You are on page 1of 9

10th AIAA/ASME Joint Thermophysics and Heat Transfer Conference

28 June - 1 July 2010, Chicago, Illinois

AIAA 2010-4784

Design of Embedded Temperature-sensor Heat Flux Gauges


for Bodies with Initial Linear Temperature Gradients

Downloaded by UNIVERSITY OF MARYLAND on August 8, 2015 | http://arc.aiaa.org | DOI: 10.2514/6.2010-4784

Daanish Maqbool1 and Christopher Cadou2


University of Maryland, College Park, MD, 20782

In some investigations of convective heat transfer in wind tunnels, it is necessary to heat


or cool the walls prior to initiating flow in the tunnel. However, the heating or cooling
process creates a thermal gradient in the wall due to some inevitable surface heat loss. This
is problematic because existing methods for inferring gas-wall heat fluxes from wall
temperature-time histories assume (require) that the initial temperature of the wall is
uniform. This paper presents a method for adapting existing techniques for inferring surface
heat flux from temperature-time measurements to situations where the initial wall
temperature is not spatially uniform. This is accomplished by adding a linear term to the
non-dimensional temperature profile used in previous work and showing that the governing
equations and boundary conditions remain unchanged with the new non-dimensionalization.
The solution for the transient temperature distribution in a wall with an initial temperature
gradient undergoing convective heat transfer is also presented. This is useful for designing
heat flux gauges and quantifying potentially significant errors (>5%) that can arise if
thermal gradients in the wall are not accounted for.

Nomenclature
a
bn
f, g, d
h
in
k
M
N
q
t
T
T0
Ti
Ts
T
T
v,
x
x1
X

1
2

=
=
=
=
=
=
=
=
=
=
=
=
=
=
=
=
=
=
=
=
=
=
=
=
=

initial temperature gradient


curve-fit coefficients
generic functions
convective heat transfer coefficient
repeated integral
thermal conductivity
confluent hypergeometric function
number of terms for curve-fit
surface heat flux
time
temperature
initial temperature/initial surface temperature
local initial temperature
surface temperature
temperature at penetration depth
gas temperature
intermediate/temporary temperature variables
distance/depth
sensor location
non-dimensional distance
thermal diffusivity
gamma function
thermal penetration depth
non-dimensional temperatures
non-dimensional time

Graduate Student, Dept. of Aerospace Engineering, 0134 Glenn L. Martin Hall, AIAA student member.
Associate Professor, Dept. of Aerospace Engineering, 3179D Glenn L. Martin Hall, AIAA senior member.
1
American Institute of Aeronautics and Astronautics

Copyright 2010 by Daanish Maqbool and Christopher Cadou. Published by the American Institute of Aeronautics and Astronautics, Inc., with permission.

I. Introduction

Downloaded by UNIVERSITY OF MARYLAND on August 8, 2015 | http://arc.aiaa.org | DOI: 10.2514/6.2010-4784

HE J-2X rocket motor is being developed by Pratt & Whitney/Rocketdyne to power the second stage of
NASAs Ares I launch vehicle. The University of Maryland is supporting the development of the engines film
cooled nozzle extension by conducting fundamental investigations of film cooling effectiveness in supersonic
environments. Part of this work involves measuring wall heat flux in a supersonic wind tunnel. The tunnel is a
blow down type in which air from the room is drawn through a nozzle and into a large vacuum chamber. Since the
recovery temperature of the flow in the test section nearly equals the ambient temperature, the difference between
the recovery and test-section wall temperatures is small and so the resulting heat flux is small and difficult to
measure. It is not practical to raise the temperature of the air entering the tunnel so instead the problem is addressed
by heating the walls above the ambient temperature. While studying the inverse (i.e. film heating) problem is
perfectly acceptable for the purpose of the project which is to develop experimentally validated CFD models for
wall heat transfer, it poses a problem for the wall heat flux measurements because the internally heated wall is no
longer isothermal due to some inevitable surface heat loss before the start of the experiment. The first section of this
paper presents a method for adapting existing techniques for inferring surface heat flux from temperature-time
measurements to situations where an initial temperature gradient is present in the wall. The second section develops
an expression for the temporal evolution of the temperature distribution in a wall with an initial temperature gradient
that is suddenly exposed to a known convective heat transfer process. This is useful in developing heat flux
instrumentation for experiments.
The review article by Diller1 discusses many methods for measuring heat flux across a surface. A common
instrument for measuring heat flux in shock tunnels is the thin-film heat flux gauge which is essentially a fastresponding thermistor that is deposited directly on the test surface. Surface heat flux is inferred from the
temperature-time history of the surface (measured by the thin-film gauge) and a 1-D model for the response of the
surface2. A similar approach uses the temperature-time history of a sensor embedded below the test surface3-9. The
latter approach is advantageous in that it does not disturb the flow over the surface and is easier to fabricate. Various
approaches have been taken to infer heat flux from an internal temperature measurement. All have some numerical
aspect; either curve-fitting to a functional form4-5, 7-8 or numerical integration in a finite-differencing scheme6, 9 and
all require that the initial wall temperature be spatially uniform. The latter constraint is the problem here because
heating the wall introduces a relatively large initial temperature gradient (~200 K/m for a surface temperature 50 K
above ambient). The correction described in the first section of this paper is an extension of the method first
presented by Chen and Thomsen4 and later extended by Chen and Chiou7. It could also be applied to most of the
other curve-fitting based interpretation methods. While the correction is developed in response to the film heating
problem described above, it is useful in any situation where one wishes to make transient heat flux measurements
over initially non-isothermal walls. In the second section of this paper, an expression for the transient temperature
distribution in a wall undergoing convective heat transfer with the aforementioned initial condition is derived. This
is useful for predicting the surface heat flux and the temperature response at a prospective sensor location given an
estimate of the flow conditions.
q(t)

II. Inverse Measurement of Surface


Heat Flux
A. Brief Review: Case of Uniform Wall
Temperature
Chen and Chiou7 considered the situation
illustrated in Fig. 1, where a semi-infinite
wall at an initially uniform temperature (i.e.
T(x,0) = constant) is suddenly exposed to an
unknown unsteady heat flux q(t) at the
surface. The objective is to infer the surface
heat flux from the temperature time-history
of a point x1 units below the surface.
The unsteady heat equation is used to
describe the temperature distribution in the
solid:

Semi-infinite
wall

Temperature x1
Measurement
Location

(A)
x

(B)
T

Figure 1. Problem Schematic. Shown above is arbitrary heat flux


into a semi-infinite wall and the internal temperature measurement
location. Also shown are the two initial temperature distributions
under consideration: (A) uniform initial temperature, and (B) initial
temperature gradient.

2
American Institute of Aeronautics and Astronautics

T
2T
= 2
t
x

(1)

Chen and Chiou7 defined the following non-dimensional variables:

t
x1

x
x1

T T0
T0

(2)

Downloaded by UNIVERSITY OF MARYLAND on August 8, 2015 | http://arc.aiaa.org | DOI: 10.2514/6.2010-4784

where is the thermal diffusivity of the material and T0 is the initial temperature. Using the new variables, the heat
equation and initial conditions become

2
=
X 2

( X ,0) = 0

(1, ) = f ( )

(3)

(, ) = 0

(4)

Using a Laplace transform technique, they showed that if the temperature-time history can be fitted to the
following function:
N
1
n
f ( ) = bn (4 ) (n + 1)i 2 n erfc

n =1
2

(5)

(where in represents a repeated integral) then the coefficients, bn (which are determined using a curve-fitting routine),
can be used to determine the non-dimensional surface temperature and temperature gradient:
N

(0, ) = bn n

(6)

n =1

(0, ) N
(n + 1)
n 1
= bn 2
X
( n + 1 )
n=1
2

(7)

The dimensional surface temperature and heat flux are then given by

Ts = T0 (1 + (0, ))

(8)

kT0 (0, )
x1 X

(9)

q=
where k is the thermal conductivity of the material.

B. Case of Initial Wall Temperature Gradient


Now, assume that the initial wall temperature profile is linear. The temperature distribution takes the following
form:

T ( x,0) = T0 + ax
where T0 is the initial surface temperature and a is a constant. Now let
3
American Institute of Aeronautics and Astronautics

(10)

T (T0 + ax)
T0

(11)

also satisfies the governing equation for unsteady heat conduction. Hence,

2
=
X 2

(12)

The initial conditions for the problem remain

Downloaded by UNIVERSITY OF MARYLAND on August 8, 2015 | http://arc.aiaa.org | DOI: 10.2514/6.2010-4784

( x,0) = 0

(1, t ) = f (t )

(, t ) = 0

(13)

Since the governing equation, initial conditions, and boundary conditions transform to the same problem studied by
Chen and Chiou7, their solution can be used directly by substituting for . The differences are that must now be
curve-fitted instead of and that a correction term must be added when solving for the dimensional value of the
surface heat flux. The surface temperature and heat flux in dimensional terms are given by

Ts = T0 (1 + (0, ))
q=

(14)

kT0 (0, )
ak
x1
X

(15)

To conclude, the procedure for determining the surface heat flux associated with a measured subsurface
temperature-time history when the wall is not initially isothermal is as follows:
1. Convert the temperature-time history into - history where the value for x in is x1, the sensor distance.
2. Curve-fit the - history to the form suggested by Chen and Chiou7 (Eq. 5).
3. Use the coefficients, bn, to determine and d/dX at the surface using Eqs. (6) and (7).
4. Recover the dimensional quantities using Eqs. (14) and (15).
C. A Note on Implementation
In order to implement the aforementioned technique, a numerical curve-fitting (optimization) routine must fit the
temperature-time data from the sensor to the form given by Eq. (5). This involves the repeated integral of the error
function. The repeated integral of an arbitrary function, g(x), is defined as:
x un 1

u1

0 0

i g ( x) =
n

... g (u)du...du

n 2

dun1

(16)

Evaluating and incorporating Eq. (16) into an iterative routine is both challenging and impractical from a
computational perspective because it requires the numerical evaluation of several nested integrals in each iteration.
The expression can be reduced to a single integral using Cauchys formula but there is an even more convenient
form for the particular repeated integral in Eq. (5); the repeated integral of the error function can be written
explicitly in terms of the confluent hypergeometric function10, M, for which explicit numerical routines are
commonly available:

2
1
z
3
n
n +1 1 2
i nerf ( z ) = e z
M
, ,z
M + 1, , z 2
2
2n n + 1 2 2 2n 1 n + 1 2

2
2
4
American Institute of Aeronautics and Astronautics

(17)

While use of the confluent hypergeometric function significantly reduces the complexity of the code and
computational time, it is still impractical for a personal computer to calculate Eq. (5) at each iteration of the
optimization routine. To further increase the efficiency of the calculation, note that
N

f ( ) = bn d (n, )

(18)

1
n
d (n, ) = (4 ) (n + 1)i 2 n erfc

(19)

n =1

Downloaded by UNIVERSITY OF MARYLAND on August 8, 2015 | http://arc.aiaa.org | DOI: 10.2514/6.2010-4784

where

This is the computationally expensive term. So, the values of d(n,) can be
calculated once and stored in an array for use by the optimization routine. This
enables the fitting procedure to be performed in an economical manner suitable
for a personal computer.

T
T0

III. Wall Temperature Distribution for Transient Convection


The development of the expression for the temperature distribution in a wall
undergoing transient convection with an initial temperature gradient (Fig. 2)
follows the same basic steps presented by Carslaw & Jaeger11 in their solution for
the temperature distribution in an initially isothermal wall. The basic idea is to
x
transform the problem into one with a known solution: a semi-infinite solid with
T
an initial non-zero uniform temperature and zero surface temperature. The
Figure 2. Surface convection
transformation is a two step procedure.
The first step is to re-scale the wall temperature in terms of the gas with an initial temperature
temperature (T), the initial gradient (a), and distance, by introducing a new gradient.
variable, v:

v = T (T + ax)

(20)

It can be shown that v satisfies the transient heat conduction equation (Eq. 1):

v
2v
= 2
t
x

(21)

The initial temperature distribution is still given by Eq. (10) and the boundary condition is given by an energy
balance at the surface. So, the initial and boundary conditions are

v( x,0) = T0 T

(22)

(0, t ) h
= (0, t ) a
x
k

(23)

It is convenient to introduce another temperature variable, :

=v

k v k
a
h x h

This also satisfies the unsteady heat conduction equation;


5
American Institute of Aeronautics and Astronautics

(24)


2
= 2
t
x

(25)

The initial and boundary conditions become

Downloaded by UNIVERSITY OF MARYLAND on August 8, 2015 | http://arc.aiaa.org | DOI: 10.2514/6.2010-4784

k
h

( x,0) = T0 T a

(26)

(0, t ) = 0

(27)

and the problem has been transformed into one of a semi-infinite plate with zero surface temperature and a uniform
initial temperature distribution. The solution to this problem is given by Carslaw & Jaeger11 and can be used directly
to write
x

( x, t ) =

k
2
T0 T a

2 t

u 2

du

(28)

The definition of , after some rearrangement, is used to solve for v:

v h h

v + + a = 0
x k
k

(29)

This is a 1st order differential equation with the solution

v = e

h x
x
k

h

k

x
h

+ a d + Ce k
k

(30)

The constant C can be determined by noting that the body is semi-infinite. So as x  , v  T0 - T. Since the value
of v is bounded, C must be zero. Simplifying Eq. (30) gives
h


h x
k
v = e k e k d + a
k
h

(31)

The first term corresponds to the convection problem for a uniform initial temperature wall, whereas the second
term is new and arises from the initial temperature gradient. The first integral has been evaluated by Carslaw &
Jaeger11 and can be used directly. Inserting their solution into Eq. (31) gives the expression for v:
hx h t
+
k x
h
x
k

k k2
v = T0 T a erf
+
e
erfc
+
t + a

h 2 t

2 t k
h

(32)

Substituting the definition of v (Eq. 20) to obtain the temperature gives


hx h t
+
k x
h
x
k

k k2
T T = T0 T a erf
+
e
erfc
+

t
+ a + x

h 2 t

2 t k
h

6
American Institute of Aeronautics and Astronautics

(33)

Eq. (33) gives the temporal evolution of the temperature distribution in a semi-infinite solid after its surface is
suddenly exposed to a convective heat transfer process. If a = 0, the solution collapses to the original expression
given by Carslaw & Jaeger11 for a semi-infinite slab with uniform initial temperature.
Finally, it should be noted that when

a=

h(T0 T )
k

(34)

the first term on the right side of Eq. (33) vanishes and the solution is time independent. This is not surprising as it
corresponds to the situation where the conductive heat flow towards the surface equals the heat removed by
convection. In this case, the initial condition is also the steady state condition.

T Ts
= 0.99
Ti Ts

(35)

Inserting Eq. (33) into Eq. (35) gives the following implicit expression for the penetration depth:
h h2t

+

k
erf
+e
2 t

k2

h t

h t
h t
k2
erfc
0
.
01
e
erfc
+

0.99
k
2 t
k
=
2

0.01
k

T0 T a
h

(36)

In general, the effect of a on the penetration depth is weak. However, at the steady state (where a is given by Eq.
34), the right side of the equation tends to infinity which means that tends to zero.

V. Numerical Verification of Results


Figure 2 illustrates the problem that motivated
342
this work: a convectively cooled MACOR plate (i.e.
the wind tunnel wall) with initial Ts = 340 K, T =
340
317.6 K, h = 435 W/m2-K, and a = 173 K/m. The
plate is exposed to these conditions for
338
approximately 6.8 seconds and we are interested in
predicting the temporal response of the temperature
336
distribution in the plate in addition to the heat flux
at the gas-plate interface. While Eq. (33) was
334
Initial Profile
developed for this purpose, it is desired to verify it
numerically. This is accomplished by solving Eq.
Numerical, a = 173 K/m
332
(1) subject to the boundary conditions of the
Eq. (33), a = 173 K/m
experiment described above. The numerical solution
Eq. (33), a = 0 K/m
330
was obtained by solving Eq. (1) using a central
differencing scheme on a uniform grid with
0
2
4
6
8
10
12
elements spaced 0.13 mm apart and with a time step
Location (mm)
of 1.1 ms.
Figure
3.
Temperature
profiles
through a MACOR plate
Figure
3
compares
plate
temperature
after
6.8
seconds.
distributions predicted using Eq. (33) with and
Temperature (K)

Downloaded by UNIVERSITY OF MARYLAND on August 8, 2015 | http://arc.aiaa.org | DOI: 10.2514/6.2010-4784

IV. Thermal Penetration Depth


The preceding solutions are built upon the assumption of a semi-infinite body. In practical situations, however,
the semi-infinite assumption can be satisfied as long as the body is thicker than the thermal penetration depth, .
Schultz & Jones2 defined the penetration depth as the distance at which the scaled local temperature differs from the
surface temperature by 1%, which can be presented as12:

7
American Institute of Aeronautics and Astronautics

without an initial temperature gradient to the results of the numerical simulation with an initial temperature gradient.
The fact that the numerical and analytical (Eq. 33) results overlap indicate that Eq. (33) is a valid solution to the
problem.
Since the principal motivation of this work is to understand the effect of an initial wall temperature gradient on
convective heat flux measurements, Fig. 4 shows heat fluxes predicted using Eq. (33) for various initial wall
temperature gradients. The heat flux decreases with time because the wall temperature decreases with time. The
figure shows that a positive initial wall temperature gradient increases the heat flux from the wall to the flow. This
means that assuming that the wall temperature is uniform when it is not will underestimate heat transfer when a>0
and overestimate heat transfer when a<0. Although this effect is not strong initially, the differences in heat flux
grow and hence become more important with time. Figure 5 shows the error resulting from assuming a uniform
temperature as a function of time for different values of the initial gradient.
6

a=
a=
a=
a=

9000
8000

0 K/m
-150 K/m
+150 K/m
+300 K/m

% Error

Surface Heat Flux (W/m2)

7000

2
0

6000

4000

a = -150 K/m
a = +150 K/m
a = +300 K/m

-2

5000

-4

0
5
10
15
Time (s)
Time (s)
Figure 4. Surface heat flux for different values of the Figure 5. Error in surface heat flux resulting
from assuming a uniform initial temperature.
initial thermal gradient, a.
Of greater interest is how well the
gradient correction technique proposed
a = 0 K/m (Eq. 33)
in section II actually works when
applied to the convective heat transfer
6000
a = 0 K/m (Inferred)
problem illustrated in figure 2 with
a = 173 K/m (Eq. 33)
boundary conditions described earlier
a = 173 K/m (Inferred)
in this section. This is assessed by
using Eq. (33) to generate the
5500
temperature-time history of a point 1.9
mm below the surface. Then, this
temperature-time history is used to
5000
infer the surface heat flux using the
method of section II, with N = 5.
Finally, the inferred heat flux (dashed
4
5
6
7
8
9
lines) is compared to the original or
actual analytical solution for the heat
Time (s)
flux (solid lines) in Fig. 6. The Figure 6. Effect of gradient correction on inferred surface heat flux.
comparison is made for two cases: one
with a uniform initial wall temperature and one with an initial wall temperature gradient. The ripple in the inferred
heat flux is an artifact of the fitting process. Figure 7 shows that the errors (or differences) in both cases are +/- 2%.
The fact that both curves lie on top of each other demonstrates that the gradient correction technique from section II
has removed any errors arising from the initial gradient. Finally, Fig. 9 compares the surface temperature predicted
using Eq. (33) to that inferred from the temperature-time history of the subsurface point. These results also
demonstrate the validity of the gradient correction technique.

10

15

Surface Heat Flux (W/m2)

Downloaded by UNIVERSITY OF MARYLAND on August 8, 2015 | http://arc.aiaa.org | DOI: 10.2514/6.2010-4784

10000

8
American Institute of Aeronautics and Astronautics

333

4
3

Surface Temperature (K)

a = 0 K/m
a = 173 K/m

% Error

2
1
0
-1
-2
-3

332

0 K/m (Eq. 33)


0 K/m (Inferred)
173 K/m (Eq. 33)
173 K/m (Inferred)

331

330

329

Downloaded by UNIVERSITY OF MARYLAND on August 8, 2015 | http://arc.aiaa.org | DOI: 10.2514/6.2010-4784

a=
a=
a=
a=

6
7
8
9
Time (s)
Figure 7. Errors in inferred surface heat flux

7
8
9
Time (s)
Figure 8. Comparison of surface temperatures
predicted using Eq. (33) and inferred from the
subsurface temperature-time measurements.

VI. Conclusion
This paper has shown that not accounting for initial wall temperature gradients that can be encountered in
actively heated or cooled walls can lead to significant (>5%) errors in surface heat flux measurements. A data
interpretation method capable of reducing these errors is described. A solution for the temporal evolution of the
internal temperature distribution of a semi-infinite solid with an initial linear temperature gradient undergoing
convective heat transfer at its surface is also presented. It is hoped that the data interpretation method and the
analytical solution presented here will be helpful in assessing errors and therefore lead to improved measurements of
surface temperature and heat flux.

Acknowledgments
This work was sponsored by NASA Constellation University Institutes Project under grant NCC3-989.
Appreciation is expressed to Claudia Meyer of the NASA Glenn Research Center, and Kevin Tucker and Joseph H.
Ruf of the NASA Marshall Space Flight Center.

References
1

Diller, T.E., Advances in Heat Flux Measurements, Advances in Heat Transfer, volume 23, 1993.
Schultz, D.L. and Jones, T.V., Heat Transfer Measurements in Short-Duration Hypersonic Facilities, AGARD Report,
AGARD-AG-165, 1973.
3
Beck, J.V., Calculation of Surface Heat Flux from an Internal Temperature History, ASME Paper 62-HT-46, 1968.
4
Chen, C.J. and Thomsen, D.M., On Determination of Transient Surface Temperature and Heat Flux by Imbedded
Thermocouple in a Hollow Cylinder, AIAA Journal, vol. 13, no. 5, 1975, pp. 697 699.
5
Imber, M. and Khan, J., Prediction of Transient Temperature Distributions with Embedded Thermocouples, AIAA Journal,
vol. 10, no. 6, 1972, pp. 784 789.
6
Beck, J.V., Blackwell, B., and Haji-Sheikh, A., Comparison of Some Inverse Heat Conduction Methods using
Experimental Data, International Journal of Heat and Mass Transfer, vol. 39, no. 17, 1996, pp. 3649 3657.
7
Chen, C.J. and Chiou, J.S., Prediction of Surface Temperature and Heat Flux from an Interior Temperature Response,
Letters in Heat and Mass Transfer, vol. 3, 1976, pp. 539 548.
8
Monde, M., Analytical Method in Inverse Heat Transfer Problem using Laplace Transform Technique, International
Journal of Heat and Mass Transfer, vol. 43, 2000, pp. 3965 3975.
9
Ozisik, M.N. and Orlande, H.R.B., Inverse Heat Transfer, Taylor & Francis, 2000.
10
Abramowitz, M. and Stegun, I.A., Handbook of Mathematical Functions, National Bureau of Standards, Applied
Mathematics Series, 7th Ed., May 1968, pp. 300.
11
Carslaw, H.S. and Jaeger, J.C., Conduction of Heat in Solids, 2nd Edition, Oxford University Press, 1959, pp. 70-73.
12
Yan, X. T., On the Penetration Depth in Fourier Heat Conduction, 8th AIAA/ASME Joint Thermophysics and Heat
Transfer Conference, 24-26 June, 2002, St. Louis, Missouri.
2

9
American Institute of Aeronautics and Astronautics