Heat Transfer Engineering, 32(10):861–875, 2011

Copyright

C Taylor and Francis Group, LLC
ISSN: 0145-7632 print / 1521-0537 online
DOI: 10.1080/01457632.2011.548592

Inverse Estimation of Surface

Heating Condition in a Finite Slab
With Temperature-Dependent
Thermophysical Properties

JIANHUA ZHOU, YUWEN ZHANG, J. K. CHEN, and Z. C. FENG

Department of Mechanical and Aerospace Engineering, University of Missouri, Columbia, Missouri, USA

Temperature and heat flux at the heated surface can be estimated by solving an inverse heat conduction problem (IHCP)
based on measured temperature and/or heat flux at the accessible locations (e.g., back surface). Most of the previous studies
used temperature measurement data in the objective function, and little work has been done for the inverse numerical
algorithm based on heat flux measurement data. In this study, a one-dimensional IHCP in a finite slab is solved by using the
conjugate gradient method. The heat flux measurement data are, for the first time, incorporated into the objective function
for a nonlinear heat conduction problem with temperature-dependent thermophysical properties. The results clearly show
that the inverse approach of using heat flux measurement data in the objective function can provide much better predictions
than the traditional approaches in which the temperature measurements are employed in the objective function. Parametric
studies are performed to demonstrate the robustness of the formulated IHCP algorithm by testing it for two different materials
under different frequencies of the imposed heat flux along with random errors of the measured heat flux at the back surface.

INTRODUCTION based on the transient temperature and/or heat flux measured at

High-energy laser (HEL) weapons can remotely deliver high-
power laser at the speed of light onto a military target. It is
critical to know the transient of temperature in the target in order
to accurately assess the resulting thermomechanical response.
However, the heated surface is either inaccessible or too hot,
so its temperature cannot be directly measured with thermal
sensors. Similar problems can be found during reentry of a
space vehicle into the atmosphere, as well as in some laser
manufacturing processes [1]. Under these circumstances, the
heated (front) surface temperature can be determined indirectly
by solving an inverse heat conduction problem (IHCP) [2–4]
The authors thank the Test Resource Management Center (TRMC) Test
and Evaluation/Science & Technology (T&E/S&T) Program for their support.
This work is funded by the T&E/S&T Program through the U.S. Army Pro-
gram Executive Office for Simulation, Training, and Instrumentation, contract
W900KK-08-C-0002. The authors also express their gratitude to Dr. James L.
Griggs for his valuable discussions.
Address correspondence to Professor Yuwen Zhang, Department of Me-
chanical and Aerospace Engineering, University of Missouri, Columbia, MO
65211, USA. E-mail: zhangyu@missouri.edu
861
the back surface.
To formulate an IHCP, either temperature or heat flux at
some locations should be measured to provide some additional
information for solving the ill-posed problem. Between them,
temperature is often preferred because it can be measured with
less uncertainty compared to the heat flux [5–8]. Recent stud-
ies, however, have shown that using the measured heat flux as
additional information in an IHCP can increase the stability of
the solution due to the less proneness to the inherent instability
of the ill-posed IHCP [9, 10].
Although IHCPs have been extensively studied for differ-
ent applications in the past decades, most of them used tem-
perature measurement data in the objective function [11–24].
Little work has been done for the inverse numerical algorithm
based on heat flux measurement data. Furthermore, in HEL
weapon applications, the laser energy is delivered to the sur-
face in a periodic way because of the target-spinning or atmo-
sphere variations. Since the formulation of the IHCP is quite
subjective, it is necessary to determine which formulation is
more appropriate for applications with a periodic heat flux for
which it may pose extra difficulties in the solution of the inverse
problems.
862 ZHOU ET AL.

Recently, the authors proposed a robust and error-insensitive (δ x)w (δ x)e

IHCP formulation to reconstruct the front-surface heating con-
dition by using the temperature measurement data as the bound-
ary condition at the back surface while incorporating the heat 1 2 3 …… M-2 M-1 M
x
flux measurement data in the objective function [25]. However, ∆x
the thermophysical properties were assumed to be constant. As
x=0 x= L
temperature significantly changes temporally during the high-
power laser interaction, thermophysical properties could vary Figure 2 A schematic diagram used to describe the calculation of heat flux
variation at x = L.
with temperature. For this case, new equations for the sensitivity
problems and adjoint problems have to be derived to incorpo-
Before the development of the mathematical formulation, the
rate heat flux measurement data in the objective function. To
following dimensionless quantities are defined:
the best knowledge of the authors, such derivations have never
been reported in existing literatures. T∗ x∗ t∗ q∗ C ∗ Tc lc
The objective of this article is to develop an effective algo- T = ,x = ,t = ,q = C = ,
Tc lc tc qc q c tc
rithm that can accurately recover the front surface temperature
based on measured temperature and heat flux at the back sur- k ∗ Tc h ∗ Tc σ∗ Tc4
face for a target with temperature-dependent thermophysical k= ,h = ,σ = (1)
qc l c qc qc
properties. The performance of the developed algorithm is in-
vestigated for different frequency of the imposed periodic heat where the quantities with the asterisk denote the true quantities
flux and different random error of the measured heat flux at the with dimensions, whereas the quantities without the asterisk
back surface. represent the nondimensional quantities. In the remainder of
this paper, all the equations are expressed in nondimensional
form unless otherwise noted.

MODEL DESCRIPTION
The Direct Problem
For the case when a laser beam size is much larger than
The direct problem can be expressed as follows:
the thickness of a heated target, the IHCP can be treated as a


one-dimensional problem. To illustrate the methodology of the ∂T ∂ ∂T
inverse heat transfer algorithms employed in this study, a finite C(T ) = k(T ) (2)
∂t ∂x ∂x
slab with a thickness of L∗ , as shown in Figure 1, is consid-
ered. Initially, the slab is uniformly at temperature T0∗ and is T = T0 for 0 ≤ x ≤ L , t = 0 (3)
subjected to a high intensity laser heating from t∗ = 0+ at its
front surface. The purpose of this study is to demonstrate the ∂T
effectiveness and accuracy of the proposed IHCP formulation −k(T ) = q1 (t) for x = 0, t > 0 (4)
∂x
in reconstructing the heat flux q1∗ (t) and temperature T1∗ (t) at
the front surface of a target with temperature-dependent thermo- T (L , t) = YT L (t) for x = L , t > 0 (5)
physical properties, based on the measured temperature and heat
flux at the back surface. Due to the fact that temperature mea- where C(T) is temperature-dependent volume specific heat, and
surement contains much less errors compared to the heat flux k(T) is temperature-dependent thermal conductivity.
measurement [5–8], the back surface temperature YT∗ L (t) is used In the direct problem described here, the front-surface heat
as the boundary condition and the back surface heat flux Yq∗L (t) flux q1 (t) and the back-surface temperature YT L (t) are consid-
is adopted in the objective function. Both the specific heat and ered to be known. The objective of the direct problem here is to
the thermal conductivity of the slab considered are temperature determine the transient temperature and heat flux distribution in
dependent. the target.

The Inverse Problem

For the inverse problem, the heat flux at x = 0 is unknown and

* * 1st B.C. Y TL*(t*) needs to be recovered, but everything else in the direct problem
2nd B.C. q1 (t )=?
O x is known. The additional information needed in the estimation
YqL*(t*) is used in
of the front surface heat flux is available from the readings of
a heat flux sensor installed at the back surface. The inverse
x*=0 x*=L* problem can thus be stated as: The heat flux measurements at
Figure 1 Physical model. the back surface, Yq L (t), is utilized to recover the front surface
heat transfer engineering vol. 32 no. 10 2011
 ZHOU ET AL. 863

Stainless Steel 304 Conjugate Gradient Method for Minimization

6000 40
Volumetric specific heat (kJ/m -K)

The iterative process based on the conjugate gradient method

Thermal conductivity (W/m-K)

35
3

5500 (CGM) [3, 4] is now derived for the estimation of unknown

30 heat flux q1 (t) by minimizing the objective function S given by
5000 Eq. (6). The front surface heat flux q1 (t) at iteration k + 1 is
25
advanced by
4500 20 q1k+1 (t) = q1k (t) − βk d k (t) (7)

4000
15 where βk is the search step size from iteration k to k + 1, which
10
is addressed in the next section, and d k (t) is the direction of
Volumetric specific heat descent (i.e., search direction), given by:
3500
5
Thermal conductivity d k (t) = ∇ S[q1k (t)] + γk d k−1 (t) (8)
3000 0
200 400 600 800 1000 1200 1400 1600 which is a conjugation of the gradient direction ∇ S[q1k (t)]
at
Temperature (K) iteration k and the direction of descent d k−1 (t) at iteration k—1.
(a) The conjugate coefficient γk is determined by:
 tf
Aluminum alloy 2024-T6 ∇ S[q1k (t)] · {∇ S[q1k (t)] − ∇ S[q1k−1 (t)]}dt
3000 240 γ = 0
k
 tf (9)
0 {∇ S[q1 (t)]} dt
k−1 2
Volumetric specific heat (kJ/m -K)

Thermal conductivity (W/m-K)

3

2800
200 with γ0 = 0. To perform the iterations according to Eq. (7), the
step size βk and the gradient of the functional ∇ S[q1k (t)] need
160 to be determined. To do so, a sensitivity problem and an adjoint
2600 problem are constructed in the following.
120
2400
80
Sensitivity Problem and Search Step Size
2200 Volumetric specific heat
40
Thermal conductivity The sensitivity and adjoint problems can be obtained by the
limiting approach described in reference [4]. It is assumed that
2000 0 the unknown heat flux q1 (t) is perturbed by an amount ξq1 (t)
200 300 400 500 600 700 800
with ξ being a real number. Thus, the temperature T(x, t) under-
Temperature (K) goes a variation ξT (t), that is:
(b)
Figure 3 Thermal properties of stainless steel 304 and aluminum alloy 2024- Tξ (x, t) = T (x, t) + ξT (x, t) (10)
T6 [24].
where the subscript ξ refers to a perturbed variable.
The perturbation of temperature causes variations on the ther-
heat flux. The front surface temperature is computed according mophysical properties. The resulting perturbed quantities are
to the temperature–heat flux relation defined by the classical linearized as:
Fourier’s law.
dC
It must be pointed out that in the inverse approach presented Cξ (Tξ ) = C(T ) + · ξ · T (11)
in this study, the temperatures at the back surface are also mea- dT
sured. But they are used as a known boundary condition at the dk
back surface. kξ (Tξ ) = k(T ) + · ξ · T (12)
dT
If the measurement data are dense in time, they can be ap-
proximated as continuous. For this case, the inverse solution can For convenience in the subsequent analysis, the differential
be obtained by minimizing the following ordinary least-squares equation (2) is rewritten as:
norm:

∂T ∂ ∂T
 tf D(T ) = C(T ) − k(T ) (13)
∂t ∂x ∂x
S[q1 (t)] = {Yq L (t) − q[L , t; q1 (t)]}2 dt (6)
0 The perturbed form of the above equation becomes:


where Yq L (t) and q[L , t; q1 (t)] are the measured and computed ∂ Tξ ∂ ∂ Tξ
heat fluxes at the back surface, respectively. Dξ (Tξ ) = Cξ (Tξ ) − kξ (Tξ ) (14)
∂t ∂x ∂x
heat transfer engineering vol. 32 no. 10 2011
864 ZHOU ET AL.

240
Front surface heat flux (W/cm 2)

1000

Front surface temperature (K)

900
220
800

700
200
600

500
180
400

300
160 200
0.0 0.5 1.0 1.5 2.0 2.5 3.0 0.0 0.5 1.0 1.5 2.0 2.5 3.0
Time (s) Time (s)
(a) (b)

3.0 1000
Back surface heat flux (W/cm2)

Back surface temperature (K)

2.5 900

800
2.0
700
1.5 600

1.0 500

400
0.5
300
0.0 200
0.0 0.5 1.0 1.5 2.0 2.5 3.0 0.0 0.5 1.0 1.5 2.0 2.5 3.0
Time (s) Time (s)
(c) (d)
Figure 4 Results for the direct problem described by Eqs. (2), (3), (40) and (41) (stainless steel 304) with measurement error φ∗ = 1% · [Yq∗Lexact (t)]max and
frequency f = 1 Hz: (a) front surface heat flux; (b) front surface temperature; (c) back surface heat flux; (d) back surface temperature.

To formulate the sensitivity problem, we apply the limiting   tf

process [4] to Eqs. (13) and (14): S q1k+1 (t) = {Yq L (t) − q[L , t; (q1k − βk d k )]}2 dt (20)
0

Dξ (Tξ ) − D(T ) Linearizing q[L , t; (q1k − βk d k )] by the Taylor-series expan-

lim =0 (15)
ξ→0 ξ sion, Eq. (20) takes the following form:

A similar limiting process is employed to the boundary and   tf
S q1k+1 (t) = {Yq L (t) − q[L , t; q1k ] − βk q(d k )}2 dt (21)
initial conditions of the direct problem. After some manipula- 0
tions, the sensitivity problem is described by:
In Eq. (21), the heat flux q[L , t; q1k ] is solved from the direct
∂(CT ) ∂ (kT )
2 problem [Eqs. (2)–(5)] with the estimated q1k as the boundary
= (16) condition at x = 0. The sensitivity function q(d k ) is the heat
∂t ∂x2
flux variation at x = L and time t, which is calculated using
T (x, 0) = 0 for 0 ≤ x ≤ L , t = 0 (17) Fourier’s law based on the temperature variation T that is
obtained from Eqs. (16)–(19) by letting q1 = d k . The calcula-
∂(kT ) tion of the heat flux variation q(d k ) will be addressed in detail
− = q1 (t) for x = 0, t > 0 (18) in the next section. The search step size βk can be determined
∂x
by minimizing the objective function, given by Eq. (21), with
T (x, t) = 0 for x = L , t > 0 (19) respect to βk :
 tf
The preceding equations are to determine the temperature {q[L , t; q1 (t)] − Yq L (t)} · q[L , t; d k (t)] · dt
variation T (x, t) that is caused by the perturbationq1 (t).
β = k 0
 tf (22)
The objective function at iteration k + 1 can be obtained by
{q[L , t; d (t)]} · dt
k 2
replacing q1 in Eq. (6) with q1k+1 given by Eq. (7): 0

heat transfer engineering vol. 32 no. 10 2011

 ZHOU ET AL. 865

Adjoint Problem and Gradient Equation 3.0

2.5

at back surface (W/cm2)

Multiplying the right-hand side of Eq. (13) by a Lagrange

Measurement heat flux

multiplier λ(x, t), integrating over the time and space domains, 2.0
and then adding the result to Eq. (6) leads to:
 tf 1.5
S[q1 (t)] = {Yq L (t) − q[L , t; q1 (t)]}2 dt
0 1.0

  
 0.5
L tf
∂T ∂ ∂T
+ C(T ) − k(T ) · λd xdt
x=0 t=0 ∂t ∂x ∂x 0.0

0.0 0.5 1.0 1.5 2.0 2.5 3.0

(23)
Time (s)
The preceding extended functional S[q1 (t)] will undergo a vari- (a)
ation S[q1 (t)] when the unknown quantity q1 (t), temperature Exact solution obtained
T, heat flux q, volume specific heat C, and thermal conductiv-
250 from Eqs.(2), (3), (40) and (41)
ity k undergo variations ξq1 (t), ξT (t), ξq(t), dC dT
· ξ · T ,
Recovered value using CGM method
dk
·ξ·T . The variation S[q 1 (t)] can be obtained by applying

at front surface (W/cm2)

dT
RMS deviation = 0.255 W/cm 2
the following limiting process:

Recovered heat flux

225 Maximum deviation = 1.338 W/cm 2
Sξ [q1 (T )ξ ] − S[q1 (T )]
S[q1 (t)] = lim (24)
ξ→0 ξ
where the term Sξ [q1 (T )ξ ] is obtained by reexpressing Eq. (23) 200
with the perturbed quantities defined in Eqs. (10)–(12).
After performing integration, applying the boundary and ini-
tial conditions of the sensitivity problems, and letting the terms
175
containing T (x, t) be zero, we get the following adjoint prob-
lem for λ(x, t): 0.0 0.5 1.0 1.5 2.0 2.5 3.0
Time (s)
∂λ ∂ 2λ q[x, t; d k (t)]
C + k 2 + 2{q[x, t; q1 (t)] − Yq L (t)} · (b)
∂t ∂x T [x, t; d k (t)]
Exact solution obtained
× δ(x − L) = 0 in 0 < x < L , for t > 0 (25) 1100 from Eqs.(2), (3), (40) and (41)
1000
Recovered value using CGM method
λ(x, t f ) = 0 in 0 < x < L , for t = tf
surface as by-product (K)

(26) RMS deviation = 0.050 K

900
Temperature at front

Maximum deviation = 0.196 K

∂λ(0, t)
=0 at x = 0, for t > 0 (27) 800
∂x 700
λ(L , t) = 0 at x = L , for t > 0 (28) 600

where δ(·) is the Dirac delta function. 500

The adjoint problem, Eqs. (25)–(28), is different from the 400
standard initial value problems, Eqs. (2)–(5), for which the final 300
time condition at time t = tf is specified instead of the cus-
tomary initial condition (t = 0). However, this problem can be 0.0 0.5 1.0 1.5 2.0 2.5 3.0
converted to an initial value problem by transformation of the Time (s)
time variables as τ = t f − t. (c)
The heat flux variation q[x, t; d k (t)] in Eq. (25) can be Figure 5 Inverse results for stainless steel 304 with measurement error of
calculated by Fourier’s law based on the temperature variation φ∗ = 1% · [Yq∗Lexact (t)]max and frequency of f = 1 Hz: (a) back-surface heat
T [x, t; d k (t)]. However, considering the fact that the thermal flux measurements with random error; (b) estimated front-surface temperature
by CGM inverse method; (c) front-surface heat flux as a by-product.
conductivity is temperature dependent, special care should be
given to the derivation of q[x, t; d k (t)], which is next described sional. In total, there are M grid points, which are located at the
with the aid of Figure 2. centers of control volumes. The dash lines represent the faces
Figure 2 shows the one-dimensional finite difference mesh of control volumes. We focus on the control volume of grid
used in this study. All the quantities in Figure 2 are nondimen- point M − 1 adjacent to the boundary x = L (shadowed area in
heat transfer engineering vol. 32 no. 10 2011
866 ZHOU ET AL.

3 Exact solution obtained

250 from Eqs.(2), (3), (40) and (41)
at back surface (W/cm )

Recovered value using CGM method

Measurement heat flux
2

2 2

at front surface (W/cm2)

RMS deviation = 5.247 W/cm

Recovered heat flux

2
225 Maximum deviation = 13.314 W/cm
1

200
0

0.0 0.5 1.0 1.5 2.0 2.5 3.0 175

Time (s) 0.0 0.5 1.0 1.5 2.0 2.5 3.0
(a)
Time (s)
Exact solution obtained (a)
250 from Eqs.(2), (3), (40) and (41)
Exact solution obtained
Recovered value using CGM method
1100 from Eqs.(2), (3), (40) and (41)
RMS deviation = 1.0 W/cm 2
at front surface (W/cm )
2

1000
Recovered value using CGM method
Recovered heat flux

225 Maximum deviation = 3.127 W/cm 2 surface as by-product (K) RMS deviation = 1.020 K
900
Maximum deviation = 2.277 K
Temperature at front

800

200 700
600
500
175 400
0.0 0.5 1.0 1.5 2.0 2.5 3.0 300
Time (s)
0.0 0.5 1.0 1.5 2.0 2.5 3.0
(b)
Time (s)
Exact solution obtained (b)
1100
from Eqs.(2), (3), (40) and (41) Figure 7 Inverse results for stainless steel 304 with measurement error of
Recovered value using CGM method φ∗ = 1% · [Yq∗Lexact (t)]max and frequency of f = 5 Hz: (a) estimated front-
1000
surface as by-product (K)

RMS deviation = 0.197 K surface heat flux by CGM inverse method; (b) front-surface temperature as a
900 by-product.
Temperature at front

Maximum deviation = 0.498 K

800
The heat flux at x = L can be approximately estimated by:
700
TM−1 − TM
600 q[L , t] = km (29)
(δx)e
500
where km stands for the mean thermal conductivity of k M−1 and
400
k M . In the present study, km is calculated as the harmonic mean
300 [26] of k M−1 and k M :
0.0 0.5 1.0 1.5 2.0 2.5 3.0 2k M−1 k M
Time (s) km = (30)
(c)
k M−1 + k M
Figure 6 Inverse results for stainless steel 304 with measurement error of
φ∗ = 5%·[Yq∗Lexact (t)]max and frequency of f = 1 Hz: (a) back-surface heat flux
Substituting Eq. (30) into Eq. (29), one gets:
measurements; (b) estimated front-surface heat flux by CGM inverse method; 2k M−1 k M TM−1 − TM
(c) front-surface temperature as a by-product. q[L , t] = · (31)
k M−1 + k M (δx)e

Figure 2) since the source term in Eq. (25) only exists at this If the thermal conductivities are constant, the calculation
control volume; x is the width of this control volume, and of the heat flux variation q[L , t] is straightforward:
(δx)w and (δx)e are the distances between grid point M − 1 and 2k M−1 k M TM−1 − TM
q[L , t] = · (32)
its neighboring points (i.e., M − 2 and M). k M−1 + k M (δx)e
heat transfer engineering vol. 32 no. 10 2011
 ZHOU ET AL. 867

Exact solution obtained 1: Exact solution obtained using T-dependent properties

from Eqs.(2), (3), (40) and (41) 2: Inverse solution obtained using T-dependent properties
250
Recovered value using CGM method 3: Inverse solution obtained using constant properties
225

at front surface (W/cm2)

3
RMS deviation = 5.793 W/cm2 1

Recovered heat flux

at front surface (W/cm2 )

2
Recovered heat flux

225 Maximum deviation = 12.522 W/cm 2

200

200

175

175 0.0 0.5 1.0 1.5 2.0 2.5 3.0

Time (s)
0.0 0.5 1.0 1.5 2.0 2.5 3.0 (a)
Time (s)
(a) 1: Exact solution obtained using T-dependent properties
1200 2: Inverse solution obtained using T-dependent properties
Exact solution obtained 1100 3: Inverse solution obtained using constant properties

surface as by-product (K)

from Eqs.(2), (3), (40) and (41) 1000
1100

Temperature at front
Recovered value using CGM method 900
1000 3
RMS deviation = 1.122 K 800
surface as by-product (K)

900 700 1, 2
Maximum deviation = 2.603 K
Temperature at front

800 600

700 500
400
600
300
500
0.0 0.5 1.0 1.5 2.0 2.5 3.0
400 Time (s)
(b)
300
Figure 9 Comparison of the inverse solutions obtained using temperature-
0.0 0.5 1.0 1.5 2.0 2.5 3.0 dependent thermal properties and using constant thermal properties (material:
Time (s) stainless steel 304, φ∗ = 5% · [Yq∗Lexact (t)]max , f = 5 Hz): (a) recovered heat
(b) flux; (b) recovered temperature.
Figure 8 Inverse results for stainless steel 304 when the measurement error
and the frequency are increased (φ∗ = 5% · [Yq∗Lexact (t)]max , f = 5 Hz): (a)
estimated front-surface heat flux by CGM inverse method; (b) front-surface It can be seen by comparing Eqs. (33) and (32) that
temperature as a by-product. when the front-surface heat flux q1 (t) is subject to a
variation q1 (t), the variation of back-surface heat flux
q[L , t] calculated with temperature-dependent thermal prop-
However, in this study, the thermophysical properties are erties will be larger than that calculated with constant thermal
considered to be temperature dependent. For this case, the heat properties.
flux q[L,t] is a function of k M−1 , k M , TM−1 , and TM . The heat After letting the terms containing T (x, t) be zero, the fol-
flux variation q[L , t] should be computed in the following lowing integral term is left:
way:  tf
∂q ∂q ∂q S[q1 (t)] = λ(0, t) · q1 (t)dt (34)
q[L , t] = k M−1 + k M + TM−1 0
∂k M−1 ∂k M ∂ TM−1
By assuming that the unknown function q1 (t) belongs to the
Hilbert space of square-integrable functions in the time domain
∂q 2k 2M TM−1 − TM
+ TM = k M−1 0 < t < t f , we can write [4, 27]:
∂ TM (k M−1 + k M )2 (δx)e  tf
S[q1 (t)] = ∇ S[q1 (t)] · q1 (t)dt (35)
2k 2M−1 TM−1 − TM
+ k M 0
(k M−1 + k M )2 (δx)e A comparison of Eq. (34) with Eq. (35) leads to the following
expression for the gradient of functional ∇ S[q1 (t)]:
2k M−1 k M TM−1 − TM
+ (33)
k M−1 + k M (δx)e ∇ S[q1 (t)] = λ(0, t) (36)
heat transfer engineering vol. 32 no. 10 2011
868 ZHOU ET AL.

Exact solution COMPUTATIONAL PROCEDURE

250 Inverse solution using
temperature measurements The solution procedure of the preceding IHCP using the
CGM is summarized as follows. Start with an initial guess q10 (t)
at front surface (W/cm2)

2
RMS deviation = 13.769 W/cm
for q1 (t), set k = 0, and then perform the steps here:
Recovered heat flux

2
225 Maximum deviation = 23.821 W/cm
Step 1. Solve the direct problem given by Eqs. (2)–(5) for T (x, t)
based on the value q1k (t).
200 Step 2. Check the stopping criterion Eq. (37). Stop the itera-
tion if satisfied. Otherwise, continue the following solution
procedure.
175
Step 3. Solve the adjoint problem given by Eqs. (25)–(28) for
λ(x, t).
0.0 0.5 1.0 1.5 2.0 2.5 3.0 Step 4. Compute the gradient of the functional ∇ S[q1 (t)] from
Time (s) Eq. (36).
(a)
Step 5. Compute the conjugate coefficient γk and the direction
Exact solution of descent d k (t) from Eqs. (9) and (8), respectively.
1100 Inverse solution using Step 6. Set q1k (t) = d k (t) and solve the sensitivity problem
1000 temperature measurements given by Eqs. (16)–(19) for T (x, t) and then q(x, t).
Step 7. Compute the search step size βk from Eq. (22).
surface as by-product (K)

900
RMS deviation = 2.614 K
Step 8. Compute the new estimation for q1k+1 (t) from Eq. (7)
Temperature at front

Maximum deviation = 4.581 K

800 and then return to Step 1.
700
One of the advantages of using the conjugate gradient method
600
to solve the IHCP is that the initial guess (i.e., q10 (t)) of the un-
500 known quantity can be chosen arbitrarily. In all the cases studied
400 in this study, the initial guess is taken as zero for convenience.
300
0.0 0.5 1.0 1.5 2.0 2.5 3.0
Time (s) RESULTS AND DISCUSSION
(b)
Figure 10 Inverse results for stainless steel 304 using temperature measure- Generation of Measurement Data
ment data in the objective function while the heat flux measurement data are
used as the back-surface boundary condition. (φ∗ = 5% · [Yq∗Lexact (t)]max , f = Instead of conducting an actual experiment, the measurement
5 Hz): (a) estimated front-surface heat flux by CGM method; (b) front-surface
data of temperature and heat flux are generated numerically by
temperature as a by-product.
solving the direct problem described by the governing Eq. (2)
with initial condition given by Eq. (3) and the following bound-
ary conditions:
∂ T (L , t)

Stopping Criterion −k = q−h(T −T∞ )−εσ T 4 −T∞
4
at x = 0, for t > 0
∂x
(40)
The discrepancy principle is used as the stopping criterion
[3, 4]: ∂ T (L , t)

−k = h(T − T∞ )+εσ T 4 − T∞ 4
at x = L , for t > 0
S[q1 (t)] < χ (37) ∂x
(41)
where χ denotes the tolerance. Assume that the absolute value where q is the periodic heat flux imposed on the front surface.
of the heat flux residuals can be approximated by: The results are obtained with the validated computer code [28].
In the inverse heat transfer analysis, the simulated measurement
Yq L (t) − q[L , t; q1 (t)] ≈ φ (38) data of back surface temperature and heat fluxes are used as the
where φ is the standard deviation of the measurements. Sub- boundary condition and are employed in the objective function,
stituting Eq. (35) into Eq. (8), the tolerance χ for the stopping respectively, to estimate the heat flux and temperature at the
criterion is obtained: front surface. These recovered heat flux and temperature will
be compared with the front-surface temperature and heat flux
χ = φ2 t f (39) calculated from the above direct problem [Eqs. (2), (3), (40)
heat transfer engineering vol. 32 no. 10 2011
 ZHOU ET AL. 869

Frequency f=6.0 Hz, δ=10%*[YqLexact(t)]max Frequency f=6.0 Hz, δ=10%*[YqLexact(t)]max

300
Exact solution obtained 1300 Exact solution obtained
from Eqs.(2), (3), (40) and (41) 1200 from Eqs.(2), (3), (40) and (41)
275

surface as by-product (K)

1100 Recovered value using CGM method
at front surface (W/cm2 )

Recovered value using CGM method

Recovered heat flux

Temperature at front
2 1000 RMS deviation = 1.773 K
250 RMS deviation = 10.137 W/cm
2
900 Maximum deviation = 3.699 K
Maximum deviation = 18.607 W/cm 800
225 700
600
200 500
400
175 300
0.0 0.5 1.0 1.5 2.0 2.5 3.0 0.0 0.5 1.0 1.5 2.0 2.5 3.0
Time (s) Time (s)
(a) (b)

Frequency f=10.0 Hz, δ=10%*[YqLexact(t)]max

300 Frequency f=10.0 Hz, δ=10%*[YqLexact(t)]max
1100 Exact solution obtained
Exact solution obtained
1000 from Eqs.(2), (3), (40) and (41)
275 from Eqs.(2), (3), (40) and (41)

surface as by-product (K)

at front surface (W/cm2)

900 Recovered value using CGM method

Recovered value using CGM method
Recovered heat flux

Temperature at front
2
RMS deviation = 1.939 K
250 RMS deviation = 14.209 W/cm 800
Maximum deviation = 3.875 K
2 700
Maximum deviation = 24.918 W/cm
225 600
500
200
400

175 300
0.0 0.4 0.8 1.2 1.6 2.0 0.0 0.4 0.8 1.2 1.6 2.0
Time (s) Time (s)
(c) (d)
Figure 11 Inverse results for stainless steel 304 with higher measurement error: (a) estimated front-surface heat flux for φ∗ = 10% · [Yq∗Lexact (t)]max and f = 6 Hz;
(b) front-surface temperature as a by-product for φ∗ = 10% · [Yq∗Lexact (t)]max and f = 6 Hz; (c) estimated front-surface heat flux for φ∗ = 10% · [Yq∗Lexact (t)]max and
f = 10 Hz; (d) front-surface temperature as a by-product for φ∗ = 10% · [Yq∗Lexact (t)]max and f = 10 Hz.

and (41)] to examine the accuracy of the present inverse heat Results of IHCP
conduction algorithm.
The simulated measurement data computed from Eqs. (2), The following numerical analyses are performed for stain-
(3), (40), and (41) provide the exact (errorless) measurement. less steel 304 and aluminum alloy 2024-T6. Their temperature-
To account for the measurement error in back surface heat flux, dependent thermophysical properties [29] are plotted in Figures
we add an error term to Yq Lexact (t) in the form: 3a and b, respectively. When the highest temperature exceeds
the maximum temperature at which thermophysical properties
Yq L (t) = Yq Lexact (t) + ωφ (42) are available, the thermophysical properties are obtained by lin-
ear extrapolation. Other parameters are: L∗ = 2.5 mm, T0∗ = 300
∗
where Yq Lexact (t) are the data simulated from the direct problem K, T∞ = 300 K, h∗ = 5 W/(m2 · K), ε = 0.92. The front surface
described by Eqs. (2), (3), (40), and (41); φ is the standard heat flux is assumed to be q ∗ = qconst
∗ ∗
+ 0.1qconst sin(2πf t ∗ )
2
deviation of the measurements and is set as a percentage of the (W/m ). Unless specified otherwise, the following simula-
∗
highest heat flux value at the back surface; and ω is a random tion parameters are used: qconst = 200 W/cm2, f = 1.0 Hz,
∗ ∗
variable having a normal distribution with zero mean and unitary φ = 1% · [Yq Lexact (t)]max . Here, the simulation parameters are
standard deviation. The measurement data obtained by Eq. (42) specified in real values, which can be converted to dimensionless
will contain random errors that have a normal distribution with quantities according to Eq. (1). The finite-difference method is
standard deviation equal to φ. used to solve the direct problem, sensitivity problem, and ad-
The bias error in heat flux measurement is ignored. In ad- joint problem. Time discretization is obtained by applying a
dition, we assume that the back surface temperature contains fully implicit scheme. Total grid number along the x direction is
no errors, since temperature can be measured with much less taken as 20 after a mesh refinement test. The final time is chosen
uncertainty compared to the heat flux [7, 8]. as t ∗f = 3.6s, and the time step t ∗ = 0.1s. Therefore, a total

heat transfer engineering vol. 32 no. 10 2011

870 ZHOU ET AL.

Exact solution obtained deviations between the recovered values and the exact solutions
250 from Eqs.(2), (3), (40) and (41) are also given in the figure. The RMS deviation between any
Recovered value using CGM method two quantities A and B is defined as:

RMS deviation = 2.099 W/cm
2

1  N
at front surface (W/cm )

RMS = 
2

(Ai − Bi )2
Recovered heat flux

2
225 Maximum deviation = 10.217 W/cm (43)
N i=1

where N is the total number of sample points.

200 As seen from Figure 5b and Figure 5c, both the recovered heat
flux and the by-product temperature at the front surface are in
excellent agreement with the exact results obtained from solving
the direct problem. The RMS deviations between the recovered
175
values and the exact solutions are so small that one cannot tell
0.0 0.3 0.6 0.9 1.2 1.5 apart the two plotting curves in Figure 5b and Figure 5c. In the
Time (s) numerical simulation, no prior information on the functional
(a)
form of the front-surface heating condition is required.
Exact solution obtained Figure 6 presents the effect of heat flux measurement error on
1000 the accuracy of the present IHCP formulation. The simulation
from Eqs.(2), (3), (40) and (41)
900 Recovered value using CGM method parameters are almost the same as those in Figure 5 except for
surface as by-product (K)

RMS deviation = 0.089 K the random error on the measured heat flux. It can be seen from
Temperature at front

800
Maximum deviation = 0.482 K
700
600
500
400
300
200
0.0 0.3 0.6 0.9 1.2 1.5
Time (s)
(b)
Figure 12 Inverse results for aluminum alloy 2024-T6 (φ∗ = 5% ·
[Yq∗Lexact (t)]max , f = 5 Hz): (a) estimated front-surface heat flux by CGM
inverse method; (b) front-surface temperature as a by-product.
of 36 unknown heat fluxes at discrete time moments are to be
determined in the present study. The heat flux and temperature
measurements at the back surface are also performed at the same
36 discrete time moments. To provide a better understanding on
the physical meanings of the simulation results, the results are
presented in dimensional form instead of dimensionless from.
Figures 4–9 present the inverse solution for stainless steel
304. Figure 4 shows the results calculated from the direct prob-
lem described by Eqs. (2), (3), (40) and (41). It can be seen from
Figure 4b that there are some fluctuations in the front surface
temperature because the front surface is subjected to a sinusoidal
heat flux heating. However, almost no fluctuations are observed
in the back surface heat flux (Figure 4c) and temperature (Figure
4d), due to damp and delay affects in heat diffusion phenomena.
To avoid the numerical instabilities in the neighborhood of final
simulation time t ∗f [4], the actual final time is taken as t ∗f = 3.6
s in all the simulations of this study.
Figure 5 shows the inverse results for the case in Figure 4.
Figure 5a gives the heat flux at the back surface with 1% random
error via Eq. (42). The root mean square (RMS) and maximum
heat transfer engineering
Figure 6b that when the heat flux measurement error is increased
to 5%, the front surface heat flux can still be reconstructed with
a good accuracy. The front surface temperature can also be
recovered as a by-product with excellent accuracy (Figure 6c).
Figure 7 tests the effect of the frequency of the sinusoidal
component on the accuracy of heat flux and temperature re-
covered from the proposed IHCP formulation. The simulation
parameters are the same as those in Figure 5 except for the fre-
quency of the sinusoidal component. It is found from Figure 7a
that when the frequency f is increased to 5 Hz, the phase of the
estimated heat flux agrees well with that of the exact solutions,
and there is a slight degradation in the accuracy of the amplitude
of the recovered heat flux. Again, the front surface temperature
can be recovered with high accuracy as a by-product of the
inverse algorithm (Figure 7b).
Figure 8 shows the inverse solutions when both the measure-
ment error and the frequency are increased (the measurement
error is 5% and the frequency is 5 Hz). It appears that the accu-
racy of the recovered heat flux is still reasonably good, and the
front surface temperature can be accurately estimated.
A focus of this paper is on the use of temperature-
dependent thermal properties. To show the effect of using
temperature-dependent thermal properties, Figure 9 shows the
comparison between the inverse solutions obtained using
temperature-dependent thermal properties and using constant
thermal properties. The constant thermal properties used in
Figure 9 are thermal conductivity k∗ = 14.9 W/(m-K) and vol-
ume specific heat C∗ = 3768.3 kJ/(m3-K). Other simulation pa-
rameters are the same as those in Figure 8. The exact solutions
in Figure 9 are obtained using temperature-dependent thermal
properties. As is seen from the comparison between Figures 8
and 9, for the same object size and laser irradiation parameters,
the constant property assumption will result in totally different
inverse solutions. This indicates that constant property assump-
tion may introduce considerable errors in IHCP analysis, which
vol. 32 no. 10 2011
 ZHOU ET AL. 871

Frequency f=10.0 Hz, δ=20%*[YqLexact(t)]max Frequency f=10.0 Hz, δ=20%*[YqLexact(t)]max

Exact solution obtained 1200
280 Exact solution obtained
from Eqs.(2), (3), (40) and (41) from Eqs.(2), (3), (40) and (41)

surface as by-product (K)

at front surface (W/cm ) 260 Recovered value using CGM method 1000

Temperature at front
Recovered value using CGM method
2
Recovered heat flux

2
240 RMS deviation = 3.515 W/cm RMS deviation = 0.191 K
2
800 Maximum deviation = 0.636 K
Maximum deviation = 13.426 W/cm
220
600
200
400
180

160 200
0.0 0.3 0.6 0.9 1.2 1.5 0.0 0.3 0.6 0.9 1.2 1.5
Time (s) Time (s)
(a) (b)

Frequency f=20.0 Hz, δ=20%*[YqLexact(t)]max Frequency f=20.0 Hz, δ=20%*[YqLexact(t)]max

280 1000
Exact solution obtained Exact solution obtained
270 900

surface as by-product (K)

from Eqs.(2), (3), (40) and (41) from Eqs.(2), (3), (40) and (41)
260
at front surface (W/cm )

Temperature at front
2

Recovered value using CGM method 800 Recovered value using CGM method
Recovered heat flux

250
2
240 RMS deviation = 10.245 W/cm 700 RMS deviation = 0.435 K
230 2 Maximum deviation = 1.068 K
Maximum deviation = 25.826 W/cm 600
220
210 500
200 400
190
300
180
170 200
0.0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1.0 0.0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1.0
Time (s) Time (s)
(c) (d)
Figure 13 Inverse results for aluminum alloy 2024-T6 with higher measurement error and the frequency: (a) estimated front-surface heat flux for φ∗ =
20% · [Yq∗Lexact (t)]max and f = 10 Hz; (b) front-surface temperature as a by-product for φ∗ = 20% · [Yq∗Lexact (t)]max and f = 10 Hz; (c) estimated front-surface
heat flux for φ∗ = 20% · [Yq∗Lexact (t)]max and f = 20 Hz; (d) front-surface temperature as a by-product for φ∗ = 20% · [Yq∗Lexact (t)]max and f = 20 Hz.

underscores the necessity of using thermal properties in practi- f = 6 Hz, the estimation of the front-surface heat flux severely
cal applications. degrades (Figure 11a), but the recovering accuracy of the front-
As mentioned earlier, one of the original contributions of surface temperature is still excellent (Figure 11b). When the
present study is using heat flux measurement data in the ob- frequency f is further increased to 10 Hz (Figure 11c and d),
jective function. To show the superiority of using heat flux the temperature estimation is still very good. This indicates
measurements in the objective function, Figure 10 presents the that the inverse formulations developed in this study are
inverse solutions using temperature measurement data in the very robust in recovering the front-surface temperature. This
objective function while the heat flux measurement data are is very important since the temperature, not the heat flux,
employed as the boundary condition. Other simulation parame- will be used as the input information when subsequent ther-
ters are the same as those in Figure 8. It is seen by comparing momechanical analysis is performed to explore the damage
Figures 8 and 10 that there is not much difference between mechanism.
the recovered temperatures obtained using these two methods The foregoing discussions were carried out for stainless steel
(comparing Figures 8b and 10b), but the accuracy in the recov- 304. To demonstrate the applicability of the proposed IHCP
ered heat flux obtained by using temperature measurements is formulation to different materials, the numerical model is tested
subject to severe degradation (comparing Figures 8a and 10a). for another commonly used industrial material, aluminum alloy
The comparison between Figures 8 and 10 clearly indicates that 2024-T6, whose thermal properties [29] are given in Figure 3b.
the inverse approach using heat flux measurements in objec- The measurement error in back-surface heat flux is 5% and the
tive function has advantage over the traditional method using frequency of the periodic heating flux is 5 Hz. Figure 12 shows
temperature measurements in objective function. the inverse results. As seen from Figure 12a, both the phase and
Figure 11 shows the calculated results when the measure- the amplitude of the estimated heat flux agree excellently with
ment error and the frequency are further increased to φ∗ = 10%· those of the exact solutions.
[Yq∗Lexact (t)]max and f = 6 or 10 Hz. It can be seen from Figure Figure 13 shows the inverse results when the measure-
11a and b that for the case where φ∗ = 10%·[Yq∗Lexact (t)]max and ment error and the frequency are further increased to φ∗ =
heat transfer engineering vol. 32 no. 10 2011
872 ZHOU ET AL.

Exact solution obtained Exact solution obtained

250 from Eqs.(2), (3), (40) and (41) 250 from Eqs.(2), (3), (40) and (41)
Recovered value using CGM method Recovered value using CGM method

at front surface (W/cm )

2 2
RMS deviation = 5.321 W/cm

2
RMS deviation = 6.825 W/cm

Recovered heat flux

at front surface (W/cm )
2
Recovered heat flux

2 2
225 Maximum deviation = 13.184 W/cm 225 Maximum deviation = 16.542 W/cm

200 200

175
175
0.0 0.5 1.0 1.5 2.0 2.5 3.0
0.0 0.5 1.0 1.5 2.0 2.5 3.0
Time (s) Time (s)
(a)
(a)
Exact solution obtained
Exact solution obtained from Eqs.(2), (3), (40) and (41)
1100
1100 from Eqs.(2), (3), (40) and (41) Recovered value using CGM method
1000
surface as by-product (K)
1000
Recovered value using CGM method Temperature at front
RMS deviation = 1.309 K
RMS deviation = 1.060 K 900
surface as by-product (K)

900 Maximum deviation = 3.454 K

Maximum deviation = 2.893 K
Temperature at front

800
700
600
500
400
300
0.0 0.5 1.0 1.5 2.0 2.5 3.0
Time (s)
(b)
Figure 14 Influence of uncertainties in thermophysical properties (stainless
steel 304, φ∗ = 5%·[Yq∗Lexact (t)]max , f = 5 Hz): (a) estimated front-surface heat
flux by CGM inverse method; (b) front-surface temperature as a by-product.
20% · [Yq∗Lexact (t)]max and f = 10 or 20 Hz. As is seen, the cal-
culating results display the similar trends as shown in Figure
11, i.e., as the measurement error and the frequency are further
increased, the heat flux estimation is subject to degradation, but
the temperature estimation still maintains a high accuracy. The
calculated results shown in Figure 13 are even better than those
in Figure 11. This indicates that for high-thermal-conductivity
materials, the proposed inverse formulation can handle cases
where the frequency of the input heating flux and the random
measurement errors are even higher.
In the foregoing discussions, the thermophysical properties
are assumed to be exactly known. In reality, the thermophysical
properties may contain uncertainties. Figure 14 examines the
influence of error-containing thermophysical properties on the
accuracy of the inverse solutions. The simulation is performed
for stainless steel 304. The measurement error in back-surface
heat flux is 5% and the frequency of the periodic heating flux
is 5 Hz. The random errors in thermophysical properties (volu-
metric specific heat and thermal conductivity) are taken as 5%,
heat transfer engineering
800
700
600
500
400
300
0.0 0.5 1.0 1.5 2.0 2.5 3.0
Time (s)
(b)
Figure 15 Influence of measurement error in back-surface temperature (stain-
less steel 304, φ∗ = 0.1K , f = 5 Hz): (a) estimated front-surface heat flux by
CGM inverse method; (b) front-surface temperature as a by-product.
which are obtained in a similar way as described in Eq. (42).
As can be seen from Figure 14, both the phase and amplitude of
the estimated heat flux agree excellently with those of the ex-
act solutions. This demonstrates that the numerical formulation
proposed in present study is insensitive to the uncertainties in
thermal properties.
As mentioned earlier, in this study, the front-surface heating
condition is recovered based on the heat flux and temperature
measurements at back surface. So far, we only consider the
random errors in heat flux measurements and the temperature
measurements are assumed to be errorless. It is necessary to
know the effects of the back-surface temperature measurement
data on the accuracy of the inverse solutions. Figure 15 tests
the influence of the random errors in back-surface temperature
measurements. The random error is introduced via Eq. (42), in
which the standard deviation of the measurements is taken as
φ∗ = 0.1 K. As can be seen in Figure 15a, the heat flux estima-
tion is subject to pronounced degradation. But the temperature
prediction still maintains a reasonable accuracy (Figure 15b).
This confirms that the boundary condition choice and inverse
vol. 32 no. 10 2011
 ZHOU ET AL. 873

algorithm formulated in this study are robust in retrieving the q1 (t) dimensionless observed heat flux at front
front-surface temperature. surface
q1∗ (t) observed heat flux at front surface, W/m2
q[L , t; q1 (t)] dimensionless computed heat flux at the back
CONCLUSIONS surface
q[L , t; d k (t)] dimensionless heat flux variation, which is
A conjugate gradient method algorithm is presented to recon- sometimes simplified as q(d k )
struct heat flux and temperature at the front (heated) surface of S dimensionless objective function
a finite slab with temperature-dependent thermophysical prop- ∇ S[q1k (t)] dimensionless gradient direction of objective
erties and under high-intensity periodic heating based on the functional at iteration k
temperature and heat flux measurement data at the back sur- S[q1 (t)] dimensionless objective function variation
face. The inverse problem is formulated in such a way that the t dimensionless time
front-surface heat flux is chosen as the unknown function to t∗ time, s
be recovered, and the front-surface temperature is computed t ∗ time step, s
as a by-product of the IHCP algorithm. New equations for the tc characteristic time, s
sensitivity problems and adjoint problems are derived to incor- tf dimensionless final time
porate heat flux measurement data in the objective function. It is t ∗f final time, s
shown that the constant thermal property assumption may intro- T dimensionless temperature
duce considerable errors in inverse solutions, which underscore T∗ temperature, K
the necessity of using thermal properties in practical applica- Tc characteristic temperature, K
tions. It is also demonstrated that the inverse approach using T0 dimensionless initial temperature
heat flux measurements in the objective function is superior to T0∗ initial temperature, K
the traditional methods in which the temperature measurement T∞ dimensionless ambient temperature
data are used. The methodologies are tested for two commonly T1 (t) dimensionless front surface temperature
used industrial materials, stainless steel 304 and aluminum al- T1∗ (t) front surface temperature, K
loy 2024-T6. The effects of the uncertainties in back-surface T [x, t; q1 (t)] dimensionless temperature variation, which is
temperature and thermal properties on inverse solutions are also sometimes simplified as T
tested. The excellent numerical results demonstrate that the pro- x dimensionless spatial coordinate variable
posed approach is a robust numerical algorithm for IHCP with x∗ spatial coordinate variable, m
temperature-dependent thermophysical properties. Y (t) dimensionless measurement data (temperature
or heat flux) with errors at back surface ob-
NOMENCLATURE tained by numerical simulations
Yexact (t) dimensionless measurement data (temperature
C dimensionless volume-specific heat or heat flux) without errors at back surface ob-
C∗ volume specific heat, J/(m3-K) tained by numerical simulations
d k (t) dimensionless direction of descent at iteration Yq L (t) dimensionless measurement heat flux at the
k back surface
f frequency of periodic laser heat flux at front Yq∗L (t) measurement heat flux at the back surface,
surface, Hz W/m2
h dimensionless convection heat transfer YT L (t) dimensionless measurement temperature at the
coefficient back surface
h∗ convection heat transfer coefficient, W/(m2-K) YT∗ L (t) measurement temperature at the back surface,
k dimensionless thermal conductivity K
k∗ thermal conductivity, W/(m-K)
Greek Symbols
lc characteristic length, m
L dimensionless thickness of one-dimensional βk dimensionless search step size at iteration level
(1-D) slab k
L∗ thickness of 1-D slab, m χ dimensionless tolerance used to stop the CGM
q dimensionless intensity of heating source at iteration procedure
front surface δ Dirac delta function
q∗ intensity of heating source at front surface, ε surface emissivity
W/m2 φ dimensionless standard deviation of heat flux
qc characteristic heat flux, W/m2 or temperature measurements
∗
qconst constant component of the front-surface peri- γk dimensionless conjugate coefficient at iteration
odic heat flux, W/m2 level k
heat transfer engineering vol. 32 no. 10 2011
874 ZHOU ET AL.
λ(x, t) dimensionless Lagrange multiplier
σ a dimensionless quantity related to
Stefan–Boltzmann constant, defined by Eq. (1)
σ∗ Stefan–Boltzmann constant, σ = 5.67 × 10−8
W/(m2-K4)
ω a dimensionless random variable having a nor-
mal distribution with zero mean and unitary
standard deviation
ξ dimensionless perturbed variable
Subscripts
0 initial
f final
q heat flux
T temperature
Superscripts
∗
real physical quantities with dimensions
k iteration level
REFERENCES
[1] dell’Erba, M., Galantucci, L. M., and Miglietta, S., An Ex-
perimental Study on Laser Drilling and Cutting of Com-
posite Materials for the Aerospace Industry Using Excimer
and CO2 Sources, Composites Manufacturing, vol. 3, no.
1, pp. 14–19, 1992.
[2] Beck, J. V., Blackwell, B., and St-Clair, C. R., Inverse Heat
Conduction: Ill Posed Problems, Wiley, New York, 1985.
[3] Alifanov, O. M., Inverse Heat Transfer Problems,
Springer-Verlag, Berlin/Heidelberg, Germany, 1994.
[4] Özisik, M. N., and Orlande, H. R. B., Inverse Heat Trans-
fer: Fundamentals and Applications, Taylor & Francis,
New York, 2000.
[5] Diller, T. E., Advances in Heat Flux Measurements, Ad-
vances in Heat Transfer, vol. 23, pp. 279–368, 1993.
[6] Childs, P. R. N., Greenwood, J. R., and Long, C. A., Heat
Flux Measurement Techniques, Proceedings of the Insti-
tute of Mechanical Engineers. Part C: Journal of Mechan-
ical Engineering Science, vol. 213, no. C7, pp. 655–677,
1999.
[7] Tong, A., Improving the Accuracy of Temperature Mea-
surements, Sensor Review, vol. 21, no. 3, pp. 193–198,
2001.
[8] Childs, P. R. N., Advances in Temperature Measurement,
Advances in Heat Transfer, vol. 36, pp. 111–181, 2002.
[9] Saidi, A., and Kim, J., Heat Flux Sensor with Minimal
Impact on Boundary Conditions, Experimental Thermal
and Fluid Science, vol. 28, pp. 903–908, 2004.
[10] Loulou, T., and Scott, E. P., An Inverse Heat Conduc-
tion Problem With Heat Flux Measurements, International
Journal for Numerical Methods in Engineering, vol. 67, pp.
1587–1616, 2006.
[11] Sparrow, E. M., Haji-Sheikh, A., and Lundgren, T. S.,
The Inverse Problem in Transient Heat Conduction, ASME
Journal of Applied Mechanics, vol. 86, pp. 369–375, 1964.
heat transfer engineering
[12] Alifanov, O. M., and Artyukhin, E. A., Identifica-
tion of Mathematical Models of Transient Heat Trans-
fer Processes, Proceedings of the 4th International
Conference on Numerical Methods in Thermal Prob-
lems, Pineridge Press, Swansea, UK, pp. 771–779,
1985.
[13] Jarny, Y., Özisik, M. N., and Bardon, J. P., A General
Optimization Method Using Adjoint Equation for Solv-
ing Multidimensional Inverse Heat Conduction, Interna-
tional Journal of Heat and Mass Transfer, vol. 34, pp.
2911–2919, 1991.
[14] Pasquetti, R., and Niliot, C. L., Boundary Element Ap-
proach for Inverse Heat Conduction Problems: Applica-
tion to a Bidimensional Transient Numerical Experiment,
Numerical Heat Transfer, Part B, vol. 20, pp. 169–189,
1991.
[15] Yang, C.-Y., and Chen, C.-K., The Boundary Estima-
tion in Two-Dimensional Inverse Heat Conduction Prob-
lems, Journal of Physics D: Applied Physics, vol. 29, pp.
333–339, 1996.
[16] Machado, H. A., and Orlande, H. R. B., Inverse Analysis
for Estimating the Timewise and Spacewise Variation of
the Wall Heat Flux in a Parallel Plate Channel, Interna-
tional Journal of Numerical Methods for Heat and Fluid
Flow, vol. 7, no. 7, pp. 696–710, 1997.
[17] Huang, C.-H, and Wang, S.-P., A Three-Dimensional In-
verse Heat Conduction Problem in Estimating Surface Heat
Flux by Conjugate Gradient Method, International Jour-
nal of Heat and Mass Transfer, vol. 42, pp. 3387–3403,
1999.
[18] Nenarokomov, A. V., Three-Dimensional Boundary In-
verse Heat Conduction Problem for Regular Coordinate
Systems, Inverse Problems in Science and Engineering,
vol. 7, no. 4, pp. 335–362, 1999.
[19] Emery, A. F., Nenarokomov, A. V., and Fadale, T. D., Un-
certainties in Parameter Estimation: The Optimal Exper-
imental Design, International Journal of Heat and Mass
Transfer, vol. 43, pp. 3331–3339, 2000.
[20] Monde, M., Arima, H., and Mitsutake, Y., Estimation of
Surface Temperature and Heat Flux Using Inverse Solution
for One-Dimensional Heat Conduction, ASME Journal of
Heat Transfer, vol. 125, pp. 213–223, 2003.
[21] Mota, C. A. A., Mikhailov, M. D., Orlande, H. R. B., and
Cotta, R. M., Identification of Heat Flux Imposed by an
Oxyacetylene Torch, 10th AIAA/ISSMO Multidisciplinary
Analysis and Optimization Conference, Albany, NY, Au-
gust 29–September 2, 2004.
[22] Xue, X., Luck, R., and Berry, J. T., Comparisons and Im-
provements Concerning the Accuracy and Robustness of
Inverse Heat Conduction Algorithms, Inverse Problems
in Science and Engineering, vol. 13, no. 2, pp. 177–199,
2005.
[23] Abboudi, S., and Artioukhine, E., Parametric Study and
Optimal Algorithm of a Simultaneous Estimation in Two-
Dimensional Inverse Heat Conduction Problem, Inverse
vol. 32 no. 10 2011
 ZHOU ET AL. 875

Problems in Science and Engineering, vol. 16, no. 4, pp. Yuwen Zhang is a professor of mechanical and
461–482, 2008. aerospace engineering at University of Missouri,
Columbia, Missouri. His research interests include
[24] Mota, C. A. A., Orlande, H. R. B., De Carvalho, M. O. M.,
phase change heat transfer, heat pipes, ultrafast, ultra-
Kolehmainen, V., and Kaipio, J. P., Bayesian Estimation intense laser materials processing, and transport phe-
of Temperature-Dependent Thermophysical Properties and nomena in materials processing and manufacturing.
Transient Boundary Heat Flux, Heat Transfer Engineering, He is the author of more than 130 journal papers and
vol. 31, no. 7, pp. 570–580, 2010. over 90 conference papers, as well as two textbooks.
He is a recipient of the 2002 Office of Naval Research
[25] Zhou, J., Zhang, Y., Chen, J. K., and Feng, Z. C., Inverse
(ONR) Young Investigator Award. He is a fellow of
Heat Conduction Using Measured Back Surface Temper- the ASME and associate fellow of the AIAA.
ature and Heat Flux, AIAA Journal of Thermophysics and
Heat Transfer, vol. 24, no. 1, pp. 95–103, 2010.
[26] Patankar, S. V., Numerical Heat Transfer and Fluid Flow, J. K. Chen is the William and Nancy Thompson
Hemisphere, New York, 1980. Professor and Director of the Center of Ultrashort,
[27] Alifanov, O. M., Solution of an Inverse Problem of Heat Ultraintense Lasers at the University of Missouri,
Conduction by Iteration Methods, Journal of Engineering Columbia, Missouri. He received his Ph.D. in aero-
nautics and astronautics from Purdue University in
Physics, vol. 26, pp. 471–476, 1974.
1984. His current research interests include laser ef-
[28] Zhou, J., Zhang, Y., Chen, J. K., and Smith, D. E., A fects and ultrafast thermomechanics. He is a fellow
Nonequilibrium Thermal Model for Rapid Heating and of the ASME.
Pyrolysis of Organic Composites, ASME Journal of Heat
Transfer, vol. 130, p. 064501, 2008.
[29] Incropera, F. P., Dewitt, D. P., Bergman, T. L., and Lavine,
A. S., Fundamentals of Heat and Mass Transfer, 6th ed.,
Z. C. Feng is a professor of mechanical and aerospace
John Wiley & Sons, New York, 2007.
engineering at University of Missouri, Columbia,
Missouri. He has conducted research on design, mod-
Jianhua Zhou is a research assistant professor in the eling, and fabrication of MEMS devices and mi-
Department of Mechanical and Aerospace Engineer- crosensors funded by NSF, NIH, and the U.S. Army
ing at University of Missouri, Columbia, Missouri.
PEO. His works include the design and fabrication of
He received his Ph.D. degree in engineering ther-
microgyroscopes, cell electrochemical sensors, and
mophysics from the Chinese Academy of Sciences,
microfluidic channels for cell sorting. He is a fellow
Beijing, China, in 2002. His current research inter-
of the ASME.
ests include optical and thermal responses in laser-
irradiated biological tissues, inverse heat conduction
problems, and heat transfer and fluid flow in particu-
late systems. He is a senior member of AIAA and a
member of ASME.

heat transfer engineering vol. 32 no. 10 2011

Copyright of Heat Transfer Engineering is the property of Taylor & Francis Ltd and its content may not be
copied or emailed to multiple sites or posted to a listserv without the copyright holder's express written
permission. However, users may print, download, or email articles for individual use.