2
Appendix 1
70- When the type of boundary function is known from equa-
tion (27) we can obtain the solution. The following
60- examples are treated further.
Example 1
Here we fit exponential functions at both points PI and
P,. Let:
*Subsequent values in the table are scaled by IO-‘to give the thermal diffusivity
sink@ZS - 111
X
sin-
-AoA, i
i=l
(-1)’
2irrcu
(in)’ cd- Az
sin (irrx) -&l&2 sin(krx) [f:(O)-(-I)‘f;‘(O)]
and
fr(t) = EoU -El evW26) -&$G sin(kx)[f1)1)(0) - (- l)‘fp(O)]
1
+ f;(t) -f;(t) + f;(t)
sin ]vVZX1 -x)1 -x
+ EoEl exp(-Eat) [ 3! 2!
sin&?&
f;‘(t) -flv) x5 I f;“(l) x4
in R3@, t> =
+2EoE,a( c sin(inx) exp [-(in)‘&] 5! 4!
i=l (i7r)*a-E2
1.
__ f;‘(t) -f;“(f) I f;‘(t) 1 x3
+RZ(x, r) + Rf(x, t) + G(x, r>+ R:(x, f> 3! ?’
I. 23
15!
-f;“(t) I fll’(t) 1 x5
+
+y ,F,$$ sin(inx) exp [-(in)‘cM] 3! 2!
Example 3 +
f;“(t) - f I”(9 + f ;“(t>
Here both functions are polynomials (quartic). Let: 7! 6!
f1(t) = co + C,t + Czt2 + C3t3 + CL/
f2(t) = A0 + A,t + A,t* + A3t3 + Az,t4
+ .fi”‘(l>-f:“(t) CO+CI*TAU+C2*TAU**2-+C3*TAU**3
(, 3! +C4*TAU**4
The next function R5(x, t) can beGobtained in a similar The program finds all coefficients
way. The other variables used in the program are:
-N- the number of values of the time variable
-TN- the first value of the time variable
Appendix 2
-DT- the time step between successive time values
One-dimensional transient conduction -ZL- the distance between PI and P2
Do you want a program explanation? -Xl - the distance between Pl and P3
If you do, type 1, otherwise type 2 -A- the thermal diffusivity
1 -TO- the initial temperature when TAU = 0
Please type any key to continue
Program explanation
This program computes either the unknown thermal
-TIM- the reference temperature greater than any tempera-
diffusivity of a material or simulates the one-dimensional
ture attained within the material
unsteady heat transfer problem to find the temperature
-APMN- and -APMX- the smallest possible value of the
distribution in a solid. The choice depends on the user.
thermal diffusivity (greater than zero) and the largest one.
Before using the program, temperatures must be taken at
These are used as a starting point for a bisection method to
points Pl, P2 and P3 (see below) within a slab of material
find the unknown thermal diffusivity
as it is heated. These points must lie on a line normal to the
surface of the slab and the temperatures must be taken over
a range of values of the time variable.
PI P3 P2 Please type the number of values of the time variable N=?
____-* _~__-----*---__--_---*-----_ 5
What is the first value of the time variable TN= [set]
Please type any key to continue 0
Now type the time-step DT=?
To determine thermal diffusivity we use the method of 2
solving the inverse heat transfer problem. For this case the
Please type the distance between PI and P2 ZL =? [m]
energy equation is solved using two functions, which deter-
.l
mine the boundary conditions at PI and P2. These func-
tions which can be exponential or polynomial are found What is the value of initial temperature TO=? [“Cl
from the temperatures measured at Pl and P2. The other 20
experimental values at P3 are compared with the analytical Now type the value of reference temperature TMAX=? [“C’J
solution to find an accurate value for the thermal diffu- 80
sivity.
The program computes the minimum of the following Please check that the values you have typed are correct
function If you wish to change them type 1, otherwise type 2
F=SUM(Z-_(A))**2 2
where: 1. The sum is taken over the time values when the Please type the values of the temperature at PI T<Pl)=? [‘C]
28 35 41 45 48
temperatures are measured at P3
2. -T(A) - is the analytical solution function and A Are you sure that these values are correct?
is thermal diffusivity
Time Temperature
3. -Z- are the values of the temperature distribution
at the interior point P3 Q.OOOQO 28.00000
2.00000 35.00000
Please type any key to continue 4.QQOOQ 41.00000
6.00000 45.00000
8.QQOOO 48.00000
There are four function types for the boundary conditions
1. Exponential 2. Parabolic If you are type 1, otherwise type 2
3. Cubic 4. Quartic 1
You can choose any combination. We now estimate the temperature function at Pl
The type of exponential function is the following: There are four function types
EU*(l -El*EXP(-E2*TAU)) 1. Exponential 2. Parabolic
3. Cubic 4. Quartic
where: EO, El, E2 are coefficients to be found by curve
fitting, and TA I/ is scaled time variable TAU = T/ZL * * 2. Please type your choice
You have to input the first coefficient EO than the program 1
These are the values of the nondimensional temperature Here are the coefficients CO, Cl,C2, C3. C4
TITA =(T- TO)/(TMAX- TO) Degree of polynomial is 4
TITA 0.0000000.2166670.3333330.3833330.416667
We now fit the function Do you wish to obtain an additional solution?
If you wish type 1, otherwise type 2
CO+Cl*TAlJ+C2*TAU**2+C3*TAU**3 1
+C4*TAlJ**4 Do you want to change the input?
to the values of the non-dimensional temperatures TZTA If you do type 1, otherwise type 2
where TAU is the scaled time variable. 2
Do you want to change the type of function Do you wish to obtain an additional solution?
If you do type 1, otherwise type 2 If you wish type 1, otherwise type 2
2 2
What kind of problem do you want to resolve? Do you want to stop the program?
1. To find unknown thermal diffusivity? - Type 1 If you do type 1, otherwise type 2
2. To find the temperature distribution? - Type 2 2
2 The next customer please! !
Type the thermal diffusivity of the solid A=? [m **2/set] Type 1 if you want to use the program, otherwise type 2
.0000785458 1
Temperature distribution
Time 0.0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1 .o
0 27.94 20.03 20.02 20.01 20.01 20.01 20.00 20.00 20.00 20.00 20.00
2 35.28 27.37 22.94 20.94 20.24 20.06 20.09 20.41 21.63 25.14 33.00
4 40.80 32.32 26.65 23.26 21.51 20.87 21.14 22.51 25.60 31.23 40.00
6 44.94 36.37 30.11 25.95 23.56 22.69 23.24 25.34 29.26 35.19 43.00
8 48.06 39.71 33.28 28.78 26.10 25.12 25.78 28.10 32.11 37.76 45.00