C. Aviles-Ramos A.

Haji-Sheikh
e-mail: haji@mecad.uta.edu Fellow ASME Department of Mechanical and Aerospace Engineering, The University of Texas at Arlington, Arlington, TX 76019-0023

Exact Solution of Heat Conduction in Composite Materials and Application to Inverse Problems
Calculation of temperature in high-temperature materials is of current interest to engineers, e.g., the aerospace industry encounters cooling problems in aircraft skins during the flight of high-speed air vehicles and in high-Mach-number reentry of spacecraft. In general, numerical techniques are used to deal with conduction in composite materials. This study uses the exact series solution to predict the temperature distribution in a two-layer body: one orthotropic and one isotropic. Often the exact series solution contains an inherent singularity at the surface that makes the computation of the heat flux difficult. This singularity is removed by introducing a differentiable auxiliary function that satisfies the nonhomogeneous boundary conditions. Finally, an inverse heat conduction technique is used to predict surface temperature and/or heat flux.

J. V. Beck
Department of Mechanical Engineering, Michigan State University, East Lansing, Ml 48824-1226 Fellow ASME

Introduction
The transient and steady-state temperature distribution in composite configurations consisting of several distinct thermally anisotropic subdomains have numerous applications to heat transfer problems in reentry vehicles, air frames, nuclear reactors, etc. The increased use of composite materials in the automotive and aerospace industries has motivated research into experimental techniques and solution methods to determine the thermal properties of these materials. The main purpose of this study is to find the surface heat flux from the knowledge of interior temperature histories when the composite bodies are heated by unknown, or partially known, surface conditions. This problem is solved by finding the exact solution of the direct heat conduction problem and subsequently using it as a model to solve the inverse heat conduction problem. The analytical solutions of heat conduction problems in composite media have been investigated by several authors. Tittle (1965) developed a one-dimensional orthogonal expansion for a composite medium while Padovan (1974) surveyed the earlier work and developed a generalized Sturm-Liouville procedure for composite and anisotropic domains in transient heat conduction problems. Ozisik (1980, Chapter 14) discussed the application of the integral transform technique for the solution of heat conduction problems in composite media. Salt (1983) investigated the transient temperature solution in a two-dimensional isotropic-composite slab which is coupled to its environment through the surface of one layer only, after the temperature of the environment has suffered a step change. Mikhailov and Ozisik (1986) analyzed the three-dimensional version of the problem considered by Salt (1983). More recently, Yan et al. (1993) published exact series solutions for three-dimensional temperature distributions in two-layer bodies subject to various types of boundary conditions. The specimen under consideration is composed of one carbon matrix-carbon fiber material (orthotropic) and a layer of mica (isotropic). Experimental measurements made on this specimen

were reported by Dowding et al. (1996). The objective is to find the exact solution of the direct heat conduction problem, and to use it with the published thermal properties and temperature measurements to determine the spatially variable and timedependent heat flux at the surface of the specimen. The emphasis of this work is to develop an alternative Green's function solution method that includes the quasi-steady-state solution as an auxiliary function (Beck et al., 1992). The quasi-steady-state solution should satisfy all homogeneous and nonhomogeneous boundary conditions of the temperature problem. For application to inverse heat conduction, the alternative Green's function solution leads to a well-behaved integral equation of the second kind. In general, problems emerge in heat transfer where heat flux and /or temperature values at a heated or cooled surface are needed and it is often physically difficult to place sensors at the surface to measure surface temperature and heat flux. The study of heat transfer in a high-temperature environment underlines this problem. One can place temperature sensors in the material domain and use the response of the sensors to predict both surface temperature and surface heat flux. In theory, using minimization principles, it is possible to calculate thermophysical properties in addition to surface temperature and/or heat flux by this method. The specific case studied here deals with heat conduction in a multidimensional two-layer body, one orthotropic and one isotropic, subject to boundary conditions of the second kind. First, transient and quasi-steady-state solutions are discussed. Next, the application of this exact solution to compute the surface heat flux is presented and the physical model employed is similar to that used in Dowding et al. (1996). The boundary conditions of the second kind make the exact solution a formidable problem. The completeness and orthogonality of the eigenfunctions are two important topics discussed here.

Analysis
Before discussing the inverse formulation, the exact series solution for the temperature distribution in a two-layer body, Fig. 1, will be presented. The orthotropic layer shown in Fig. 1 is carbon-carbon composite and the isotropic layer is mica. The diffusion equation in the carbon-carbon layer, region 1, is Transactions of the ASME

Contributed by the Heat Transfer Division for publication in the JOURNAL OF HEAT TRANSFER and presented at '97 IMECE, Dallas. Manuscript received by the Heat Transfer Division, Aug. 18, 1997; revision received, Apr. 16, 1998. Keywords: Analytical, Composites, Conduction, Heat Transfer, Inverse, Properties. Associate Technical Editor: A. Lavine.

592 / Vol. 120, AUGUST 1998

Copyright 1998 by ASME

Downloaded From: http://electronicpackaging.asmedigitalcollection.asme.org/ on 03/08/2014 Terms of Use: http://asme.org/terms

y,,

I 8T2 k2{

V dy

?<X 0-

(7)

mww

Isotropic Region 2
'jr^1 ; r ^

I Orthotropic Region 1 x=a

A

The initial temperatures in regions 1 and 2 are designated as Tx{x, y, 0) and T2(x, y, 0 ) . Using a standard transformation for orthotropic bodies y = y* (8)

Fig. 1 Schematic of a two-layer body: an orthotropic layer and an isotropic layer

klx a 2 + k'y ~^T dy2

<92r,

Q "2 j

r 1

"IT

Si(x, y, t) = p,cp} -^ . dt

(1)

and a methodology based on the separation of variables similar to the one discussed by Yan et al. (1993), both solutions for Ti and T2 have the common forms, T,(x,y,t) = X X A,(t) cos (nirx/a) cos (ymkry) e x p ( - \ L 0 (9) and T2(x, y, t) = X 2 Amn{t) cos (nirx/a)[Bm sin (rjmny) + Cm cos (r)my)] exp(-\lmt), (10)

Now, consider an isotropic mica layer in perfect thermal contact with a carbon-carbon layer. The diffusion equation in the mica layer, region 2, is , 82T2 , d2T2 , dT2 h ~r-Y 2 + k2 T + g2(x, y, t) = p2c2 ox dy dt (2)

where g\(x, y, t) and g2{x, y, t) are volumetric heat source functions that could include the surface heat flux and a contribution from the quasi-steady-state temperature. Equations (1) and (2) are subjected to the following homogeneous boundary and interface conditions: dTt '- = 0 at x = 0 ox and x = a for / = 1 or 2 (3)

where /c,. = 4klxlkXy and -y, and r;,,,,, are related to \, by the equations
7i

= ^iJalx

(nn/a) (nixla)2

(11) (12)

BT -7T = Q at y = 0 ay

(4)

Vm = v \ | / a 2 -

Tl(x,b)

= T2(x,b).

(6)

The nonhomogeneous boundary condition at y = c is

in which alx = kulpicpi and a 2 = k2lp2cp2. Notice that for y,m and ?7m to be real, X, must be equal to or larger than the largest of the quantities [(nir/a)ia^x, (nitIa){a2]. It is noted that ym and r\mn can be real or imaginary depending on the values of alx, a2, n, and m; further discussion concerning ym and r?, will appear later. The contact conditions at y = b yield the values of B,m and C, as

Nomenclature
a, b, c, d = dimensions in Fig. 1, cm Gij = Green's function A, B, C = constants H(t) =Heaviside function a = a/fe i,j - indices A, = Fourier coefficients, Eq. k\x = thermal conductivity in region 1 (28) along x, W/cmK B, = coefficient, Eq. (13) kXy = thermal conductivity in region 1 c = clb along y, W/cmK Cmn coefficients, Eq. (14) k2 = thermal conductivity in region 2, cpi specific heat in region i = 1 W/cmK or 2, J/kgK *,= (*!,/*!,) "2 dj_ = coefficients in Eq. (47) / = index to designate thermocouple d = d/a number D, E = constants m, n = indices in eigenfunctions D = vector of coefficients Eq. M = number of temperature measure(49) ments D = coefficient, Eq. (38) N = number of unknowns minus 1 E = coefficient, Eq. (39) N, = norm, Eq. (20) F,mn = eigenfunctions, Eqs. (17), q = input heat flux, W/m 2 (18), i = 1,2 s(x) = spatial function / ( f ) = a function of time Sj = constants g0 = constant related to surface t = time, s heat flux Tj = temperature in regions 1 and 2, K g, = volumetric heat source in Tf = quasi-steady temperature in reregion i, W/cm 3 gions 1 and 2, K gt = constant related to heat Tu = temperature, Eq. (31), K source Journal of Heat Transfer TGii = temperature, Eq. (32), K V = volume, cm3 x, y = coordinates, cm y = yK Y = column vector of measured data Greek au = thermal diffusivity along the xdirection in region 1, cm 2 /s a2 = thermal diffusivity in region 2, cm 2 /s Pn = eigenvalue for x-direction, tin/ a, ctrr 1 6 = Dirac delta function ymn = eigenvalue for y -direction in region 1, cirT1 r\m = eigenvalue for y-direction in region 2, cnT 1 X, = eigenvalue for time, s~' A, = initial condition constant 9X, 62 = solutions of Laplace equation p{, p2 = density of regions 1 and 2, kg/ cm3 2 a = variance i// = notation, Eq. (40) * = matrix, Eq. (49) AUGUST 1998, Vol. 120 / 593

Downloaded From: http://electronicpackaging.asmedigitalcollection.asme.org/ on 03/08/2014 Terms of Use: http://asme.org/terms

Bmn = cos ( y

sin (Vmb) sin (ymnk,b) cos (7? (13)

^ '= ="

^ ^

y)

dA^

- (ym/Vmn)(JkJc^/k2) Cm = cos (y,M cos ( W )

= &(*, y> OF^x, sin ( y m ^ ) sin (Vmnb).

y). (26)

(14) Integrating Eqs. (25) and (26) over regions 1 and12, respectively, adding the resulting equations, and applying the orthogoA homogeneous boundary condition at v = c yields a relation nality condition, Eq. (19), and Eq. (20), one gets tan (r)mc) = BmnICm that reduces to tan [rjmn(c - b)] - ( W ^ X V W f c ) tan (7mU) (16)
+

+ ( W ^ x C V f c )

(15) ^ dt

\ r t> i>b i f Jo f Jo g,(x,y, N

mn

t)Fhm(x,y)dydx

I *><*' >'

?)F2 (

- *<

y)dydx

exp(>Lt).

(27)

for finding the eigenvalues. In a numerical scheme to calculate the eigenvalues, the parameters ym and T]m in Eqs. (13), (14), Integrating Eq. (27) yields the value of the Fourier coefficients and (15) were replaced by km using Eqs. (11) and (12) and the roots of Eq. (15) are obtained numerically. , . . 1 f \ f f6 , ,,P , , w J ^ Am = A, +- J J J giU.y, t)FUm(x,y)dydx Orthogonality. In the subsequent formulations, the follow"" T=0 L J o J ing shorthand notation will identify the eigenfunctions: FXmn{x, y) = cos (rawx/a) cos (ymnkry) (17) + g2(x,y, t)F2im(x,y)dydx exp(\2mr)dT.
Jo Jb

and F2mn{x, y) = cos (mrx/a)[Bmn sin (rjmy) + Cm cos {r\mny)], ,. A study of orthogonality shows that ( ^ - Xi,)[/o I P^F^F^dydx (28) (18) The integration constant, Am, is calculated by substituting Eq. (28) into Eqs. (9) and (10) and applying the initial conditions t 0 0btajn (19) ^ ^ y t 0 ) = l z KmnFimn{Xt y) for t = 1; 2 ( 2 9 )
n=0 m=0

IT
M = "

PiCp2F%mnF%ja/dydx

- 0.

According to Eq. (19), the term in the square bracket is equal to zero p = ni m and v =n ; therefore, ciu except cAt^cpi when wiicii fj, aiiu is n, uicit/iuic, the uic norm iiuiiii is is = f
'

. _ . 2. . . . PlcpX(Fx,mn) dydx

P ' multiplying Eq. (29) by piCpiFi:liU, integrating the resulting equations over the z'th region, adding them, and applying the orthogonality relation, Eq. (19), yields the constant "-m, i rr /* / " > -> a -^ PiCpiFUmn(x', y')Ti(x', y', 0)dy'dx' . . . . . ""' M ^ L J o J o + p2cp2F2,mn(x',
Jo Jb

r eacrl

^0 */0

P2Cp2(F2,m)2dydx.
Jo Jb

(20)

y')T2(x',

y', 0)dy'dx'

Fourier Coefficients and Temperature Solutions. The solution is obtained by substituting Eqs. (9) and (10) into Eqs. (1) and (2) and making use of the two-dimensional eigenvalue problem dx2 Fimn ox dy3 9 F%mn dy ,~, .. klx Kmn k2 '''""

(30) Th e final solutions are obtained by inserting A, from Eq. (30) in Eq. (28) and then substituting Amn into Eqs. (9) and (10). The resulting equations satisfy the homogeneous boundary conditions over all surfaces and each equation has two parts. The first part is the fundamental solution that provides the contribut i o n 0 f the initial condition
y

for m a 0 and n > 0, to obtain llFl.mn(x,y)~exp(-XLt)=fgi(x,y,t) I X Fi(x,y)^S!exp(-XLO = 72ft(Jic.3'. 0(23)

/.' - Z< ZJ
= 0 m =o

~
""

exp(.-A.mf n i

L,o 'o

p!Cpl

IJV'

(24)

+ f

f p2cp2F2,mn(x',

y')T2(x',

y', 0)dy'dx

Both sides of Eqs. (23) and (24) are multiplied by PiCpiFUli(x,y) and p2cp2F2,lll/(x, y), respectively, to obtain V V c vF . >^Amn 2 PiCpi L L Fiimn(x, y)t\,^KX, y) ^ exp(-X m f)
=0 n=0

f r = 1 or 2.

(31)

= gi(x, y, t)FltlJI/(x, y) 594 / Vol. 120, AUGUST 1998

The second part accounts for the effect of the volumetric heat source while the initial temperature is zero; the heat flux at the wall multiplied by the Dirac delta function will be viewed as a (25) volumetric heat source, Transactions of the ASME

Downloaded From: http://electronicpackaging.asmedigitalcollection.asme.org/ on 03/08/2014 Terms of Use: http://asme.org/terms

TGJ = Fi'"'"(X' y) f exp[-\L(f - r)]

,,=(1 m=n "mi JT=0

Jo

gi(x', y', T)FUnn(x',

y')dy'dx'

n
<l(s lib

g2(x',y',T)F2,m(x',y')dy'dx' 2. (32)

dr;

for

( = 1 or

The final temperature solution is the sum of these two solutions T,(x,y,t) = T,j(x,y,t) for + 2 TGJ(x,y,t); (33)

/ = 1 or

that includes the effect of initial condition and internal volumetric heat source. Equations (1) through (33) are for two-dimen- Fig. 2 Three-dimensional representation of the quasi-steady-state solusional bodies and they can readily be modified for three-dimen- tion, r*(x, y, 0)/q sional bodies. A two-dimensional analysis is employed for convenience of the presentation and to facilitate a comparison with the experimental data of Dowding et al. (1996). * g + S^ ~ fo)2 + Sibjy - b) For a solid illustrated in Fig. 1, a constant heat flux, q, is T = t applied to the surface, y = c, within 0 & x < d, at / = 0 and then cut off at time ts. This condition is accounted for by 2 ~ q sin (nnd/a) c o s , W7rxy, s c o s n, , , , s considering Eq. (32) having the volumetric heat sources gi(x, + ( ") (nirkryla) a Zd ~T~, ; 2 =i k2(mr/a) ip y,t) = 0 and g2(x, y, t) = q6(y - c)[H(t) - H(t - t.)] where <5(y - c) is the Dirac delta function, and H(t) - H(t (36) ts) is the Heaviside function. Furthermore, Eq. (31) provides the contribution of the initial temperature. Because of an inherent and singularity at the surface, Eq. (32) converges slowly to the * _ _ . , giiy - b)2 gibjy - b) solution. This can be improved by introducing an auxiliary dif- T ' 2 - go* + 7T, + ; ferentiable function that satisfies both homogeneous and nonho2k2 k2 mogeneous boundary conditions, such as the quasi-steady-state ,'2\ ^ q sin (mrd/a) , , s solution. + I _ I ZJ ~TZ 7~7T, c o s (nirx/a) / ,,= i k2(mr/a)2il/ Quasi-Steady-State Solution. Under quasi-steady-state conditions, the temperature throughout the material domain changes X [D sinh (niry/a) + E cosh (niry/a)] (37) linearly with time if the heat flux at the wall, q, is independent of time, i.e., dT*/dt = g0. Using the energy balance, the value where T* = T*(x,y, t) and T* = Tf(x, y, t). The contact conditions at y = b and boundary condition at y = c result in of go is the relations q(x)dx (acpecpe) go = Ai = (ykixkiy/k2) sinh (mrkrb/a) cosh (nwb/a)

f
Jo

- cosh (nirk,b/a) sinh (nnb/a) E = cosh (nirkrb/a) cosh (rnrbla) iy/k2) sinh (nirk,b/a) sinh (mrb/a) < ] > = D cosh (nnc/a) + E sinh (nixcla).

(38)

[p\Cpib + p2cp2(c b)]/c. When dTfldt where pe g0, the homogeneous versions of Eqs. (1) and (2), gt = 0, reduce to the following Poisson's equations: d Tf d T? + k\ dy2 dx2
2
2

(39) (40)

piCpii

(34)

When gi = gpxcpX, g2 = gp2cp2, the standard transformations Tf = 9, + g,y2/2fc,y + Sty + S2 and T* = 82 + g2y2/2k2 + S3y + S4 reduce the above Poisson's equations into two Laplace equations for 8{ and 02. Because of boundary conditions of the second kind at y = 0 and y = c, either S2 or S4 is an arbitrary constant. The insulated boundary at y = 0 requires Si = 0 and to simplify the algebra, the arbitrary constant is S2 = -gib2/ 2kiy. The contact condition at y = b yields S3 = gib/k2 g2bl k2 and S4 = g2b2/2k2 - g,b2/k2; the transformations reduce to Tf(x, y, t) = 6i+ gtlptcIK + gi(y - b)2!2kiy + gMy - b)l kiy and T*(x, y, t) = 82 + gBtlpcpe + g2(y - b)2/2k2 + gi(y - b)/k2. Notice that the transformations are augmented by the term g"f/pecpe. The substitutions of T* in Eq. (34) and T* in Eq. (35) yield two Laplace equations and the solutions for 8{ and 62 result in the final equations for T* and T*, Journal of Heat Transfer

d Tt d T$ dx + k dy2 2

PlCp2go-

(35)

In the numerical computations, it is assumed that heat is entering the system and therefore q is negative. A three-dimensional plot of T*, when t = 0, is presented in Fig. 2 that indicates a satisfactory behavior of the quasi-steady-state solution. Equations (36) and (37) give the quasi-steady-state solution within a constant. This constant can be selected so that the temperature is the correct transient temperature for the transient problem after the initial transients have disappeared and it is found by making the volumetric heat-capacity-weighted temperature equal to zero at t = 0. The auxiliary functions Tf and T* become the true quasi-steady-state temperature when the constant gib3
3ci,

g2(c - b)3 6ck2

gMc - b)2 2ck2

(41)

is added to Eqs. (36) and (37). This constant has no influence on the value of Tt and T2 since it will reduce the coefficient A00 by exactly the same constant. AUGUST 1998, Vol. 120 / 595

Downloaded From: http://electronicpackaging.asmedigitalcollection.asme.org/ on 03/08/2014 Terms of Use: http://asme.org/terms

17 15

,' -

i'i

'i-r-r

i i-i

i i i i i i i i i i i i i .

Table 1 A sample of convergence for (T - Tt)/q within 0.1 percent for (0.889, 0.9144) location in Fig. 4
Eq. (43) Eq. (33) Time, seconds 6 12 24 36 60 T - T C Exact 5.1525 8.2792 11.737 14.208 7.4233

/ * /
>7 5 r-

Temp. 5.1566 8.2822 11.737 14.208 7.4231

m
11 3 3 3 3

Temp. 5.1551 8.2783 11.745 14.196 7.4231

A't
46 34 37 33 3

~ Equation (42) Equation (32) Experimental Data

t N is the number of terms for x and y-directions.

:
i,

1 1 1 1 1 1 1 1

<

20

40

60

80

100

120

Time, Seconds
Fig. 3 A comparison of two temperature solutions at (0.889, 0.9144) with the experimental data, Dowding et al. (1996)

Now, the superposition of two solutions, Eq. (31) and Eq. (36) or (37) for ;' = 1 or 2 yields the transient temperature solution Ti(x, y, t) = T*(x, y, t) + Tu(x, y, t). (42)

The superposition of solutions requires using a new initial condition T,(x, y, 0) - Tf(x, y, 0) instead of T,(x, y, 0) in Eq. (31). Solution Verification. Equations (42) and (32) are employed to solve for the effect of surface heat flux. To check the validity of these solutions, the following numerical values that are consistent with the experimental information, Dowding et al. (1996), are selected: a = 7.62 cm, b = 0.9144 cm, c = 0.958 cm, d = 0.31a cm, ku = 0.593 W/cmK, kXy = 0.0484 W/cmK, piCpl = 2.509 J/cm 3 K, k2 = 0.001 W/cm, and p2cp2 = 2 J/cm 3 K. It was noted that, for the above parameters, the imaginary values of y, would significantly influence the temperature solution and a numerical study confirms this. For a solid shown in Fig. 1, Fig. 3 shows the effect of imaginary y,m values when using data obtained from Eqs. (42) and (32). The input heat flux is q 1 W/cm 2 from t = 0 to 40 seconds and the solid is initially at a uniform temperature. Although the solid line, Eq. (42), and the dash line, Eq. (32), in Fig. 3 should be identical, the results are quite different and, in fact, unacceptable. A comparison with the experimental data of Dowding et al. (1996), for the referenced location, confirmed that neither solution is correct. A study of the basis functions revealed that this set is

' i ' i "

'

'

''-i>

I ' i ' i ' i ' .

not complete since the solution excluded \mn values that will yield imaginary values for y, when n > 0. An imaginary ymn value makes cos (ymnkry) in Eqs. (13) and (17) to become cosh (\ymn\kry), and cos (ymnk,b) and ym sin (ymnk,b) in Eqs. (13) and (14) to become cosh (\ym\hb) and ~|y,| sinh (\ym\krb). Figure 4 is prepared to show the solutions using Eqs. (42) and (32) when all eigenvalues are retained. The temperature at three surface locations with (x, y) coordinates of (1.270, 0), (0.889, 0.9144), and (1.905, 0.9144) and for q =1 W/cm 2 is calculated using Eqs. (42) and (32) while retaining the contribution of imaginary ymn. The calculated temperature, (T - T,)/ q, is plotted in Fig. 4. The results show excellent agreement except for extremely small time. The difference in these two solutions is in the fifth significant figure when using 65 terms for both m and n. Typically, the solution that uses T* converges faster at large time. The computed temperature converged using a smaller number of terms. For a typical set of data, Table 1 shows the number of terms needed for Eq. (42) to provide a temperature solution at x = 0.889 cm and y = 0.9144 cm within 0.1 percent of the exact solution. For comparison, Table 1 includes similar data using Eq. (32). For time larger than six seconds, only three terms for x and y directions will provide accurate results. For the same accuracy a solution that uses Eq. (32) needs significantly more terms. A numerical study of the surface heat flux shows the reason for better convergence of Eq. (42). Unlike the temperature solution given by Eq. (32), Eq. (42) is free from singularities when calculating the surface heat flux. The value of the heat flux, when using Eq. (32), becomes equal to zero over the entire y = c surface whereas Eq. (42) provides a surface heat flux equal to that of Eq. (37). Accordingly, the surface heat flux is calculated by differentiating T* in Eq. (37) with respect to y and surface heat flux for q = - 1 W/cm2 is plotted in Fig. 5. Note that the summation in Eq. (37) contains n 2 in the denominator but following differentiation, the heat flux solution will have n in the denominator. Therefore, the convergence of the heat flux solution is slower than the convergence of the temperature solution. When 50 terms are used, the surface heat

15 13 11 9 7 5 3 1 -1

(0.889,0.9144)

: :

\
\\ (1.905,0.9144) :

o 0.0
X -0? 3 U. -0.4
*J

0.2

jytyVVwwwwwvA
n = S000 n = 500 n = SO

^ (1.270,0) Solution, Eq. (42) Solution, Eq. (32)

i , i , i , i i . i . i . i . i . i

-0.6

10 20 30 40 50 60 70 80 90 100

: . -

3 -1.2 (0

rfa

a -0.8 o -1.0

wwwv
3 4 5

Time, Seconds
Fig. 4 A comparison of two solutions using Eqs. (42) and (32) Fig. 5

X, cm
Calculated surface heat flux using different number of terms

596 / Vol. 120, AUGUST 1998

Transactions of the ASME

Downloaded From: http://electronicpackaging.asmedigitalcollection.asme.org/ on 03/08/2014 Terms of Use: http://asme.org/terms

Carbon-Carbon Specimen Fig. 6 Specimen with the location of the seven thermocouples, see Dowding et al. (1996)

the domain to solve for the spatially variable and time-dependent surface condition. In general, this solution is found using the Green's function and an alternative Green's function solution (AGFS) formulation, Beck et al. (1992). The AGFS uses an auxiliary temperature function T* that satisfies the homogeneous and nonhomogeneous boundary conditions. The functions, Tf, Eqs. (36) and (37), are constructed by solving an auxiliary quasi-steady-state problem defined using the same boundary conditions and physical properties of the original problem. Using the quasi-steady-state solution, Eq. (31) yields the AGFS solution, T,(x,y,t) = Tf{x,y,t) - i f Gu(x,y,t\x',y',0)Tf(x\y',0)dV; i = 1 or 2 (46)

flux visibly fluctuates about 1 and 0. Using 500 terms, the fluctuations are diminished and limited to the vicinity of x = d. There is no visible fluctuation when 5000 terms are used. Green's Function, One can generalize Eqs. (31) and (32) by expressing the solution in terms of the Green's function. Consider a single source gj{x, y, r ) = pjCPJ6(x - x*)6(y - y*)6(t - T * ) at (x*, y*) in region; = 1 ovj = 2. Substitution of a source term gi(x', y', T) or g2(x', y', r ) in Eq. (32) followed by the integrations using the properties of the Dirac delta function (Beck et al., 1992) yields the Green's function Gjj(x, y, t\x*, y*, r * ) . After replacing**, y*, and r* by x', y', and T, respectively, one obtains G^x^, t\x',y',r)

ft^-^'^^^^')exp[-XL(r-r)]
(43)

where i = 1 or 2; and j = 1 and 2. Equations (31) and (32), in terms of the Green's function, become Tij(x,y, t) = 1 f Gli{x,y,t\x',y,,Q)Tj(.x',y',0)dV = 'L
dr

(44)

for this inverse heat conduction study. In general, the value of temperature at a location designated by (x, y) and at time, t, is known and the unknown is the heat flux in the definition of Tf(x, y, t). This formulation permits the heat flux to be a continuous or piecewise continuous function of spatial coordinates along the boundary. Also, the heat flux can be a continuous or a discrete function of time. In this inverse heat conduction study, q(x, t) is the unknown to be determined and different methods of computation are discussed in Beck et al. (1996). In practice, one selects a complete set of independent basis functions to describe the variation of q(x, t) and the size of the set depends on the complexity of the heat transfer process, the extent of the heated surface, and the duration of the process. For the experimental data of Dowding et al. (1996), the switchon and switch-off process is adequately described by using the Heaviside function. It is assumed that the heat flux at the surface y = c has the form q = s(x)f(t) where s(x) = (d0 + d,x + d2x2 + . . .)[H(x) - H{x - d)),f(t) = H(t) - H(t - / , ) , the parameter d describes the size of the heated zone, Fig. 1, and ts is the time when the power to the heater is switched off. These two Heaviside functions describe the abrupt changes in the spatial and temporal variation of the surface heat flux. Following appropriate substitutions, Eq. (46) reduces to a system of linear equations, each for a given set of {x, y; t}.

Matrix Representation. For simplicity in this presentation, let (xh yt) designate the location of thermocouple /, and tv be the time at which each temperature measurement is taken. Since X gj(x',y\T)dV (45) all thermocouples were placed in the carbon-carbon composite, the left side of Eq. (46), Ti(xh y,, tp), becomes a known quanwhere dV = 1 dx'dy' and i takes the value of 1 or 2. Often it is convenient to express the variables in dimen- tity. The measured temperature is denoted as Yx(xh yh tp) and sionless form. For instance, if b is the characteristic length, it is an approximation to the temperature Ti(xi, yi, tp). By then, in dimensionless form, x = xlb, y = ylb, c c/b, a = substituting q in Eqs. (36) and (37) to obtain T* and T* for alb, d ~ dlb, y,m = y,mb, rj, = r]mnb, and /? = f3b = {nn/ inclusion in Eq. (46), the right side of Eq. (46) becomes d0<&o + rf,$i + d2$2 + + dN$N. Physically, ^represents tempera)b. Accordingly, \2m reduces to \2m = [(ym)2 + (xh y,) and time t if the surface heat flux (0)2]ajb2 in region 1 and to \2nn = [(r},,,)2 + (P)2]a2/b2 ature at the location is equal to xJf(t) for j = 0, 1, 2, . . . , N and there is no in region 2. The exact series solution obtained in this manner exhibited internal heat source. In practical applications, the temperature no peculiarities. However, computation of the eigenvalues using measurements Y{(x,, y,, tp) are known while the % values are Eqs. (15) and (16) needs special attention since numerical obtained numerically. The least-squares estimate for the values truncation can produce false eigenvalues when some y, values of dj is the one that makes the sum of squares for all seven thermocouples are imaginary and n is large. TGJ(x,y,t)
j = l PiCPJ ^ = 0

G,j(x,y,t\x',y',T)

JVj

Inverse Heat Conduction The extension of the direct solution to inverse heat conduction is the final objective of this study. The inputs to the inverse problem are the experimentally obtained temperature data in Dowding et al. (1996). The calculation of the heat flux at the surface is carried out in the two-layer domain, Fig. 6. The computed heat flux is compared with available experimental information. An inverse technique uses temperature data within Journal of Heat Transfer

X I l S dj$j(x y t) - dx y tp)]2
(=1 p=\ j=0

(47)

a minimum and this quantity is a minimum if

a 1 M N

- l I S H

dj$j(x y tp) - FiU, y t)]2} = 0,

for i = 0, 1, . . . , N (48)

AUGUST 1998, Vol. 120 / 597

Downloaded From: http://electronicpackaging.asmedigitalcollection.asme.org/ on 03/08/2014 Terms of Use: http://asme.org/terms

1 M M

320

II

TC No. 1 TC No. 2 TC No. 3 TC No. 4 T TC No. 5 O TC No. 6 o TC No. 7

A V

300 -

290

I . I . I . I . I . I . I 20 40 60

80

I 100

120

Time, Seconds
Fig. 7 A comparison of the calculated and measured temperatures at seven thermocouple locations using a spatially constant heat flux

flux of 1.730 W/cm 2 reported by Dowding et al. (1996). The temperature variance for this case is a2 = 0.078. The temperatures at the seven thermocouple locations are calculated using the computed surface heat flux and the results are shown in Fig. 7. The variation of the heat flux in the ^-direction on the y = c surface is investigated assuming an ^-dependent heat flux of the form q(x, t) = (d0 + d2x2)[H(x) - H(x - d)][H(t) H(t - ts)]. The constants d0 and d2 are computed using the thermocouple locations shown in Fig. 6. A comparison between experimental and calculated data is in Fig. 8. The constants and the variance for this case are d0 = 1.633 W/cm 2 , d2 = 0.0393 W/cm 4 , and a2 = 0.097. This correction slightly improved the agreement between calculated and measured temperatures for thermocouple number 4 in Fig. 8. However, the variation in the heat flux is of the order of the data noise and is too small to be significant. The temperature variance for this case is slightly higher than the variance of the constant heat flux case.

Conclusion and Remarks

This paper presents a relatively convenient quasi-steady solution for the case of a two-layer material with one of the layers having orthotropic properties. Also, the transient solution for a two-dimensional problem with a surface partially heated is given. As a generalization, the Green's function for this problem is given and can be used for other problems. The numerical computations presented in this paper show that the exact analytical solution serves as a useful tool for solving an interesting class of inverse heat conduction problems. Furthermore, the solution was developed for insulated boundaries but the same methodology applies to other boundary conditions. There are a few differences between the solution discussed earlier and classical solutions for isotropic bodies. First, a product solution (Beck, et al, 1992; p. 92) cannot be used as observed by Tittle (1965); also, the product solution cannot be used to obtain the Green's function. Second, a solution of this type, that uses the separation of variables, is restricted to the boundary conditions of the first or second kind on x = 0 and x = a surfaces. However, there is no restriction on the type of boundary conditions at y = 0 and y = c. Third, contact resistance at the interface can be included in the solution with additional efforts. This exact analysis serves as a powerful tool for solving inverse heat conduction problems and the inverse procedure combined with the exact analysis, described above, produces results with remarkable accuracy. This procedure satisfies all the criteria for the evaluation of the inverse heat conduction

where N + 1 is the number of constants to be determined. Equation (48) reduces to a set of linear equations described by *D = Y (49)

where Y is a column vector of M measured temperature data, D is a column vector that contains N + 1 unknown coefficients, dj, and $ represents a matrix of calculated < T > ; values whose size is M by (N + 1). Using standard least-squares algebra suggested by Eq. (49), *'(*D) = $'Y where $ ' is the transpose of <&. The solution for D is D = (#'*)~1($'Y) (51) (50)

Because of the measurement errors and the approximate nature of the product of $ by D, 3>D does not equal Y. The vector Y - <&D represents the residual vector and the estimated variance, a2, of the measured temperatures is obtained using # 2 = (Y - * D ) ' ( Y - $ D ) / M (52)

where M is the total number of temperature data. In addition to random noise, other parameters, such as imperfection of heat flux modeling and thermophysical properties, will affect a2. Example. The temperature data for this numerical example are those of Dowding et al. (1996). Region 1 is a carboncarbon composite layer and region 2 is a mica layer. Measured temperature data at the seven thermocouple sites shown in Fig. 6 are used during calculations. The location of thermocouples 1 through 7, using (x, y) axes shown in Fig. 6, are (1.270, 0), (6.35, 0 ) , (0.889, 0.9144), (1.905, 0.9144), (3.175, 0.9144), (4.290, 0.9144), (6.731, 0.9144). It is assumed all thermocouples are located on both surfaces of the carbon-carbon composite layer and all units are in centimeters. The measured temperature data are shown in Fig. 7 and the mean variance for the measured temperatures is a2 = 0.0204. This represents an error of 0.2C with 95 percent probability which is smaller than the size of the symbols shown in Figs. 7 and 8. The thermophysical properties mentioned earlier, reported by Dowding et al. (1996), are used in the inverse conduction calculations. The effective thermal conductivity (Dowding et al., 1996), k2, includes the contribution of the contact conductance at the interface. First, it is assumed that the heat flux at y = c has no spatial dependence in the interval 0 s x =s d. Using this assumption, the inverse heat conduction procedure described above produced a surface heat flux of 1.682 W/cm 2 using 50 X 50 eigenvalues. The calculated heat flux differs only 2.8 percent from an experimental heat 598 / Vol. 120, AUGUST 1998

320

I"
I290
I I I

TC TC TC TC T TC 0 TC O TC

A V

No. No. No. No. No. No. No.

1 2 3 4 5 6 7

20

I I I I I I I I I I I 40 60 80 100 120 Time, Seconds

Fig. 8 A comparison of the calculated and measured temperatures at seven thermocouple locations using a spatially variable surface heat flux

Transactions of the ASME

Downloaded From: http://electronicpackaging.asmedigitalcollection.asme.org/ on 03/08/2014 Terms of Use: http://asme.org/terms

problem (IHCP) solution methods outlined in Beck (1979) except for accommodating the temperature-dependent thermophysical properties. The temperature-dependent thermophysical properties need special attention not addressed here. This exact analysis is well suited to design experiments for the estimation of thermal properties as a part of IHCP studies. Also, the exact solution is ideal to calculate the sensitivity coefficients, to serve as a standard for evaluating approximate numerical methods, and to carry out simulations before performing the actual experiments. Acknowledgment The authors wish to acknowledge the support of the Mechanical and Aerospace Engineering Department at the University of Texas at Arlington and the National Science Foundation, Grant No. CTS-9400647.

References
Beck, J. V., 1979, "Criteria for Comparison of Methods of Solution of the Inverse Heat Conduction Problem," Nucl. Eng. Desg. Vol. 53, No. 1, pp. 1 1 22.

Beck, J. V., Blackwell, B., and Haji-Sheikh, A., 1996, "Comparison of Some Inverse Heat Conduction Methods Using Experimental Data," International Journal of Heat and Mass Transfer, Vol. 39, No. 17, pp. 3649-3657. Beck, J. V., Cole, K. D., Haji-Sheikh, A., and Litkouhi, B 1992, Heat Conduction Using Green's Functions, Hemisphere, Washington, DC. Dowding, K. J., Beck, J. V., and Blackwell, B. F., 1996, "Estimation of Directional-Dependent Thermal Properties in a Carbon-Carbon Composite," Int. J. Heat and Mass Transfer, Vol. 39, No. 15, pp. 3157-3164. Mikhailov, M. D., and Ozisik, M. N., 1986, "Transient Conduction in a ThreeDimensional Composite Slab," Int. J. Heat and Mass Transfer, Vol. 29, No. 2, pp. 340-342. Ozisik, M. N., 1993, Heat Conduction, 2nd Ed., John Wiley and Sons, New York. Padovan, J., 1974, "Generalized Sturm-Liouville Procedure for Composite Domain Anisotropic Transient Heat Conduction Problems," AIAA Journal, Vol. 12, No. 8, pp. 1158-1160. Salt, H., 1983, "Transient Conduction in a Two-Dimensional Composite SlabI. Theoretical Development of Temperature Modes," Int. J. Heat and Mass Transfer, Vol. 26, No. 11, pp. 1611-1616. Salt, H., 1983, "Transient Conduction in Two-Dimensional Composite Slab II. Physical Interpretation of Temperature Modes," Int. J. Heat and Mass Transfer, Vol. 26, No. 11, pp. 1617-1623. Tittle, C. W., 1965, "Boundary Value Problems in Composite Media," Journal of Applied Physics, Vol. 36, No. 4, pp. 1486-1488. Yan, Ling, Haji-Sheikh, A., and Beck, J. V., 1993, "Thermal Characteristics of Two-Layered Bodies With Embedded Thin-Film Heat Source," ASME Journal of Electronic Packaging, Vol. 115, No. 3, pp. 276-283.

Journal of Heat Transfer

AUGUST 1998, Vol. 120 / 599

Downloaded From: http://electronicpackaging.asmedigitalcollection.asme.org/ on 03/08/2014 Terms of Use: http://asme.org/terms