Experimental Thermal and Fluid Science 28 (2004) 903908

www.elsevier.com/locate/etfs

Heat ux sensor with minimal impact on boundary conditions

Arash Saidi a, Jungho Kim
b

b,*

a
ATEC Inc., 387 Technology Dr., College Park, MD 20742, USA
Department of Mechanical Engineering, University of Maryland, 2181Glenn L Martin Hill, College Park, MD 20742, USA

Received 10 August 2003; received in revised form 20 December 2003; accepted 22 January 2004

Abstract
A technique for determining the heat transfer on the far surface of a wall based on measuring the heat ux and temperature on
the near wall is presented. Although heat transfer measurements have previously been used to augment temperature measurements
in inverse heat conduction methods, the sensors used alter the heat ow through the surface, disturbing the very quantity that is
desired to be measured. The ideal sensor would not alter the boundary condition that would exist were the sensor not present. The
innovation of this technique in that it has minimal impact on the wall boundary condition. Since the sensor is placed on the surface
of the wall, no alteration of the wall is needed. The theoretical basis for the experimental technique as well as experimental results
showing the heat ux sensor performance is presented.

2004 Elsevier Inc. All rights reserved.
Keywords: Heat ux sensor; Inverse heat conduction; Experimental technique; Boundary conditions

1. Introduction
Inverse heat conduction methods can be used to
determine heat ux and temperatures on an inaccessible
surface of a wall by measuring the temperature on an
accessible boundary (TS method, Fig. 1a). The noise
present in any measure of temperature, however, can
cause instabilities in the predicted heat uxes. It has
been suggested [1] that the prediction can be greatly
improved by measuring temperature at two locations
(Fig. 1b). Altering the wall to include an interior thermocouple cannot be performed in many applications,
however, and installation of an interior thermocouple
can result in material inhomogeneities that change the
heat ow through the wall. By numerical experiments
and a sensitivity analysis we show that incorporating a
measurement of the heat ux at the accessible boundary
(TS/HFS method, Fig. 1c) can be used to improve the
calculation.
The objective of this work was to develop a method
by which stable predictions of the heat transfer on an
inaccessible boundary could be obtained without alter*

Corresponding author. Tel.: +1-301-405-5437; fax: +1-301-3149477.

E-mail address: kimjh@eng.umd.edu (J. Kim).
0894-1777/$ - see front matter
2004 Elsevier Inc. All rights reserved.
doi:10.1016/j.expthermusci.2004.01.004

ing the thermal boundary condition that would have

existed were a sensor not present. As mentioned above,
it is known that increasing the number of the sensors
will enhance the stability of the inverse solution by
incorporating more data about the heat ux at the
inaccessible boundary. An alternative to inserting a
thermocouple within the wall is to measure the heat ux
at the accessible surface. It will rst be shown below via
numerical experiments with an inverse method that this
extra input (heat ux) increases the stability of the
procedure and is less prone to the inherent instability of
the ill-posed problem of inverse heat conduction. A
sensitivity analysis is then carried out to show that the
inverse conduction results are less sensitive to errors in
heat ux sensor values than errors in temperature input.
A sensor that measures both heat transfer and temperature on an accessible boundary with minimal impact on
the boundary condition is then described and results of
experiments with this sensor are presented.

2. Numerical experiments
In order to demonstrate the advantage of using both
heat ux and temperature data at an accessible boundary to predict heat transfer at an inaccessible boundary,

904

A. Saidi, J. Kim / Experimental Thermal and Fluid Science 28 (2004) 903908

Nomenclature
h
K
q
S
T
Y

coecient of heat transfer

gain coecient in Eq. (5)
heat ux value
sensitivity to errors
temperature
measured values of temperature

h
s
c
x
1, h
1, s

active sensor (heater)

surface
cooler device
unknown heat ux
bulk temperature of water in cooler
bulk temperature of air

Subscripts
1
inaccessible surface
2
accessible surface

inaccessible
wall (point 1)

accessible
wall (point 2)

T2 (t)

q (t)
1

(a) TS method
inaccessible
wall (point 1)

q (t)
1

T2 r; tM T0

accessible
wall (point 2)

T2 (t)

(b) Two temperature method

q (t)
1

M
X

qn D/r; tMn

n1

T (t)
3

inaccessible
wall (point 1)

In this integral, qk is the partial eect of q1 t on an

interval of time as an impulse of heat ux and /r; t is
the resulting eect of that partial heat ux, so the total
eect of q1 t is equal to summation of all these eect, or
the integral of all partial eects. One can approximate
this integral numerically, and obtain:

accessible
wall (point 2)

T2 (t), q2(t)

(c) TS/HFS method

Fig. 1. Schematic of inverse heat conduction methodologies.

a program was developed that uses both temperature

and heat ux data on an accessible boundary (point 2)
to estimate the heat ux at an inaccessible boundary
(point 1), see Fig. 1.
The eect of the heat ux q1 t on the temperature
and the heat ux at the other side of the wall (the
location of the sensor) is quantied by a superposition
method (see Fig. 1). There are dierent methods to derive Duhamels theorem (see [1]), but the nal direct
form of the theorem becomes:
Z t
o/r; t  k
dk
1
T2 r; t T0
qk
ot
0

One can use this integral in reverse to estimate the q1 t

from measurements at a given point, namely T r; t.
Several methods exist for inverting Eq. (1) or (2). One
method that we used is the exact matching of measured
temperature, and it is usually called Stolz algorithm [2].
In this method, discrete transient temperatures from a
single sensor are used to estimate the surface heat ux as
a function of time, and the calculated temperatures are
made equal to measured values. This method is called
the exact method because the calculated temperatures
are matched to the measured values. In contrast, several
approximate methods have been developed that instead use a least squares technique. Stolzs algorithm
is given by:
P
YM  i1 qi D/Mi  T0
qM
3
/1
where M denotes the time steps, Y are the measured
values of temperature, and / are exact solutions to
partial heat uxes (these values must be calculated at
any given time step). This method is sequential in nature, and each q at given time step is determined based on
the previous values of q and Y .
An extension of this method is achieved by using
several sensors at point 2. In our case, the measurements
of temperature and heat ux at point 2, in the mixed
method (Fig. 1c) are handled by this extension. The heat
ux values are estimated using [1]:
qM

2
X
i1

Ki YMi1  TMi1 jqM...0

A. Saidi, J. Kim / Experimental Thermal and Fluid Science 28 (2004) 903908

K is called the gain coecient and dened as:

/
Ki P2 i 2
j1 /j

In Eq. (5) /1;2 are eects of an impulse heat ux at point

1 on temperature and heat ux values at point 2 (Fig. 1),
respectively. These values are calculated using a direct
heat conduction solution of a one dimensional slab.
A series of numerical experiments were conducted
using the described method. In the rst of these eorts,
we calculated the heat ux at point 1 using two methods,
rst using only temperature sensor data and then using
both temperature and heat ux data at point 2. These
data were articially generated from a solution of the
problem assuming the exact input at point 1. The heat
conduction equation was then solved to determine the
temperature at point 2. To simulate the eect of the
actual errors in temperature, a random number generator was used to generate numbers between 0 and 1.
These numbers were multiplied by a percentage (5%, 8%
and 12%) of the exact value and added to this value.
It was noticed that the TS/HFS method resulted in a
more accurate estimation of the heat ux at far
boundary q1 than the TS method. In order to quantify
this eect an error value was dened:
Error

dq1
q1

1
TS method
TS/HFS method

0.8

one case was analyzed. The sensitivity of the estimated

values to each of these inputs was dened as: For T2 :
ST2

0.6

dq1 =q1
dT2 =T2

For q2 :
Sq2

dq1 =q1

dq2 =q2

These values indicate how sensitive the estimated values

of q1 are to the relative errors in input values separately.
The values for the same problem discussed above were
recalculated resulting in these values for ST2 and Sq2 :
ST2 13:7
Sq2 1:4
It is evident from this sensitivity study that the errors in
estimation of heat ux at a far boundary q1 is one
order of magnitude more sensitive to errors in a temperature sensor than to errors in heat ux at point 2,
demonstrating the importance of incorporating heat ux
data in the inverse heat conduction estimation process.
It is known that the estimation process becomes
unstable when the time steps become smaller than a
certain value in Stolzs algorithm. We carried out a
numerical experiment to investigate the eect of the
mixed method on this instability. We chose the problem
of triangular heat ux input and changed the total time
of the triangular heat imposition. Accordingly, the time
steps were chosen shorterthe results are shown on Fig.
3. The time is normalized with respect to the material
and thickness of the material, as t at=L2 . The error
shown in this gure is dened and normalized as in Eq.
(6). It is seen that when t becomes smaller that about 5,
the error in the method using only temperature sensor at
point 2 (Fig. 1a) becomes increasingly larger and after
some limit the results are no longer useful. This gure

10
TS method
TS/HFS method

Error in q1

Error in relative estimatimated

q1

where, dq1 is the average of the absolute dierence between estimated and exact values of q1 over the whole
time domain. This value was normalized with respect to
the input error in the temperature and heat ux values.
The results shown on Fig. 2 show that the TS/HFS
method decreases the error levels considerably. For a
12% error (in temperature at point 2), the average error
in the estimated value of q1 is about 85% of the average
value for the TS method while it is about 42% for the
TS/HFS method, demonstrating the superiority of the
mixed method with respect to errors in estimated heat
ux values.
In order to investigate this eect, the errors in temperature T2 and heat ux input q2 data were separated and the sensitivity of q1 to each of these inputs in

905

10

100

0.1

0.4
0.2
0
4

10

11

12

13

14

0.01

t_plus

Error (%) in sensor values

Fig. 2. Relative eect of erroneous data on estimated q1 values.

Fig. 3. Eect of the time step and time of the rise of the triangular heat
ux input on the estimated heat ux error.

906

A. Saidi, J. Kim / Experimental Thermal and Fluid Science 28 (2004) 903908

also shows that errors become larger in the mixed

method as well (Fig. 1c), but the increase is much more
moderate. This numerical experiment once again indicates the superiority of using both temperature and heat
ux data as input values.

wall q00x as shown by the simple analysis of the heat ux

sensor performance given below. The passive temperature sensor is cooled through convection by a heat
transfer coecient hs . An energy balance on this sensor
yields
q00x hs Ts  T1;s

3. Heat ux sensor

Solving for the passive sensor temperature yields

Several heat ux sensors have been developed to

measure the thermal radiation in aerodynamic ows.
For example, Borell and Diller [3] designed an apparatus
for measuring heat uxes in convective airows. A different type of heat ux sensor was developed by Hager
et al. [4], which consisted of several layers of thin lms
that form a dierential thermopile across a thin oxide
layer. A sensor developed by Leclercq and Thery [5]
measured the heat ux by determining the temperature
gradient over a tangent plane to the heat ux surface.
Physical asymmetries are used to deect the heat ux
lines and generate a temperature gradient over a planar
thermopile. This gradient is directly proportional to the
imposed heat ux and produces a voltage across the
thermopile. A number of such thermopiles are fabricated in series to amplify the voltage proportional to the
heat ux.
Any heat ux sensor is only able to measure the heat
transfer through it. All of the sensors described above
alter the heat ow through the surface, disturbing the
very quantity to be measured. The ideal sensor would
minimally alter the boundary condition that would exist
were the sensor not present. Such a sensor can be made
using the concept described below.
A schematic diagram of the sensor is shown on Fig. 4.
A small resistance heater (active heater in Fig. 4) is
attached to the accessible boundary of a wall, and its
temperature is controlled by an electronic feedback loop
to track the temperature of a passive temperature sensor
(passive sensor in Fig. 4) mounted on the same
boundary a short distance away. The passive temperature sensor is very thin so the wall boundary condition is
only minimally altered. The active heater is cooled by an
ecient and substantial cooling mechanism from behind
(for example, circulating chilled water or an impinging
air or water jet).
By measuring the heat added to the active heater (q00h
in Fig. 4), we can determine the heat ux through the

Ts

T,s

Passive sensor

hs

T,h

q''

q''c

hh

Ts

Th

q''
x

q ''
x

Active heater

Wall

Fig. 4. Schematic representation of heat ux sensor.

q00x
T1;s
hs

10

An energy balance on the active heater is given by

q00x q00h q00c

11

where
q00c hh Th  T1;h

12

If the heater temperature Th is kept at the same temperature as the sensor (this is done using a feedback
circuit as described below), then Th Ts and Eqs. (10)
and (12) can be substituted into Eq. (11) to yield


00
00 hh
q h qx
13
 1 hh T1;s  T1;h
hs
Examination of Eq. (13) indicates the following properties of the heat ux sensor:
(1) If the heat transfer coecients (hs and hh ) and the
environment and coolant temperatures (T1;s and
T1;h ) are constant and hh > hs , then the heat supplied to the heater is linearly proportional to the
heat transfer through the substrate. By measuring
q00h , q00x can be determined.
(2) If hh is larger than hs , the heat ux sensor acts to amplify theh heat i transfer through the wall by an
amount hhhs  1 without disturbing the wall temperature.
(3) In order to avoid the possibility of negative q00h , we
should keep T1;s > T1;h and hh =hs > 1.
(4) If hs and hh are constant, then drifts in T1;s and T1;h
simply result in an oset to q00h . The heat ux sensor
can be operated with dierent T1;s and T1;h by using
Eq. (13) to correct the output.
A schematic of the electronic feedback circuit is
shown on Fig. 5. The voltage applied to the active heater
is controlled using a feedback control circuit similar to
that described by Bae et al. [6]. This circuit maintains the
temperature of the active heater equal to the temperature of the passive sensor. The op-amp in the control
circuit measures the imbalance in the bridge and outputs
the voltage needed to keep ratio Ractive =Ru equal to the
resistance ratio on the right side of the bridge. The
heater resistance (Ractive ), and thus the heater temperature are controlled by changes in the resistance of the
passive sensor (Rpassive ) due to temperature changes on
the surface of the wall. The voltage across the heater

A. Saidi, J. Kim / Experimental Thermal and Fluid Science 28 (2004) 903908

907

Coolant in

24 V
1K

Coolant out
Ru

Passive sensor

Rt

Active sensor
10 mm copper block

+
-

+
Ractive

R active,0

Fig. 6. Schematic of test rig.

R passive
V
A
R passive,0

Fig. 5. Electronic feedback loop to control heater resistance.

(Vout ) is measured and used to determine q00h . The resistance of the passive sensor can be determined by measuring VA and the current through the small resistor
Rpassive;0 . Since the temperature of the active heater
tracks the temperature of the accessible boundary (as
measured by the passive heater), the presence of the
active heater and the cooling of this heater do not alter
the thermal boundary condition on this boundary.
The sensor described above is preferable to other heat
transfer measuring systems for the following reasons:
The sampling rate of each heater can reach a frequency as high as 15 kHz, allowing for rapid temporal discrimination of changes in heat ux, if needed.
The measurement technique could be capable of handling high temperatures.
The incorporation of a feedback loop to maintain the
temperature of the heater at the same temperature as
the surface that is undisturbed by the active sensor
eliminates the problem of sensor-structure interaction
that can occur with other heat ux sensing techniques
and the resulting heat ux errors.
Unlike thin lm sensors, the output of this sensor is
directly measurable, with no need for amplication
by using either a series of sensors and/or ampliers.
The proposed sensor can be easily designed to survive
hostile environments.

4. Heat ux sensor performance

The concept has been tested using the setup shown in
Fig. 6. A 5.2 cm 7.6 cm copper plate 10 mm thick was
used for the wall, and heated using a Minco foil heater
connected to a variable voltage source. A pair of RTDs
for the active and passive sensor were specially made for
this application by Vishay Electronics. The RTDs consist of a 2.54 lm thick etched platinum foil sandwiched
between two 0.0254 mm thick Kapton lms. The passive

and active sensor resistances were 988.9 and 98.6 X at 20

C. The RTDs had a nominal TCR of 0.0035 X/X C
with dimension 10 mm 5 mm. The active sensor was
cooled from the back using an impinging jet of water at
a constant ow rate (160 ml/min) and temperature (30
C) to provide a high hh , and a fan was used to provide a
constant hs to cool the plate. The air and water inlet
temperatures (T1;s 25 C and T1;h 30 C) were
measured along with the voltage across the heater Vout
as q00x was increased to verify that they remained constant. A plot of q00h vs. q00x is shown on Fig. 7, and indicates a linear variation as expected. The sensitivity is
seen to be quite high.
The results of the steady state heat ux measurement
using the sensor as the power to the foil heater was increased are shown on Fig. 8. The actual heat ux values
were computed using the measured voltage across the
foil heater and its resistance, while the sensor values
were obtained using the feedback circuit. The agreement
is observed to be well within 10%.
This sensor was then used to estimate the transient
heat ux into the inaccessible side of the 10 mm thick
copper slab for the case where the input heat ux was
suddenly decreased from 6500 to 0 W/m2 . The heat ux
sensor provided heat ux and temperature data on the
accessible side of the copper slab, and the inverse heat
conduction model and software were used to estimate
the heat ux variation with time on the inaccessible side.
These results are presented in Fig. 9. The estimated heat
ux agrees very well with the actual heat ux variation.

15000

q''h (W/m2)

Vout

Thin film heater

10000

5000

0
3000

4000

5000

6000

q''x (W/m2)

Fig. 7. A plot of q00h vs. q00x .

7000

908

A. Saidi, J. Kim / Experimental Thermal and Fluid Science 28 (2004) 903908

x=y
7000

Sensor heat flux (w/m )

+10%
6000

-10%

5000

4000

3000
3000

4000

5000

6000

7000

Actual heat flux (W/m )

Fig. 8. Comparison of actual heat ux into the copper plate vs. the heat
ux measurement using the active and passive sensors.

Heat flux (W/m )

Acknowledgements
The support of Kirtland AFB under SBIR F2960102-C-0138 is gratefully acknowledged. The grant monitor was Dr. Choon Tham.

12000
ACTUAL

10000

on an inaccessible wall was conrmed via numerical

experiments. A novel heat ux sensor that has a minimal
eect on the wall thermal boundary condition has been
designed and built. Experiments were conducted to
verify the feasibility and accuracy of the concept. It was
observed that the proposed sensor is quite sensitive to
the input heat ux value and possesses good accuracy.
In this stage of our eort, other problems associated
with the sensor such as the surface conditions and the
temporal response of the cooling device were not considered. These issues can aect repeatability of the sensor and time response of it to changes, but they do not
alter the main conclusions of this study.

ESTIMATED

8000

References

6000
4000
2000
0
0

100

200

300

400

500

600

Time (s)

Fig. 9. Estimated heat uxes (Heat ux decreased from 6500 to 0

W/m2 ).

5. Conclusions
The advantage of incorporating heat ux into an inverse heat conduction method for predicting heat uxes

[1] J.V. Beck, B. Blackwell, C.St. Clair Jr., Inverse Heat Conduction:
Ill-posed Problems, Wiley, New York, 1985.
[2] G. Stolz, Numerical solutions to an inverse problem of heat
conduction for simple shapes, ASME Trans. J. Heat Transfer 82
(1960) 2026.
[3] G.J. Borell, T.E. Diller, A convection calibration method for local
heat ux gages, ASME Trans. J. Heat Transfer 109 (1987) 8389.
[4] J.M. Hager, L.W. Langley, S. Onishi, T.E. Diller, Microsensors for
high heat ux measurements, J. Thermophys. Heat Transfer 7 (3)
(1993) 531534.
[5] D. Leclercq, P. Thery, Apparatus for simultaneous temperature
and heat ux measurements under transient conditions, Rev. Sci.
Instrum. 54 (1983) 374380.
[6] S. Bae, M.H. Kim, J. Kim, Improved technique to measure time
and space resolved heat transfer under single bubbles during
saturated pool boiling of FC-72, Exp. Heat Transfer 12 (3) (1999)
265278.