Professional Documents
Culture Documents
Inverse Estimation of Boundary Heat Flux For Heat Conduction Model
Inverse Estimation of Boundary Heat Flux For Heat Conduction Model
)
DOI: 10.4197 / Eng. 21-1.5
1. Introduction
An inverse problem in heat conduction consists of solving a direct
problem and then finding some way to determine the inverse solution [1].
In inverse heat conduction problem [1], the direct problem is used to
determine the temperature distribution inside the solid body when certain
initial and boundary conditions are given, such as temperature or heat
flux, which are known as a function of time. In practice, the temperature
in the solid can be monitored by sensors; however the surface heat flux
cannot be determined experimentally. Theoretically, the surface heat flux
73
74
A. S. A. Alghamdi
75
T(x, t)
=0
x
x=0 L
T (x, t)
= q(t)
x
x=L
A. S. A. Alghamdi
76
For the case where the boundary condition at x = L , i.e., q(t ) , the
initial condition T0 ( x) , and the thermo-physical properties , and k are
all specified, the problem given by Equation (1) is concerned with the
determination of the temperature distribution T ( x, t ) in the interior
region of the solid as a function of time and position.
2.2 Inverse Problem
The inverse problem is concerned with the determination of the
unknown function q(t ) at the surface x = L . To determine the heat flux
is given below:
T ( x, t )
2T ( x, t )
=
,
in 0 < x < L , for 0 < t t f
(2a)
t
x 2
T (0, t )
at x = 0 , for 0 < t t f
(2b)
= 0,
x
T ( L , t )
(2c)
k
= q ( t ) = ( unknown ) , at x = L , for 0 < t t f
x
T ( x,0) = T0 ( x) ,
for t = 0 , 0 < x < L
(2d)
and temperature measurements at an interior location x meas at different
times t i are given by
(3)
T ( x meas , t i ) Ti at x = x meas for t = t i (i = 1,2, , I )
In our problem, the boundary surface function heat flux q(t ) is
unknown. Therefore, this version of the problem is referred to as a
boundary inverse heat conduction problem. The main objective of the
direct problem is to construct the temperature field ( T ( x, t ) , the effect) in
the plate, when all parameters (causes) are specified ( , k , q, T0 ). On the
other hand, the objective of the inverse problem is to estimate heat flux
(the cause) from the knowledge of the measured temperature (the effect)
at some specified section of the medium ( x meas = 0 ).
Alam et al. [12] and Kumar [13] developed an analytical solution,
which consists of finding the temperature in a one-dimensional plate for a
77
(4)
u ( x, t )
2 u ( x, t )
=
,
in 0 < x < L , for t > 0
t
x 2
u (0, t )
at x = 0 , for t > 0
= 0,
x
u ( L, t )
at x = L , for t > 0
= 0,
x
T ( x,0) = T0 ( x) ,
for t = 0 , 0 < x < L
while v ( x, t ) is the solution of
v( x, t )
2 v ( x, t )
=
,
t
x 2
v(0, t )
= 0,
x
v ( L, t )
k
= q (t ) ,
x
T ( x,0) = 0 ,
(5a)
(5b)
(5c)
(5d)
(6a)
at x = 0 , for t > 0
(6b)
at x = L , for t > 0
(6c)
(6d)
The method of separation of variables can be used to find the solution of (5)
A. S. A. Alghamdi
78
u ( x, t ) =
C0
+ C n cos( n x). exp( 2n t )
2 n =1
(7)
where
L
2
(8)
T0 ( ). cos( n )d ,
n = 0,1,2,...
L 0
Making use of the solution of homogenous problem (5), we look for a
solution in the form of a Fourier cosine series with time-dependent
Fourier coefficients, that is
C (t )
(9)
v( x, t ) = 0 + C n (t ) cos( n x)
2
n =1
Using the orthogonality property of eigenfunction, differentiating with
respect to time, integration along x-axis and applying the boundary
conditions leads to
dC n (t )
2
+ 2n C n (t ) =
(1) n .q(t )
(10)
dt
LK
The solution of the above first order linear ordinary differential equation
subject to C n (0) = 0 is given by
Cn =
2
( 1)n exp 2n (t ) .q( )d
C n (t ) =
LK
0
(11)
q( )d
(12)
L
L
1
2
2
+
T
(
)
d
T
(
).
cos(
)
d
n
0
0
L0
n =1 L 0
t
t
2
n
(
)
(
)
+
+
1
exp 2n (t ) .q( )d . cos( n x )
LK 0
n =1 LK
0
(13)
where
n
, n = 1,2,3,...
L
For a constant initial condition, i.e., T ( x,0) = T0 , the solution becomes
n =
T (x, t ) = T0 +
79
t
t
2
n
(
)
(
)
q
1
exp 2n (t ) .q ( )d . cos( n x ) (14)
+
LK 0
n =1 LK
0
The solution (14) for q(t ) = constant agrees with solution presented by
Beck et al. [2].
4. Inverse Problem Solution
4.1 Linear Spline Model
In the previous approaches, the analytical solution was developed
by representing the heat flux as a sixth-degree polynomial, Kumar [13],
and as a cubic spline by Zhong [14]. In this study, the heat flux q(t ) in the
analytical solution is expressed as a linear spline function in t , i.e., as a
piecewise straight line representation. Here the best fit straight line
method is used to match the experimental data with such analytical data
where the least square error between the linear spline and the
experimental data is minimized.
In this method, a polynomial of degree 1 is the simplest polynomial
to use which produces a polygonal path that consists of line segments
that pass through the points. A point-slope formula for a line segment to
represent this piecewise linear curve is given by the following expression
q qi 1
q(t ) = qi 1 + si 1 (t t i 1 ) , where si 1 = i
(15)
t i t i 1
The resulting curve looks like a broken line, as seen in Fig. 2. The
resulting linear spline function can be written in the form
q 0 + s 0 (t t 0 )
q + s (t t )
1
1 1
q (t ) =
qi 1 + si 1 (t t i 1 )
q N 1 + s N 1 (t t N 1 )
for t in [t 0 , t1 ]
for t in [t1 , t 2 ]
for t in [t i 1 , t i ]
for t in [t N 1 , t N ]
(16)
It
is
assumed
that
the
abscissas
are
ordered
t 0 < t1 < t 2 < t N 1 < t N . For a fixed value of t , the interval [t i 1, t i ]
A. S. A. Alghamdi
80
for
t i 1 t t i .
qi 1
qi
q0
q(t )
q2
q1
qN
q N 1
q qi 1
(t t i 1 )
q(t ) = qi 1 + i
t i t i 1
t1
t0
t2
t i 1 t
ti
tN
LK
q
0
i 1
q i q i 1
( t i 1 )d
t i t i 1
t
2
( 1)n exp 2n (t ) . q i 1 + q i q i 1 ( t i 1 ) d cos ( n x )
+
LK
t
t
n =1
i
i 1
(17)
81
j 1 1
(qi + qi 1 )(t i t i 1 )
i =1 2
T (x, t ) = T0 +
LK
1 q j q j 1
2
t t j 1
+ q j 1 t t j 1 +
2 t j t j 1
n (t ti )
e
qi qi 1
(
)
t
t
i 1
i 1 + t t
2
n
i
i 1
j 1
2n (t ti 1 )
e
qi qi 1
(
)
t
t
+
i 1
i 1 t t
n
i
i 1
i =1
cos ( n x )
( 1)n 1 qi qi 1 e 2n (t ti ) e 2n (t ti 1 )
+
LK
n =1
2 2 ti ti 1
2(
)
t
t
n
j 1
q j q j 1
e
+ q j 1 +
t t j 1
q j 1
2
t j t j 1
2n
n
2 (t t
q
q
j
j
1
n
j
1 e
t j t j 1
( )
( )
(18)
4.2 Finite Difference Solution
To check validation of the above solution (18), the direct problem
(1) can be solved using a finite difference method. The method will start
as follows:
Rewriting the partial differential equation in terms of finite
difference approximations to the derivatives
T jn+1 T jn
T jn+1 2T jn + T jn1
(19)
t
x 2
Thus, if for certain n we know the values of T jn for all j, ( initial
condition T ( x,0) = T0 ( x) ), we can solve the equation above to find T jn +1
for each j :
T jn +1 = T jn +
t n
T j +1 2T jn + T jn1 = F0 T jn+1 + T jn1 + (1 2 F0 )T jn
x 2
(20)
A. S. A. Alghamdi
82
where
T (0, t )
= 0:
x
T3
T2
T1
T jn +1 = F0 [2T jn1 + T jn + T jn ] + [1 4 F0 ]T jn
T
T
n +1
j
n +1
2
= F0 [2T ] + [1 2 F0 ]T
n
j 1
= 2 F T + [1 2 F0 ]T
n
0 3
(21)
n
j
(22)
n
2
(23)
n +1
T ( L, t )
= q (t )
x
x T j T j
k
+ h T T = c
(24)
x
2
t
n +1
n
T jn1 _ T jn
x T j T j
k
+ q (t n +1 ) = c
(25)
x
2
t
Rearranging the above equation, we get
1
x
(26)
T jn +1 = F0 2T jn1 + T jn + 2 q (t n +1 )
F
k
0
k
t
where
=
,
F0 =
c
x 2
For stability purposes, we must have the following condition in the
Fourier number
1
t
(27)
F0 =
2
(x ) 2
n
j
400
400
350
T anl
300
anl
T ( C)
fd
300
83
T fd
200
100
250
T ( C)
0
200
400
150
10
20
time (s)
T emperature evolution at x=L
T anl
300
o
T ( C)
100
50
30
T fd
200
100
10
20
time (s)
30
10
20
30
time (s)
Fig. 3. Comparison between the analytical and the finite difference solutions for heat flux
q (t ) = q 0 t 2
S (q) = T i Ti (q)
i =1
(28)
A. S. A. Alghamdi
84
where
T i T (t i ) = measured temperature at t i
N = total number of the unknown parameters
I =total number of measurements, where I N
After defining the direct and inverse formulations, the Matlab code
is written to solve the inverse problem with the help of optimization tool
box [19] which impelements the criteria, and the Levenberg-Marquardt
method. The iterative procedure, the stopping criteria, and the
computational algorithm were chosen based on the Levenberg-Marquardt
Matlab function capability.
5. Test Cases: Inverse Solution for Known Heat Flux Profile
In this section, two test cases are discussed. One test case is for a
heat flux that increases in a linear fashion at time. Exact values of the
simulated temperature history are used. The second case is for a heat flux
which varies at time in a triangular fashion; for this case both exact
temperatures and temperatures with random errors are used.
For the time interval 0 < t 120 , we consider 100 transient
measurements of a single sensor located at x meas = 0 .
5.1 Simulated Measurements
Simulated measurements are obtained from the solution of the
direct problem at sensor location by using prior prescribed values for the
unknown parameters qi of heat flux q(t ) .
85
T (t i ) = Tex (t i ) +
where
(29)
With use of such simulated measurements as the input data for the
inverse analysis, we expect the code to return the same values of
parameters used to find the direct solution.
5.2 Case One: Linear Increase in Heat Flux
The surface heat flux q(t ) increases linearly in time in the form
q(t ) = q 0 t over the interval 0 < t t f . The results of the inverse
calculation for case = 0 are shown in Fig. 4. A comparison of the
exact simulated and estimated values of the temperatures profile at
insulated surface of probe shown in the figure reveals that they are in
excellent agreement. Also, it is clear from the figure the recovered heat
flux coincides with the exact heat flux. To test the code accuracy, a
simulated measurement containing random errors was used. Figure 5
compares the heat flux profiles for both cases to the exact profile, a
A. S. A. Alghamdi
86
500
500
400
400
300
200
100
0
300
200
100
50
100
Time ( s)
150
200
50
10
100
Time ( s)
150
200
150
200
x 10
T-exp
500
q-exact
T-anl
Temperature (oC)
400
300
200
q-anl,=0
100
0
600
Temperature (oC)
Temperature (oC)
600
50
100
Time ( s)
150
200
50
100
Time ( s)
Fig. 4. Results from IHCP algorithm using time-varying heat flux in the form
q(t ) = q0 t .
87
(30)
Heat Flux Profile
x 10
q-exact
q-estimated,=0
q-estimated,=0.001Tmax
8
7
6
5
4
3
2
1
0
20
40
60
80
100
Time ( second)
120
140
160
Fig. 5. The estimation of the time-varying boundary heat flux as straight line fashion.
A. S. A. Alghamdi
88
thermal treatment process that involves rapid cooling of metal alloys for
the purpose of hardening. Experimental measurements of temperature
history of quenching experiments can be used to determine the surface
heat fluxes. The developed model was used to estimate the heat flux for
the quenching experiment developed at Ohio University [20]. The
developed algorithm was used to solve this problem [15], but for the sake
of completeness, the description of the experiments and the results will
be presented.
Exact Simulated Temperature
200
150
T em perature (o C )
T em perature (o C )
150
100
50
200
100
50
50
100
150
200
50
Time ( s)
200
3.5
T-anl
200
q-exact
q-anl,=0
2.5
H eat F lux (w /m 2 )
T em perature (o C )
150
x 10
T-exp
150
100
Time ( s)
100
50
2
1.5
1
0.5
0
50
100
Time ( s)
150
200
-0.5
50
100
150
200
Time ( s)
Fig. 6. Results from IHCP algorithm using time-varying heat flux in the form of triangular
function.
3.5
89
x 10
q-exact
q-estimated,=0
q-estimated, =0.001Tmax
2.5
2
1.5
1
0.5
0
-0.5
20
40
60
80
100
Time ( second)
120
140
160
Fig. 7. The estimation of the time-varying boundary heat flux as triangular function fashion.
A. S. A. Alghamdi
90
Fig. 9. The heat flux values from the plate are very high initially, as
expected and gradually decays to zero. The heat flux is seen to have a
maximum value of 450 kw/m2.
400
Texp
350
Temperatuer( c)
Temperature ( 00C)
300
T ( x = 0 , ti ) = Ti
250
t = ti (i = 1, 2 , , I )
200
150
100
50
0
20
40
60
80
100
120
Time( second)
400
Temperature (oC)
Temperature (oC)
300
exp
200
100
400
300
91
anl
200
100
20
40
60
80
100 120
20
40
Time ( s)
60
80
100 120
Time ( s)
400
x 10
0
T-exp
Heat Flux (w/m2)
Temperature (oC)
300
T-anl
200
100
q(t)
-1
-2
-3
-4
20
40
60
Time ( s)
80
100 120
-5
20
40
60
80
100 120
Time ( s)
Fig. 9. Results from IHCP algorithm using experimental data from quench probe [15].
A. S. A. Alghamdi
92
English Symbols
k
L
q
thermal conductivity
[ W m.K ]
thickness of plate
heat flux
[m]
[W m2 ]
S
T
t
x
u
[-]
temperature
time
space coordinate
[ 0C ]
[s]
[m]
temperature
total number of the unknown parameters
[ C]
[-]
[-]
thermal diffusivity
[ m2 s ]
temperature
eigenvalues
dummy variable
[-]
[-]
dummy variable
[-]
93
Greek Symbols
C]
Subscripts
i
n
0
f
meas
Superscripts
n
T
measured temperature at
ti
References
[1]
[2]
[3]
[4]
[5]
[6]
Ozisik, M. N. and Orlande, R. B. H., Inverse Heat Transfer, Taylor and Francis, New
York ( 2000).
Beck, J. V., Blackwell, B. and St. Clair, C. R., Inverse Heat Conduction, III Posed
Problems, A Wiley Interscience Publication, New York (1985).
Beck, J., Blackwell, V. B. and Sheikh-Haji, A., Comparison of Some Inverse Heat
Condition Methods Using Experimental Data, Int. J. Heat Mass Transfer, 39(17): 36493657 (1996).
Taler, J. and Duda, P., Solving Direct and Inverse Heat Conduction Problems, Springer,
Berlin (2006).
Tervola, P., A Method to Determine the Thermal Conductivity from Measured Temperature
Profiles, Int. J. Heat Mass Transfer, 32:1425-1430 (1989).
Lesnic, D., Elliott, L. and Ingham, D. B., The Solution of an Inverse Heat Conduction
Problem Subject to the Specification of Energies", Int. J. Heat Mass Transfer, 41: 25-32
(1998).
94
[7]
[8]
[9]
[10]
[11]
[12]
[13]
[14]
[15]
[16]
[17]
[18]
[19]
[20]
A. S. A. Alghamdi
Tseng, A. A., Chang, J. G., Raudensky, M. and Horsky, J., An Inverse Finite Element
Evaluation of Roll Cooling in Hot Rolling of Steels, Journal of Material Processing and
Manufacture Science, 3: 387-408 (1995).
Huang, C. H., Ju, T. M. and Tseng, A. A., The Estimation of Surface Thermal Behavior of
the Working Roll in Hot Rolling Process, Int. J. Heat Mass Transfer, 38: 1019-1031 (1995).
Keanini, R. G., Inverse Estimation of Surface Heat Flux Distributions During High Speed
Rolling Using Remote Thermal Measurements, Int. J. Heat Mass Transfer, 41: 275-285
(1998).
S. Abboudi, E. Artiouknine and H. Riad, Computational and Experimental Estimation of
Boundary Conditions for a Flat Specimen, Preliminary program inverse problems in
engineering iii, Port Ludlow Resort & Conference Center, 200 Olympic Place, Port Ludlow,
WA 98365 (1999).
Stolz, G., Jr., Numerical Solutions to an Inverse Problem of Heat Conduction for Simple
Shapes, J. Heat Transfer, 82:20-26 (1960).
Alam, M. K., Pasic, H., Anugarthi, K. and Zhong, R., Determination of Surface Heat Flux
in Quenching, ASME IMECE Proceedings, Nashville, November (1999).
Kumar, Analytical Solution for Inverse Heat Conduction Problem, M.S. Thesis, Ohio
University (1998).
Zhong, R., Inverse Algorithm for Determination of Heat Flux, M.S. Thesis, Ohio University
(1999).
Alghamdi, A. S. A., and Alam, M. K., Inverse heat transfer solution for the flat plate probe,
Proceedings of the 3rd BSME-ASME International Conference on Thermal Engineering,
Dhaka, Bangladesh (2006).
Poulikakos, D., Conduction Heat Transfer, Prentice-Hall, Inc (1994).
Levenberg, K., A Method for the Solution of Certain Non-linear Problems in Least
Squares, Quart. Appl. Math., 2:164-168 (1944).
Mrquardt, D. W., An Algorithm for Least Squares Estimation of Nonlinear Parameters, J.
Soc. Ind. Appl. Math, 11: 431-441 (1963).
Chong, K. P. E. and Zak, H. S., An Introduction to Optimization, Prentice-Hall, Inc (1996).
Zajc, D., Experimental Study of a Quench Process, M.S. Thesis, Ohio University (1998).