You are on page 1of 23

See

discussions, stats, and author profiles for this publication at: https://www.researchgate.net/publication/230931493

MEMS-based pressure and shear stress sensors


for turbulent flows

Article in Measurement Science and Technology July 1999


DOI: 10.1088/0957-0233/10/8/302

CITATIONS READS

76 60

2 authors, including:

Lennart Lfdahl
Chalmers University of Technology
118 PUBLICATIONS 932 CITATIONS

SEE PROFILE

All content following this page was uploaded by Lennart Lfdahl on 23 March 2017.

The user has requested enhancement of the downloaded file. All in-text references underlined in blue are added to the original document
and are linked to publications on ResearchGate, letting you access and read them immediately.
Meas. Sci. Technol. 10 (1999) 665686. Printed in the UK PII: S0957-0233(99)99136-1

MEMS-based pressure and shear


stress sensors for turbulent flows

Lennart Lofdahl and Mohamed Gad-el-Hak
Thermo and Fluid Dynamics, Chalmers University of Technology, 412 96 Goteborg,
Sweden
Department of Aerospace and Mechanical Engineering, University of Notre Dame,
Notre Dame, IN 46556, USA
E-mail: lelo@tfd.chalmers.se and mohamed.gad-el-hak.1@nd.edu

Received 10 November 1998, in final form and accepted for publication 2 February 1999

Abstract. From a fluid dynamics perspective, the introduction of microelectromechanical


systems (MEMS) has considerably broadened the spectrum of workable experiments. A
typical MEMS sensor is at least one order of magnitude smaller than traditional sensors used
to measure instantaneous flow quantities such as pressure and velocity. The microsensors can
resolve all relevant scales even in high-Reynolds-number turbulent flows, and arrays of
microsensors make it feasible, for the first time, to achieve complete information on the
effective small-scale coherent structures in turbulent wall-bounded flows. In this paper we
focus on the use of MEMS for the diagnosis of turbulent shear flows and survey the status and
outlook of microsensors as used for measurements of fluctuating wall pressure and wall shear
stress, two quantities which we deem particularly difficult to measure with conventional
probes. For both wall pressure and wall shear stress sensors, we give general background,
design criteria and calibration procedure. Examples of measurements conducted with
MEMS-based sensors are provided and the minute devices are compared to their larger
cousins.

Keywords: pressure, shear stress, microsensors, MEMS, turbulence measurements

Contents 1. Introduction

1. Introduction 665 Microelectromechanical systems (MEMS) can be considered


2. Pressure sensors 666 as the next logical step in the silicon revolution
2.1. Background 666 which eventually started when silicon microelectronics
2.2. Pressure sensor principles 667 revolutionized the semiconductor and computer industry
2.3. Requirements for turbulent flows 669 with integrated circuits (ICs). An additional dimension is
now being added by MEMS since it allows ICs to break
2.4. Piezoresistive pressure sensors 670
the confines of the electronic world and interact with the
2.5. MEMS pressure sensors 672 environment through sensors and actuators. This is a giant
2.6. Outlook for pressure sensors 673 step and it cannot be excluded that MEMS in the near
3. Wall shear stress sensors 674 future will have the same impact on society and economy
3.1. Background 674 as the IC has had since the early 1960s. Generally, a
3.2. Thermal sensor principle 675 typical MEMS device has a characteristic length of less
3.3. Calibration 677 than 1 mm but more than 1 m. It combines electrical
3.4. Spatial resolution 678 and mechanical components and can be either individual
3.5. Temporal resolution 679 units or parts of a more complex mechanical system with
3.6. MEMS thermal sensors 680 integrated electronics. Motors, turbines, actuators, valves,
gears and transducers of less than 100 m size have been
3.7. Floating element sensors 681
successfully fabricated. During the last decade there has
3.8. MEMS floating element sensors 682 been a tremendous interest in developing MEMS, resulting
3.9. Outlook for shear stress sensors 683 in major impacts in both industrial and medical applications.
4. Concluding remarks 684 From a fluid dynamics perspective, the introduction
Acknowledgments 684 of MEMS has considerably broadened the spectrum of
References 684 workable experiments. A typical MEMS sensor is less

0957-0233/99/080665+22$30.00 1999 IOP Publishing Ltd 665


L Lofdahl and M Gad-el-Hak

than 100 m in size, and this is at least one order of structures with submicrometre resolution are formed. More
magnitude smaller than traditional sensors used to measure details on MEMS fabrication can be found in the numerous
instantaneous flow quantities such as velocity, pressure and surveys that have been published; here we cite only a few:
temperature. The small spatial extension implies that both Petersen [3], Bryzek et al [4] and Tien [5].
the inertial mass and the thermal capacity are reduced, thus In this paper, our attention is focused on MEMS
making MEMS sensors suitable for dynamic measurements sensors for measuring fluctuating wall pressure and wall
of flow quantities in high-Reynolds-number turbulent flows shear stress beneath a turbulent boundary layer. The
where a high-frequency response and a fine spatial resolution most prominent advantage of MEMS sensors is the small
are essential. For instance, pressure and velocity sensors measuring volume achieved, and this suggests an extremely
with, respectively, diaphragm side length and wire length good resolution of even the smallest eddies in the turbulent
of less than 100 m are in use today. Another important flow field. Section 2 gives a general background, design
feature is that MEMS sensors are not hand-made, but are criteria and calibration procedure for the wall pressure sensor,
produced by photolithographic methods which implies that and provides examples of measurements conducted with
each unit is fabricated to extremely small tolerance and at MEMS-based sensors. Significant quantities like rms values,
very low cost. The latter trait makes it possible to use a correlations and advection velocities obtained with MEMS
large number of sensors to cover large areas/volumes of the sensors, yielding a spatial resolution of 510 viscous wall
flow field. This in turn makes it feasible to study coherent units, are discussed. Section 3 gives a general background,
structures and to effectively execute reactive control for design criteria and calibration procedure for wall shear stress
turbulent shear flows. An additional important advantage sensors, and provides examples of measurements conducted
of microfabrication is that it enables the packing of sensors with both direct and indirect methods. Finally, section 4
in arrays on the same silicon chip. These characteristics are provides some concluding remarks for both these types of
crucial for measurements of fluctuating quantities such as
MEMS sensors and outlines the outlook for MEMS in fluid
wall pressure and shear stress beneath a turbulent boundary
dynamics.
layer. By measuring the fluctuating wall pressure, which is
coupled via a complex interaction to gradients of both mean
shear and velocity fluctuations as described by the transport 2. Pressure sensors
equations for Reynolds stresses [1, 2], much information of
vital importance for turbulence modelling, reactive control 2.1. Background
and aeroacoustics can be obtained without disturbing the
In turbulence modelling, flow control and aeroacoustics, the
interior of the flow. The same is true for the time-resolved
fluctuating wall pressure beneath a wall-bounded flow is a
part of the fluctuating wall shear stress which provides a
crucial parameter. By measuring this random quantity much
measure of the unsteady structures in the flow, is responsible
information can be gleaned about the boundary layer itself
for the individual momentum transfer and is an indicator of
without disturbing the interior of the flow. The fluctuating
turbulence activities.
wall pressure is coupled via a complex interaction to gradients
MEMS are created using specialized techniques derived
of both mean shear and velocity fluctuations as described by
and developed from IC technology in a process called mi-
cromachining. Usually three main technologies are distin- the transport equations for Reynolds stresses [1, 2].
guished: bulk micromachining, surface micromachining and The characteristics of the fluctuating wall pressure field
micromoulding. Bulk micromachining involves different beneath a turbulent boundary layer have been extensively
techniques which use simple, single-crystal, silicon wafers as studied in both experimental and theoretical investigations,
structural material. By using anisotropic silicon etching and and reviews of earlier work may be found in Blake [6],
wafer bonding, three-dimensional structures such as pres- Eckelmann [7] and Keith et al [8]. However, from
sure sensors and accelerometers have been fabricated. In the experimental perspective, our knowledge of pressure
surface micromachining, the silicon substrate is used as sup- fluctuations is far from being as comprehensive as that
port material, and different thin films such as polysilicon, of velocity fluctuations since there is lack of a generally
silicon dioxide and silicon nitride provide sensing elements applicable pressure-measuring instrument which can be used
and electrical interconnections as well as structural, mask in the same wide variety of circumstances as the hot-wire
and sacrificial layers. The basis of surface micromachining anemometer. In spite of this, some general facts have been
is sacrificial etching where free-standing, thin-film structures established for wall pressure fluctuations. For example, the
are free-etched by the lateral underlying sacrificial layer. Sur- order of magnitude of the rms value, the general shape of the
face micromachining is simple but powerful and, despite its power spectra and the spacetime correlation characteristics
two-dimensional nature, different complex structures such (see, e.g. Harrison [9], Willmarth and Wooldridge [10], Bull
as pressure sensors and actuators have been fabricated. The [11], Bull and Thomas [12], Schewe [13], Blake [6], Lauchle
third technique, micromoulding (the LIGA (lithographie gal- and Daniels [14], Farabee and Casarella [15].
vanoformoung abformung) technique), is more like conven- A clear shortcoming in many of the experiments
tional machining in concept. The fabrication principle is designed to measure pressure fluctuations is the quality of the
based on forming a metal mould using lithographic tech- data. In the low-frequency range, data may be contaminated
niques, which allow fine feature resolution. Typically, tall by facility-related noise, while in the high-frequency range
Normally the fabrication of a prototype sensor, i.e. the first batch, is very
the spatial resolution of the transducers limits the accuracy.
costly, but once the fabrication principle has been outlined the unit cost drops The former difficulty is usually circumvented by noise
dramatically. cancellation techniques (Lauchle and Daniels [14]) or by

666
MEMS-based pressure and shear stress sensors for turbulent flows

using a free-flight glider as an experimental platform. This


problem may not be considered as a major obstacle today. At
the other end of the spectrum, the spatial resolution problem
is more difficult to handle. The main criticism raised is that
in many experiments the size of the pressure transducer used
is far too large in relation to the thickness of the boundary
layer in context, let alone in relation to the characteristic small
scale. The ultimate solution is of course to use small sensors, Figure 1. Cross-sectional view of a piezoelectric silicon
but testing in tunnels containing fluids of high viscosity microphone. From Royer et al [19].
to increase the viscous length scale in the flowhas also
been tried. Using highly viscous fluids creates its own set of fabrication is also compatible with other IC techniques so
of problems. Very specialized facilities and instrumentation electronic circuitry like preamplifiers can be integrated close
are needed when oil or glycerin, for example, are used as to the sensoran important factor for improving pressure
the working fluid. An oil tunnel is expensive to build and sensor performance. Here we give a short background of the
to operate. Moreover, the Reynolds numbers achieved are above mentioned principles for pressure sensor operations.
generally low, and it is not clear how to extrapolate the We use the word microphone to mean a device whereby
results to higher-Reynolds-number flows (Gad-el-Hak and sound waves are caused to generate an electric current for
Bandyopadhyay [16]). the purpose of transmitting or recording sound.
Pinhole microphones have also been utilized in
an attempt to improve the sensors spatial resolution. 2.2.1. Piezoelectric sensors. A piezoelectric sensor
Unfortunately, results from this type of arrangement are in consists of a thin diaphragm which is either fabricated from
general questionable, and to this end the use of pinhole a piezoelectric material or mechanically connected to a
microphones must be considered inconclusive. Bull and cantilever beam consisting of two layers of piezoelectric
Thomas [12] concluded that the use of pinhole sensors in material with opposite polarizations. Vertical movement of
air may lead to severe errors in the measured spectra, while the diaphragm causes a stress in the piezoelectric material
Farabee [17], Gedney and Leehey [18] and Farabee and and generates an electric output voltage. Royer et al [19]
Casarella [15] all claimed that pinhole sensors are most presented the first MEMS-based piezoelectric sensor shown
effective for wall pressure measurements. in figure 1. This transducer consists of a 30 m thick
Based on the above arguments, it seems then that the only silicon diaphragm with a diameter of 3 mm. On top of the
realistic solution to improve measurements of fluctuating diaphragm a layer of 35 m ZnO is deposited, sandwiched
wall pressure is to use small sensors. For this reason, between two SiO2 layers that contain the upper and lower
microfabrication offers a unique opportunity for reducing aluminium electrodes. The sensor can be provided with an
the diaphragm size by at least one order of magnitude. integrated preamplifier, and is basically used for microphone
MEMS also provide an opportunity to fabricate inexpensive applications. A sensitivity of 50250 V Pa1 and a
(for large production volumes) dense arrays of pressure frequency response in the range of 10 Hz10 kHz (flat within
sensors for correlation measurements and studies of coherent 5 dB) were recorded. Other researchers have presented
structures in turbulent boundary layers. In this section, we similar piezoelectric silicon microphones (Kim et al [20],
describe the basic principles used for MEMS-based pressure Kuhnel [21], Schellin et al [22]), and the sensitivities of
sensors/transducers/microphones. We look specifically at these microphones were reported to be in the range of
the design of pressure sensors utilizing the piezoresistive 0.0251 mV Pa1 . However, applications of the piezoelectric
principle, which is particularly suited to measurements of microphone to turbulence measurements are strongly limited
wall pressure fluctuations in turbulent flows. This section also because of its high noise level, which has been found to be in
contains a summary of measurements conducted with MEMS the range of 5072 dB(A) SPL (Sessler [23], Kim et al [20]).
pressure sensors and, where possible, comparisons with (These noise levels are most commonly measured using an
conventional data. Finally, we discuss the outlook for A-weighted filter, in dBs relative to 2 105 Pa, which is
the use of MEMS-based pressure sensors for turbulence the lowest sound level detectable by the human ear. The
measurements. A-weighted filter corrects for the frequency characteristics
of the human ear and provides a measure of the audibility of
the noise.)
2.2. Pressure sensor principles
Many different methods have been proposed for the detection 2.2.2. Piezoresistive sensors. A piezoresistive sensor
of pressure fluctuations. The principles available are consists of a diaphragm that is usually provided with four
based on detecting the vibrating motion of a diaphragm piezoresistors in a Wheatstone bridge configuration. One
using piezoelectric, piezoresistive and capacitive techniques. common arrangement is to position two of the gauges in
These principles were already known at the beginning of the the middle and two at the edge of the diaphragm. When
twentieth century, but the introduction of photolithographic the diaphragm deflects, the strains at the middle and at
fabrication methods over the last two decades has provided a the edge of the diaphragm have opposite signs, which
strong impetus since this technology offers sensors fabricated causes an opposite effect on the piezoresistive gauges. The
to extremely small tolerance, increased resolution due to a most important advantage of this detection principle is its
high degree of miniaturization and low unit cost. This method low output impedance and its high sensitivity. The main

667
L Lofdahl and M Gad-el-Hak

2.2.3. Capacitive sensors. Most MEMS-fabricated


pressure sensors are based on the capacitive detection
principle and the majority of these sensors are used as
microphones. Figure 3 shows a cross-sectional view of
such a capacitive microphone together with the associated
electrical circuitry. The latter must be included in the
discussion since the preamplifier constitutes a vital part in
determining the sensitivity of the capacitive probe. Basically,
Figure 2. Cross-sectional view of a piezoresistive silicon a capacitive microphone consists of a backchamber (with a
microphone. From Schellin and Hess [24]. pressure equalizing hole), a backplate (with acoustic holes),
a spacer, and a diaphragm covering the air gap created by
the spacer located on the backplate. The capacitor Cm and
a dc voltage source Vb constitute the sensing part of the
microphone. Fluctuations in the flow pressure field above
the diaphragm cause it to deflect which in turn changes
the capacitance Cm . These changes are amplified in the
preamplifier Ho which acts as an impedance converter with a
bias resistance Rb and an input capacitance Ci . In figure 3,
Cp is a parasitic capacitance which is of interest when
determining the microphone attenuation. In discussing the
sensitivity of a capacitive sensor, the open-circuit sensitivity
is a relevant quantity and is considered to consist of two
components, namely the mechanical sensitivity Sm and the
electrical sensitivity Se . The total sensitivity is a combination
of the two. The mechanical sensitivity is defined as the
increase of the diaphragm deflection dw resulting from an
increase in the pressure dp acting on the microphone
Sm = dw/dp. (1)
From figure 3 we can obtain the relation dw = dsa0 , where
sa0 is the depth of the air gap between the diaphragm and
the backplate. The electrical sensitivity of the microphone is
given by the change in the voltage across the air gap resulting
from a change in the air gap thickness dsa0 . Thus,
Se = dV /dsa0 (2)
The quasi-static, open-circuit sensitivity Sopen of a condenser
Figure 3. Cross-sectional view of a capacitive microphone. The microphone may be defined as Sopen = Sm Se . The output
microphone is connected to an external dc bias voltage source, and signal em can then be expressed as
loaded by a parasitic capacitance, bias resistor and a preamplifier
with an input capacitance. From Scheeper et al [28]. em = Sm Se p (3)
where p is the fluctuating pressure. For details of this
drawback is that the piezoresistive material is sensitive to derivation, see Scheeper et al [28].
both stress and temperature which, unfortunately, gives Hohm [29] presented the first electret microphone
the piezoresistive sensor a strong temperature dependency. based on MEMS technology. The backplate, (1 1) cm2
Schellin and Hess [24] presented the first MEMS-fabricated silicon, was provided with one circular acoustic hole of
piezoresistive sensor, shown in figure 2. This particular diameter of 1 mm. A 2 m thick SiO2 layer was used
sensor was used as microphone and it had a diaphragm as the electret and was charged to about 350 V. The
made of 1 m thick, highly boron-doped silicon with an diaphragm was a metallized 13 m thick foil of diameter
area of 1 mm2 . The diaphragm was equipped with 250 nm of 8 mm. Later, polymer foil diaphragms were used in
capacitive microphones by Sprenkels [30] and Murphy et al
thick, p-type polysilicon resistors, which were isolated from
[31]. In these microphones, the fabrication was made more
the diaphragm by a 60 nm silicon dioxide layer. Using
compatible with standard thin-film technology. Bergqvist
a bridge supply voltage of 6 V, this transducer showed a [32] and Bergqvist and Rudolf [33] showed that MEMS-
sensitivity of 25 V Pa1 and a frequency response in the fabricated microphones can achieve a high sensitivity. For
range of 100 Hz5 kHz (+3 dB). However, the sensitivity example, microphones with a (2 2) cm2 diaphragm were
was lower than expected by a factor of ten, which was shown to have an open-circuit sensitivity in the range
explained by the initial static stress in the highly boron- of 1.413 mV Pa1 . In these microphones, diaphragms of
doped silicon diaphragm. To improve the sensitivity of thickness 58 m were fabricated using anisotropic etching
piezoresistive sensors, different diaphragm materials, such as in a KOH solution and applying an electrochemical etch-stop.
polysilicon and silicon nitride, have been explored (Guckel More details on the design and performance of capacitive
[25], Sugiyama et al [26], Liu et al [27]). pressure sensors can be found in Scheeper et al [28].

668
MEMS-based pressure and shear stress sensors for turbulent flows

Table 1. Summary of the chronological development silicon micromachined microphones.


Diaphragm Upper Equivalent
Transducer area frequency Sensitivity noise level
principle (mm2 ) (kHz) (mV Pa1 ) (dB(A))
Royer et al [19] Piezoelectric 33 40 0.25 66
Hohm [29] Capacitive 0.8 0.8 >20 4.3 54
Muller [81] Piezoelectric 33 7.8 0.05 72
Franz [82] Piezoelectric 0.8 0.9 46 0.025 68
Sprenkels [30] Capacitive 33 >10 7.5 25
Sprenkels et al [83] Capacitive 33 >10 25
Hohm and Hess [84] Capacitive 0.8 0.8 2 1
Voorthuyzen et al [85] Capacitive 2.45 2.45 15 19 60
Murphy et al [31] Capacitive 33 >15 48 30
Bergqvist and Rudolf [33] Capacitive 22 4 13 31.5
Schellin and Hess [24] Piezoresistive 11 10 0.025 at 6 V
Kuhnel [21] FET 0.85 1.3 >20 5 62
Kuhnel and Hess [86] Capacitive 0.8 0.8 16 0.410 >25
Scheeper et al [87] Capacitive 1.5 1.5 14 2 35
Bourouina et al [88] Capacitive 11 2.5 3.5 35
Ried et al [89] Piezoelectric 2.5 2.5 18 0.92 57
Kalvesten [39] Piezoresistive 0.1 0.1 25 0.0009 at 10 V 90
Scheeper et al [28] Capacitive 22 14 5 30
Bergqvist [32] Capacitive 22 >17 11 30
Schellin et al [22] Piezoelectric 11 10 0.1 60
Kovacs and Stoffel [90] Capacitive 0.5 0.5 20 0.065 58
Kronast et al [91] FET 0.5 0.5 25 0.01 69

2.2.4. Comparative overview. Table 1 gives an overview wavenumbers, the largest and most energetic eddies occur,
of the most significant dimensions, measured sensitivity, and normally there are no problems associated with resolving
noise level and high-frequency response of first-generation these eddies. Basically, this is a question of having access to
MEMS pressure sensors. The following remarks can be computers with sufficiently large memory to store the amount
made from inspection of the data. The piezoelectric sensors of data that may be necessarily acquired from a large number
seem to have the highest noise levels. Although simple in of distributed probes, each collecting data for a time period
design, they have fairly low sensitivity and relatively large long enough to reduce the statistical error to a prescribed
diaphragm spatial extension. The piezoresistive sensors seem level. At the other end of the spectrum, however, both the
to be most flexible since the major advantage of locating spatial and the temporal resolutions are crucial, and this puts
piezoresistive gauges on a diaphragm is the high sensitivity severe limitations on the sensors to be used.
that is accomplished with this arrangement. The drawback It is possible to obtain a relation between the small and
is the high temperature sensitivity, although this problem large scales of the flow by substituting the inviscid estimate
can be circumvented by a Wheatstone bridge arrangement of the total dissipation rate into the expressions for the
of the gauges. The capacitive sensors are most common Kolmogorov microscales of length , time and velocity
since these sensors have the largest commercial interest and (see Tennekes and Lumley [1] and Kolmogorov [34, 35].
are mainly designed to operate in conventional microphone Thus,
applications such as music and human communications with  3/4
u`
pressure amplitudes as low as 20 Pa. The capacitive sensors = Re3/4 (4)
`
have relatively large spatial extension and are commonly
equipped with on-chip electronics for signal amplification.  1/2
u u`
This implies a reduction in noise levels but unfortunately also = Re1/2 (5)
`
necessitates a very complex fabrication process. The main
 1/4
advantages of capacitive sensors are their high sensitivity to v u`
pressure and low sensitivity to temperature. Their primary = Re1/4 (6)
u
disadvantages are the relatively large diaphragm areas and
the decreased sensitivity for high frequencies due to the air- where Re is the Reynolds number based on the speed of
streaming resistance of the narrow air gap inherent in the the energy containing eddies u and their characteristic length
principle. `. Since turbulence is a high-Reynolds-number phenomena,
these relations show that the small length, time and velocity
2.3. Requirements for turbulent flows scales are much less than those of the larger eddies, and that
the separation in scales widens considerably as the Reynolds
The subject of turbulence is both complex and vast. A number increases. Moreover, this also implies that the
good primer for readers unfamiliar with its richness is the assumptions made on the statistical independence and the
book by Tennekes and Lumley [1]. It is the ultimate goal dynamical equilibrium state of the small structures will be
of all measurements in turbulent flows to resolve both the most relevant at high Reynolds numbers. Another interesting
largest and smallest eddies that occur in the flow. At lower conclusion that can be drawn from the above relations is that

669
L Lofdahl and M Gad-el-Hak

if two turbulent flow fields have the same spatial extension simulations is that a sensor spacing of 35 Kolmogorov
(i.e. same large scale) but different Reynolds numbers, there lengths is recommended. When designing arrays for
would be an obvious difference in the small-scale structure in correlation measurements the spacing between the coherent
the two flows. The low-Reynolds-number flow would have structures will be the determining factor. For example, when
a relatively coarse small-scale structure, while the high-Re studying the low-speed streaks in a turbulent boundary layer,
flow would have much finer small eddies. several sensors must be situated along a lateral distance of
To spatially resolve the smallest eddies, sensors which 100 wall units, the average spanwise spacing between streaks.
are of approximately the same size as the Kolmogorov All this requires quite small sensors, and many attempts have
length scale for the particular flow under consideration are been made to meet these conditions with conventional sensor
needed. This implies that as the Reynolds number increases designs. However, in spite of the fact that conventional
smaller sensors are required. For instance, in the self- sensors like hot-wires have been fabricated in the micrometre
preserving region of a planecylinder wake at a modest size range (for their diameter but not their length), they are
Reynolds number (based on the cylinder diameter) of 1840, usually hand-made, difficult to handle and are too fragile. It
the value of varies in the range of 0.50.8 mm (Aronson is here that MEMS technology has really opened a door for
and Lofdahl [36]). For this case, conventional hot-wires new applications.
can be used for turbulence measurements. However, an There is no simple way to calculate the required pressure
increase in the Reynolds number by a factor of ten will range of a sensor for turbulence and flow control applications.
yield Kolmogorov scales in the micrometre range and call Tennekes and Lumley [1] estimate the fluctuating pressure
for either extremely small conventional hot-wires or MEMS- to be a weighted integral of the Reynolds stresses, so its
based sensors. Another illustrating example of the Reynolds length scales should in general be larger than those of the
number effect on the requirement of small sensors is a simple velocity fluctuations. Moreover, it is plausible to assume
two-dimensional, flat-plate boundary layer. At a momentum that the order of magnitude of the fluctuating pressure would
thickness Reynolds number of Re U / = 4000, the not be less than the Reynolds stresses, giving a good hint
Kolmogorov length scale is typically of the order of 50 m, of the intensity of the fluctuating pressure. This intensity
and in order to resolve these scales it is necessary to have of course depends strongly on the particular flow under
access to sensors that have a characteristic active measuring consideration, but for a typical flat-plate boundary layer at
length of the same spatial extension. say Re = 4000, this implies that the fluctuating pressure rms
Severe errors will be introduced into the measurements would be of the order of 10 Pa. Higher Reynolds numbers
by using too large a sensor, since such a sensor will integrate yield higher magnitudes of the fluctuating wall pressure. As
the fluctuations due to the small eddies over its spatial compared to other applications, for example in combustion
extension, and the energy content of these eddies will be processes where the fluctuating pressure is of major interest,
interpreted by the sensor as an average. When measuring the pressure magnitudes in incompressible turbulent flows
fluctuating quantities, this implies that these eddies are are extremely small. There is also a spatial constraint in the
counted as part of the mean flow and their energy is lost. The sensor design. For the same boundary layer considered here,
result will be a lower value of the turbulence parameter, and we have a Kolmogorov length scale of about 50 m, requiring
this will be wrongly interpreted as a measured attenuation of a diaphragm size in the range of 100300 m. The required
the turbulence (see for example Ligrani and Bradshaw [37]). temporal resolution of the pressure sensor is probably the
However, since turbulence measurements deal with statistical easiest to estimate since we have access to good turbulence
values of fluctuating quantities, it may be possible to loosen kinetic energy spectra and these show that the energy content
in a flow above 10 kHz is almost negligible. Based on these
the spatial constraint of having a sensor of the same size as
physical arguments, a framework for the design of a pressure
, to allow sensor dimensions which are slightly larger than
transducer for turbulence applications can be established,
the Kolmogorov scale, say of the order of .
namely that the sensor should have a pressure sensitivity of
For boundary layers, the viscous wall unit has been used
10 Pa, a diaphragm size of 100 m, and a flat frequency
to estimate the smallest possible sensor size for accurately
characteristics in the range of 10 Hz10 kHz. Of course, the
resolving the smallest eddies. For instance Keith et al
signal-to-noise ratio must be sufficiently high so that ordinary
[8] state that ten wall units or less is a relevant sensor
data acquisition can be performed.
dimension for resolving small-scale pressure fluctuations.
Critical scrutiny of the different principles used for
Measurements of fluctuating velocity gradients, essential for
design a pressure sensor discussed in section 2.2 reveals that
estimating the total dissipation rate in turbulent flows, are
the most suitable principle for turbulence applications is the
another challenging task. Gad-el-Hak and Bandyopadhyay
piezoresistive sensor, since the required spatial and temporal
[16] argue that turbulence measurements with probe lengths
resolutions can easily be achieved, simple fabrication in
greater than the viscous sublayer thickness (about five wall
MEMS is possible and the temperature drift can be controlled.
units) are unreliable, particularly near the surface. Many
In the next section we outline the design procedure of such
studies have been conducted on the spacing between sensors a piezoresistive pressure transducer for turbulence and flow
necessary to optimize the formed velocity gradients (see control applications.
Aronson et al [38] and references therein). A general
conclusion from both experiments and direct numerical
2.4. Piezoresistive pressure sensors
This is the inner length scale in a wall-bounded flow and is given by
the ratio of the fluid kinematic viscosity to friction velocity; denoted by + Micromachined piezoresistive pressure sensors were intro-
superscript. duced to the market in the late 1950s by companies like

670
MEMS-based pressure and shear stress sensors for turbulent flows

(streamwise) hL i and transverse (spanwise) hT i directions


can, according to Guckel [40], be expressed as

1R/R = Gpar hL i + Gper hT i (7)

where Gpar and Gper are the longitudinal and transverse


gauge factors. The pressure sensors fabricated by Kalvesten
[39] use two active strain gauges positioned at the points of
maximum stress on top of the diaphragm (off the neutral line),
and the other two are integrated beside the diaphragm to serve
as passive reference gauges (see figure 5). This arrangement
makes it possible to connect them in a half-active Wheatstone
bridge for high-pressure and low-temperature sensitivity.
Figure 4. Isometric view of a pressure sensor/microphone for The expected theoretical acoustical pressure sensitivity,
turbulence applications. The diaphragm side length is 100 m and
its thickness is 0.4 m. From Lofdahl et al [45]. S, for small pressure amplitudes has been derived by
Kalvesten [39] and reads
1 Vout
Kulite, Honeywell and MicroSystems. These early transduc- S= 0.077Gpar (1 2 )
ers were all very simple devices and typically the piezoresis- Vin p
 2 
tive gauges were glued by hand on the diaphragm. As the Et 3(1 + )E0 1 (1 2 )a 2
+ + (8)
technology developed, more advanced fabrication methods a2 4 16 2 td
were introduced, and today pressure transducers are all fab- where Vin and Vout are respectively the bridge supply and
ricated using batch processing technologies where no hand output voltages, p is the pressure acting on the diaphragm,
assembly is required. A large variety of different pressure is the isothermal compressibility coefficient of air, d is the
transducers are commercially available and the range of ap- cavity depth and Gpar is the longitudinal gauge factor of the
plications is very wide, spanning from relatively slow but polysilicon piezoresistor. For the polysilicon diaphragm,
stable tools for meteorological observations to sensors used is Poissons ratio, E is Youngs modulus, 0 is the built-in
to optimize rapid combustion processes. However, examina- strain, a is the side length and t is the thickness. Assuming
tion of these commercial sensors with turbulence and control a linear diaphragm deflection, this equation can be used to
applications in mind indicates that very few, in fact almost derive the static pressure sensitivity of the sensor. In doing so
none, can yet be applied for these specific purposes. If the it is necessary to assume a vacuum-sealed cavity by setting
sensors have either a frequency response and sensitivity that the third term of the denominator (air gap compressive forces)
is acceptable, then the sensing area is far too large and the to zero, see Rossi [41]. Table 2 gives a comparison of the
necessary spatial resolution is not fulfilled. Or vice versa, if measured and calculated characteristics of the microphone
the diaphragm size is acceptablewhich is very rarethen developed by Kalvesten [39].
either the frequency response or the sensitivity are far too A piezoresistive sensor is a combination of mechanical
low for the fulfillment of, respectively, the required temporal (the small and thin diaphragm), acoustical (the air gap
resolution or an acceptable signal-to-noise ratio. A literature cavity, the air in front of the diaphragm and the narrow vent
review confirms this statement. The sensors listed in table 1 channel) and electrical (polysilicon piezoresistors) elements.
are fabricated using MEMS technology and fall in the range In order to optimize the dimensions of the sensor, an
of having a certain interest for turbulence and flow control electrical analogy model is very useful for the task, see
applications. Rossi [41]. The important advantage of this analogyfrom
Based on the criteria discussed in section 2.3, it may which the microphone frequency response and the diaphragm
be concluded that a piezoresistive sensor for turbulence centre deflection can be derivedis that it allows a good
applications should be fabricated with a diaphragm size of overview of even very complex systems. The analogous
100 100 0.4 m3 . In order to give the pressure sensor model relies on the fact that there is a relationship between
the necessary sensitivity, the diaphragm must be very thin, the geometry of the mechano-acoustical elements and their
and the air gap behind the diaphragm (the cavity) must be equivalent electrical impedance. Hence, pressure p (N m2 )
relatively deep (say, 4 m) to reduce the air stiffening effect. is equivalent to voltage, flowrate q (m3 s1 ) to current, and
For equalization of the static pressure between the cavity and acoustic mass M (kg m4 ) to inductance. Damping in the
the ambient, a narrow vent channel with a length of 3 mm system, which is due to the viscosity of the gas, is described
and a cross-section of 5 0.1 m2 must be included in the by resistances R (N s m5 ) and capacitors C (m5 N1 ).
design. An isometric view of an acoustic sensor built to those These quantities then describe the diaphragm flexibility
specifications by Kalvesten [39] is shown in figure 4. and the gas compressibility. However, a limitation of this
When a diaphragm is deflected by a pressure load, stress equivalent circuit method is that analytical expressions for
is induced in the diaphragm and this should be detected by the equivalent electrical impedance are available only for
piezoresistive strain gauges located on top of the diaphragm simple mechanical and acoustical structures. One possibility
at appropriate positions off the neutral line. Noting that the for the derivation of the equivalent impedance has been
geometrical piezoresistive effect in semiconductors can be demonstrated by Kalvesten [39] and Kalvesten et al [4244],
neglected, the general relation between the relative change where an energy method is used for the calculation of the
of resistance and the strain components in the longitudinal energy and power contributions of the different elements.

671
L Lofdahl and M Gad-el-Hak

Figure 5. Top view of the piezoresistor layout with the electrical Wheatstone bridge configuration. From Lofdahl et al [45].

Table 2. Some calculated and measured characteristics of the MEMS-based microphone fabricated by Kalvesten [39].
Theory Measurements
Diaphragm side length, a (m) 100 300 100 300
Diaphragm thickness, t (m) 0.4 0.4 0.4 0.4
Lower frequency limit (Hz) 2 103 5 106 10a 10a
Upper frequency limit (kHz) 1310 894 25a 25a
Acoustic sensitivity at 10 V (V Pa1 ) 1.0 0.2 0.9 0.3
Static sensitivity at 10 V (V Pa1 ) 1.7 2.2 1.2 1.4
Noise (A-weighted) (dB(A)) 90 90
a
Limits set by the calibration measurement set-up.

the approximations made in the sensitivity calculation and


the uncertainty of the level of the built-in stresses in the
polysilicon diaphragm. As can be seen from table 2, and
as is further elucidated in Kalvesten et al [43], the required
frequency range for turbulence applications of 10 Hz < f <
10 kHz is expected to be covered based on the theoretical
range of about 1 Hz < f < 1 MHz. The noise levels of
the sensors are nearly the same for the two sensor versions,
corresponding to an acoustic level of about 90 dB(A). More
Figure 6. An electrical analogy describing the equivalent details on this sensor can be found in Kalvesten [39], or
acoustical circuit of the pressure sensor/microphone. From Lofdahl et al [45].
Lofdahl et al [45].

2.5. MEMS pressure sensors


These expressions are then identified as equivalent electrical
impedance. Figure 6 shows a simplified circuit diagram of Genuine MEMS sensors for turbulence measurements have
their microphone. been introduced only in this decade and accordingly very few
The piezoresistive pressure probe fabricated by reliable measurements for studying wall pressure fluctuations
Kalvesten [39] has perhaps the smallest diaphragm built beneath turbulent boundary layers have been conducted.
specifically for turbulence diagnosis. Its spatial resolution Otherwise, many MEMS devices have been launched, but the
is adequate to measure the small scales, and the small fluid dynamics validation of these devices have been made at
diaphragm has a correspondingly high resonance frequency a surprisingly low level, typically in some undefined channel
which leads to a wide frequency bandwidth. The main or pipe flow.
drawbacks of this small sensor are its relatively low acoustical Schellin et al [22] presented a subminiature microphone
sensitivity and high equivalent noise level. It should be noted, which was based on the piezoresistive effect in polysilicon
however, that for turbulence applications, the pressure is using only one chip. This sensor was classified as an
of the order of 1 Pa, while pressure amplitudes as low as acoustical sensor which was fabricated with a CMOS process
20 Pa are encountered in human communication and music and standard microfabrication technology. The diaphragm
applications. area was one square millimetre and the sensitivity was
The frequency characteristics of Kalvestens pressure 0.025 mV Pa1 at 6 V, so for turbulence applications this
sensor/microphone [39] were determined by using a Bruel sensor is a bit on the large side as well as too insensitive. The
& Kjaer Type 4226 closed coupler with a built-in reference frequency response was determined to be flat from 100 Hz
microphone at a sound pressure level of 114 dB. All to 5 kHz. Unfortunately, no fluid dynamics measurements
measurements were performed for a bridge supply voltage have been reported with this sensor.
of 10 V, and show acoustical sensitivities of about 0.9 Kalvesten [39], Kalvesten et al [4244] and Lofdahl
and 0.3 V Pa1 for diaphragm side lengths of 100 and et al [45] have designed, fabricated and used a silicon-
300 m respectively. The deviations from the theoretical based pressure transducer for studies focused on the high-
estimates for the acoustical sensitivities can be explained by frequency portion of the wall pressure spectrum in a

672
MEMS-based pressure and shear stress sensors for turbulent flows

Figure 8. Local advection velocity of the pressure producing


eddies for various frequency ranges. Uncertainty less than 10%.
From Lofdahl et al [45].
Figure 7. Dependence of normalized rms pressure fluctuations on
Reynolds number. The numbers next to selected data points
indicate the diaphragm side length in wall units. From Lofdahl Each sensor element consists of a FabryPerot etalon
et al [45]. fabricated from single-crystal silicon. Illuminating the
sensor array with a near-infrared tunable diode laser and
two-dimensional, flat-plate boundary layer. The momentum detecting the reflected light intensity with an infrared
thickness Reynolds number in their study was Re camera, allow remote, simultaneous reading of the reflected
U / = 5072. A large value of the ratio between the signals from the entire sensor array. The results obtained
boundary layer thickness and the diaphragm side length d demonstrate the basic feasibility of optically measuring
was used. The side lengths of the diaphragm were 100 m surface pressure using micromachined FabryPerot pressure
(d + = 7.2) and 300 m (d + = 21.6). This gives a ratio sensor arrays. Future work in this context will concentrate on
of the boundary layer thickness (outer length scale) to the demonstrating temporally resolved pressure measurements.
diaphragm side length of the order of 240 and a resolution
This technology is interesting since it offers the possibility
of eddies with wavenumbers less than ten viscous wall
of putting arrays on curved surfaces.
units. Power spectra were measured for the frequency range
13 Hz < f < 13 kHz, and these were scaled in outer and
inner variables. A clear overlap region between the mid- 2.6. Outlook for pressure sensors
and high-frequency parts of the spectrum was found, and
MEMS pressure transducers have certainly opened many
in this region the slope was estimated to be 1 . For the
doors in formulating relationships between wall pressure
high-frequency region, the slope was proportional to 5 .
and turbulence. In this closing subsection we focus on
The normalized rms pressure fluctuations were shown to
depend strongly on the dimensionless diaphragm size with some issues that should be addressable in the near future by
an increase connected to the resolution of the high-frequency further miniaturization and array arrangements of pressure
region as shown in figure 7. Classical data in the field are transducers. The Reynolds number effect on the rms values
also plotted in the same figure for comparison. of the wall pressure is one key issue. The data shown
Correlation measurements in both the longitudinal and in figure 7 seem to suggest a trend that is also depicted
transverse directions were performed by Lofdahl et al [45]. schematically in the figure, namely that they come from a
Longitudinal spacetime correlations, including the high- family of d + . In a review of Reynolds number effects, Gad-
frequency range, indicated an advection velocity of the order el-Hak and Bandyopadhyay [16] concluded that meaningful
of half the freestream velocity as can be seen in figure 8. trends in the inner layer can be extracted only when the
In this figure, Uc is the advection velocity of the pressure- sensors are of the order of the viscous sublayer thickness.
producing eddies, U is the freestream velocity, x1 is the At all Reynolds numbers in the figure, asymptotic values are
streamwise distance between two probes and is the local reached by sensors approaching the size d + = 5, and the slope
displacement thickness. The advection velocities computed of the family of d + lines drops as d + values are increased,
by Lofdahl et al [45] are consistently lower than those indicating a decreasing sensitivity of sensors to Reynolds
obtained by Willmarth and Wooldridge [10] and Bull [11]. A number effects. It should be noted that the data shown in
broad-band filtering of the longitudinal correlation confirmed figure 7 come from various sources with varying degrees of
that the high-frequency part of the spectrum is associated background noise, differences in data acquisition, corrections
with the smaller eddies from the inner part of the boundary applied, and different inherent errors in the sensors and
layer, resulting in a reduction of the correlation, as shown in instrumentation. Hence, the uncertainties in each data set
figure 9. are different and probably not known accurately. There are
Recently, Miller et al [46] have designed, fabricated also gaps in the data, so a useful contribution would be a
and tested a unique micromachined sensor array for making systematic variation of the sensor dimension and Reynolds
optical measurements of surface pressure distributions. number so the data of figure 7 could be completed.

673
L Lofdahl and M Gad-el-Hak

Using arrays of wall pressure sensors will offer


possibilities of more extensive studies of spacetime
correlations of pressure fluctuations. The primary
contribution of small transducers is that the eddies of the
flow field can be resolved with a spatial resolution of about
five viscous units, i.e. the thickness of the viscous sublayer.
Based on the measured correlations, it is possible to examine
the advection velocities of the different wavenumbers. It has
been shown earliere.g. by Bull [11]that the advection
velocities of turbulent eddies based on broad- and narrow-
band frequency analysis approach 80% and 60% of the
freestream velocity respectively. This can be questioned
since in all earlier investigations sensors that were too large
were used. In other words, the ratio between the boundary
layer thickness and the characteristic dimension of the sensor
was too low. With MEMS sensors, values of this ratio about Figure 9. Longitudinal spatial correlation of the pressure
ten times larger than those of conventional sensors can be fluctuations in three frequency ranges, 30 Hz < f < 1000 Hz,
realized. Having access to well designed pressure sensor 300 Hz < f < 13 kHz, and 13 Hz < f < 13 kHz. Momentum
arrays, the advection velocities for the different wavenumbers thickness Reynolds number is Re = 5072. From Lofdahl et al
[45].
can be determined. Such data also offer the possibility to
examine whether the structures are produced by pressure
sources inside the boundary layer with a wide range of stress studies conducted rely on the premise that the mean
advection velocities. The so-called two-family concept velocity gradient and the heat transfer rate near or at the wall
suggested by Bull [11] can then be scrutinized. It was are both proportional to the wall shear stress. The former
claimed by Bull [11] that one family of high wavenumber relation was established in the beginning of the 1920s by
components was associated with the turbulent motion in the Stanton et al [47] and the latter a few years later by Leveque
constant stress layer. Those components were longitudinally [48].
coherent for times proportional to the time needed for the Over the last few decades, numerous experiments have
structure to be advected distances equal to their wavelengths, been conducted where measurements of wall shear stress
and laterally coherent over distances proportional to their own have been the kernel. The success of those endeavours
wavelengths. The other wavenumber family was associated depends basically on the complexity of the flow, the geometry
with large-scale eddy motion in the boundary layer, and loses of the solid boundaries and the limitations of the measuring
coherence as a group independent of the wavelength. To device used. Some facts have been established, like the wall
this end, it is not possible to question Bulls hypothesis, shear stress distribution along flat plates and simple bodies of
mainly because of the lack of small pressure transducers and revolution, but in general our knowledge of wall shear stress
associated arrays. and in particular its fluctuating component is limited. An
The concepts discussed above in relation to MEMS- explanation of this is that it is both difficult and cumbersome
based pressure sensors are all key issues and significant to measure wall shear stress. In general it is a parameter
to aeroacoustics, energy transport in vibrating structures of small magnitude. Some typical values are worth bearing
interacting with turbulent flows, turbulence modelling of in mind for the following discussion. A submarine cruising
PVC terms, and last but certainly not least to reactive control at 30 km h1 has an estimated value of the shear stress of
of the flow. about 40 Pa; an aircraft flying at 420 km h1 , 2 Pa; and a
car moving at 100 km h1 , 1 Pa. Such small forces per unit
area put heavy demands on the sensitivity of the measuring
3. Wall shear stress sensors
devices used. Those estimates are approximate and were
obtained by Munson et al [49] who treated the body as a
3.1. Background
collection of parts, for instance the drag of an aircraft was
From scientific and engineering perspectives the wall shear approximated by adding the drag contributions produced by
stress is an essential quantity to compute, measure or infer wings, fuselage, tail, etc.
in a wall-bounded turbulent flow. Time-averaged values Many different methods for the determination of wall
of this quantity are indicative of the global state of the shear stress have been developed, and the required spatial
flow along a surface and can be used to determine body- and temporal resolutions are specific for each application
averaged properties like the skin friction drag. The time- and environment. In laminar flows the sensors must be
resolved part of the wall shear stress is a measure of the capable of measuring the time-averaged shear stress, while
unsteady structures in the flow which are responsible for the in turbulent flows both the mean shear and its fluctuating
individual momentum transfer events and is an indicator of component are of interest. An attempt to classify the
the coherent portion of the turbulence activities. Spatially techniques available for wall shear stress measurements was
distributed values of the instantaneous wall shear stress can made by Haritonidis [50], who divided the technologies into
be used in a feedforward or feedback control loop to effect direct and indirect methods, depending upon whether the wall
beneficial changes in the boundary layer. Most wall shear shear stress is measured directly or is inferred from other

674
MEMS-based pressure and shear stress sensors for turbulent flows

Figure 10. Classification of wall shear stress measurements techniques. After Haritonidis [50].

measured properties. Figure 10 shows a schematic diagram the amount of energy necessary for keeping a wall-mounted,
of Haritonidis classification. electrically heated resistor at constant temperature, despite a
Since the number of different methods is enormous, it time-dependent, convective heat transfer to the flow. A high
is beyond the scope of this work to review each technique. freestream velocity yields steep velocity and temperature
A trend common to all measurements and methods can, gradients at the heated surface and thereby a good cooling
however, can be highlighted. Since the mid 1950s, the which requires more energy, while the opposite is true for
evolution of the probes used has been directed towards a low freestream velocity. In the floating element method,
utilizing smaller and smaller sensors in order to improve a flush-mounted wall element, movable in the plane of the
the accuracy, flexibility and resolution. There are several wall shear stress, is displaced laterally by the tangential
review papers that describe shear stress sensors and discuss viscous force and this movement can be measured using
in detail the merits and drawbacks of the methods used in a either resistive, capacitive or optical detection principles.
vast variety of flow situations. To cite a few, Winter [51] gives In this section, we describe the principles of the thermal
a comprehensive review of available conventional methods sensor, since it is particularly suitable for measurements of
and a good discussion of measurements in turbulent flows. fluctuating wall shear stress as well as for MEMS fabrication.
Haritonidis [50] summarizes conventional methods and In the description we briefly recall the classical equations
introduces the first micromachined wall shear stress sensor. and based on these we highlight possible uncertainties and
Hakkinen [52] lists the merits and drawbacks of conventional sources of errors in measuring instantaneous wall shear stress.
techniques. Most recently, Hanratty and Campbell [53] Additionally, we review current MEMS-based sensors. Since
discuss the relevant experimental issues associated with the
calibration is a crucial part of wall shear stress measurements,
use of various wall shear stress sensors.
particularly in the determination of fluctuating quantities,
A major problem in wall shear stress measurements
we summarize some common calibration methods and
is that conventional fabrication of sensors and balances
discuss possibilities for dynamic calibration. We provide a
allows only a certain degree of miniaturization since these
brief discussion on floating element sensors for measuring
devices are more or less hand-made. Better spatial and
instantaneous wall shear stress. Finally, we outline the
temporal resolutionsthan those provided by conventional
outlook for the use of MEMS-based wall shear stress sensors
probesare needed in typical turbulence studies and reactive
for turbulence diagnosis and flow control.
flow control. MEMS has the potential of circumventing
conventional sensor limitations, since microfabrication offers
a high degree of miniaturization and associated increased 3.2. Thermal sensor principle
resolution together with the fact that the sensors are not hand-
made, implying that each unit is fabricated to extremely small The thermal or hot-film sensor benefits from the fact that the
tolerance at reasonable cost. Table 3 lists the operation and heat transfer from a sufficiently small heated surface depends
detection principles as well as the active sensor area of some only on the flow characteristics in the viscous region of the
recently developed MEMS-based shear stress probes. boundary layer. A schematic drawing of a thermal sensor is
A glance at the classification scheme of figure 10 shows provided in figure 12. The hot-film gauge consists of a thin
that the floating element principle and the heat transfer metallic film positioned into a substrate. Usually the gauge
method (or more correctly thermal element method) are forms one part of a Wheatstone bridge, and an electric current
both well suited to MEMS fabrication. Figure 11 shows a is passed through the film in order to maintain it at a constant
schematic diagram of the principle of operation of the two temperature as heat is continuously being transferred from
techniques. The thermal element method relies on measuring the film to the moving fluid as well as to the films substrate.

675
L Lofdahl and M Gad-el-Hak

Table 3. Operation and detection principle of silicon micromachined wall shear stress sensors (in chronological order of development).

Operation Detection principle Active sensor area (mm2 )


Oudheusden and Huijsing [92] Thermal Thermopile 43
Schmidt et al [73] Floating element Capacitive 0.5 0.5
Ng et al [74] Floating element Piezoresistive 0.12 0.14
Liu et al [27] Thermal Anemometer 0.2 0.2
Pan et al [76] Floating element Capacitive 0.1 0.1
Padmanabhan et al [77] Floating element Optical 0.12 0.12
Kalvesten [71] Thermal Anemometer 0.3 0.06

Figure 11. Wall shear stress sensors of the hot-film and floating element type.

Figure 12. Illustration of a surface mounted hot-film sensor.

The ohmic (Joule) heating in the device QJ is transferred This expression has been derived by many researchers, e.g.
both to the fluid and to the surrounding substrate. Ludwieg [54], Liepmann and Skinner [55], and Bellhouse
This can be expressed as and Schultz [56]. This simple relation breaks down, however,
QJ = Qf + Qs (9) in unsteady and turbulent flows as has been pointed out by
Bellhouse and Schultz [56]. In a turbulent environment, most
where Qf represents the heat transferred to the fluid directly assumptions made in the derivation are violated and the heat
from the heated surface (Qf1 ) and indirectly through the conduction to the substrate changes instantaneously. In spite
heated portion of the substrate (Qf2 ), and Qs represents the of this, we will use an extended version of equation (10)
heat lost irretrievably to the substrate. If the flow is steady to illustrate the assumptions made and to pinpoint possible
and laminar and if the streamwise pressure gradient can be
pitfalls in using hot-film gauges for fluctuating wall shear
neglected, the resulting heat transfer rate to the fluid Qf has
stress measurements. In the following argument we will
been shown to be related to the wall shear stress according to
follow the path outlined by Bellhouse and Schultz [56] and
the classical relation
later by Menendez and Ramaprian [57], who derived an
w Q3f . (10) extended version for boundary layer flows subjected to a

676
MEMS-based pressure and shear stress sensors for turbulent flows

periodic freestream velocity of the form obtained as a function of the heat transfer rate from the wall
to the fluid
Ua (t) = U0 (1 + sin t) (11)  
cos t Qf
w = A1 Q3f + A2 + A3 (17)
where is the the angular frequency of the oscillation, U0 is Qf t
the mean freestream velocity, and is the relative amplitude
of the oscillation. The derivation below is highly abbreviated, where
L
and the reader is referred to Menendez and Ramaprian [57] A1 (18)
for a more complete version. (1 n)acp f2 (1T0 )3
The problem to be considered is shown in figure 12. A bf (1T0 )aU0
thermal boundary layer develops within a turbulent boundary A2 (19)
2a
layer over a heated film having an effective streamwise length
L. The thermal boundary layer is produced by a sudden jump Lc
A3 . (20)
in the surface temperature, from a constant value equal to the (1 + n)af (1T0 )
ambient temperature Ta to the higher constant value Tw at In these expressions the shape parameters a, b and c
the location x = x0 . It is assumed that the thermal boundary are basically functions of the streamwise coordinate x.
layer is totally embedded in the viscous region, so effects According to Menendez and Ramaprian [57], equation (17)
of turbulent diffusion can be neglected. A relation between can be regarded as the basic relationship between the
the local wall shear stress w and the heat transfer rate from instantaneous value of the wall shear stress and the wall
the wall to the fluid Qf is derivable from the first law of heat flux. By taking ensemble average of equation (17) over
thermodynamics. For a two-dimensional flow, neglecting a large number of cycles and assuming that the turbulent
dissipation and compressibility effects and applying the usual fluctuations are small, a linearized expression for the periodic
boundary layer approximation, the energy equation reads wall shear stress is obtained.
To summarize, the classical expression, equation (10),
T T T f 2 T
+U +V = (12) gives a relation between the wall shear stress and the heat
t x y cp y 2 transfer from the wall. This expression assumes steady,
where T (x, y, t) is the desired temperature distribution, t is laminar, zero-pressure-gradient flow, and is not valid in
time, U and V are velocity components along the coordinates a turbulent environment. Menendez and Ramaprian [57]
x and y respectively and , f and cp are respectively the fluid derived an extended version of equation (10) valid for a
density, thermal conductivity and specific heat. Using the periodically fluctuating freestream velocity, equation (17).
equation of continuity and integrating equation (12) across The latter relation contains, however, some assumptions
the thermal boundary layer up to a point y = ye beyond the which are questionable for turbulent flows. For instance, in
edge of the thermal boundary layer yields the thermal boundary layer it is assumed that the temperature
distribution is self-similar and that the local thickness varies
Z ye Z ye
1 linearly. These assumptions are relevant for a streamwise
(T Te ) dy + U (T Te ) dy = Qf
t 0 x 0 cp velocity oscillation and a weak fluctuation, but certainly not
(13) in a turbulent flow which is strongly unstable in all directions.
where Te is the temperature at the edge of the thermal Following a brief look at available calibration methods, we
boundary layer. It is now assumed that within the viscous discuss the spatial and temporal resolutions of wall-mounted
region where the thermal boundary layer resides, inertial hot-films together with some MEMS-based sensors for wall
effects can be neglected. The velocity profile can be shear stress determination.
approximated by
  3.3. Calibration
w 1 p
U= y+ y2 (14)
2 x Several methods and formulae are in use for calibrating hot-
film shear probes operated in the constant-temperature mode
where p is the pressure and is the dynamic viscosity of the and the choice of method depends on flow conditions and
fluid. It is further assumed that the temperature distribution sensor used. In this section, two static calibration methods
in the thermal boundary layer is self-similar at any instant of are discussed. Both are based on the theoretical analysis
time, and that the thermal boundary layer thickness T can leading to the relation between rate of heat transfer and wall
be expressed as follows shear stress, as discussed in section 3.2. The challenge is of
T Te course to be able to use the shear stress sensor in a turbulent
= f (, t) (15) environment.
Tw Te
If a laminar flow facility is used to calibrate the wall
 
T (x, t) x x0 n shear stress sensor, then equation (10) can be re-written more
= (16) conveniently in time-averaged form
T (x0 + L, t) L
where y/T is the similarity coordinate and n is an w = Ae + B
1/3 2
(21)
unknown exponent which can in general depend on time.
After some algebra and assuming n is independent of time, where w is the desired mean wall shear stress, e2 is the
an expression for the local wall shear stress at x = x0 + L is square of the mean output voltage, and A and B are calibration

677
L Lofdahl and M Gad-el-Hak

constants. The term B represents the heat loss to the substrate Bellhouse and Rasmussen [63] and Bellhouse and Schultz
in a quiescent surrounding, and this procedure is similar [56] achieved this in two different ways. One method is
to a conventional calibration of a hot-wire [58]. However, to mount the hot-film on a plate, which can be oscillated
laminar flow is often difficult to realize in the desired range at various known frequencies and amplitudes. The main
of turbulence wall shear stress, and it is more practical to drawback to this arrangement is the limited amplitudes
perform the calibration without moving the sensor between and frequencies that can be achieved when attempting to
calibration site and measurement site. In that case, the vibrate a relatively heavy structure. An alternative strategy
calibration is made in a high-turbulence environment and the is to generate shear stress variations by superimposing a
high-order moments of the voltage must also be considered. monochromatic sound field of different frequencies on a
Ramaprian and Tu [59] proposed an improved steady, laminar flow field. Padmanabhan [64] described
calibration method, with the instantaneous version of how oscillating wall shear stress of a known magnitude and
equation (21) re-written and time-averaged to give frequency can be generated using an acoustic plane wave
tube. A schematic diagram of the calibration experiment
w = (Ae2 + B)3 = A3 e6 + 3A2 Be4 + 3AB 2 e2 + B 3 . (22) is shown in figure 13. The set-up comprises an acrylic
tube with a speaker/compression driver at one end and a
This can be written as wedge-shaped termination at the other end. The latter is
designed to minimize reflections of sound waves from the
w = C6 e6 + C4 e4 + C2 e2 + C0 (23)
tube end and thereby set up a purely travelling wave in the
where the Cs are the new calibration constants. The high- tube. A signal generator and an amplifier drive the speaker
order moments of the voltage can easily be calculated with to radiate sound at different intensities and frequencies. At
a computer, but care must be taken that they are fully some distance downstream the waves become plane and
converged. The relation between e and w can also be at this location a condenser microphoneused to measure
represented by a full polynomial function, so if the frequency the fluctuating pressureand the shear stress sensor are
response of the sensor is sufficiently flat, it is possible to write mounted. The flow field inside the plane wave tube is
very similar to a classical fluid dynamics problemStokes
w = C0 + C1 e + C2 e2 + + CN eN . (24) second problem. The only difference is that instead of
an oscillating wall with a semi-infinite stationary fluid, the
When M is the number of calibration points and N is the plane wave tube has a stationary wall and oscillating fluid
order of the polynomial above, a linear equation system is particles far away from the wall. Solutions to the Stokes
obtained where the calibration coefficients can be computed problem can be found in many textbooks, e.g. Brodkey [65].
by a numerical least-squares method. The mean wall shear Padmanabhan [64] converted the boundary conditions and
stress on the left-hand side of equation (24) can be measured derived corresponding analytical expression for the plane
with, for example, a Preston tube using Patels method wave tube. The analytical solution of the fluctuating wall
[60]. The second calibration technique described here was shear stress was compared to the measured output of the shear
called stochastic calibration by Breuer [61], who also stress sensor as a function of frequency and a transfer function
demonstrated that the validity of the calibration polynomial of the sensor was determined. As expected, the measured
may extend well beyond the original calibration range, shear stress showed a square root dependence on frequency.
although this requires careful determination of the higher-
order statistics as well as a thorough understanding of the 3.4. Spatial resolution
sensor response function.
The experiments of Bremhorst and Gilmore [62] showed Equation (17) is an extended version of equation (10), and
that the static and dynamic calibration coefficients for by assuming a steady flow and zero-pressure-gradient in
hot-wires agree to within a standard error of 3% for the streamwise direction it reduces to the classical formula.
the velocity range of 3 to 32 m s1 . Thus, they The latter expression can be used to estimate the effective
recommended the continued use of static calibration for streamwise length of a hot-film
dynamic measurements. As pointed out in section 3.2, this
w = A1 Q3f . (26)
is not necessarily true for a wall-mounted hot-film. One
solution to this problem may be to calibrate the hot-film By assuming a stepwise temperature variation and
in pulsatile laminar flow with a periodic freestream velocity introducing the average heat flux q f over the heated area
Ua (t), and make use of the formula proposed by Menendez which is assumed to have a streamwise length L, the desired
and Ramaprian [57] relation reads
 
c1 dUa de2 w 1/3
w = (Ae2 + B)3 + + c2 A (25) q f = 0.807f 1Tw P r1/3 (27)
Ae2 + B dt dt L
to obtain the additional calibration constants c1 and c2 . This where P r is the Prandtl number. It is convenient to re-write
step is necessary in order to characterize the gauge dynamic the above expression into a non-dimensional form, and in
response at relatively high frequencies. The constants A and doing so we keep L explicit
B are obtained from a steady state calibration.
One difficulty in performing a dynamic calibration is hc L 2
Nu = 0.807(P rL+ )1/3 (28)
to generate a known sinusoidal wall shear stress input. f

678
MEMS-based pressure and shear stress sensors for turbulent flows

Figure 14. Pedleys relation [69] showing the region of validity of


the boundary layer solution as a function of Peclet number, P e. In
the area in between the two full curves, the boundary layer
approximation predicts the heat transfer accurately to within 5% of
the correct value.

figure 14, and it can be seen that Peclet numbers larger than
40 correspond approximately to 1/3 w > 0.95. A relatively
Figure 13. An acoustic plane wave tube for the calibration of a simple calculation of the upper limit of L+ can then be made
floating element shear sensor. (a) Overall experiment set-up; by assuming that the viscous sublayer is about five viscous
(b) probe region. From Padmanabhan [64]. units, and this yields an upper limit of L+ which for air can
be estimated to be approximately 47.
where N u is the Nusselt number averaged over the heated In summary then, it can be concluded that the streamwise
area, L+ is the streamwise length of the heated area extent of the hot-film cannot be too small, otherwise the
normalized to the viscous length scale /u , and hc is boundary layer approximations made are not applicable. On
the convective heat transfer coefficient. This equation was the other hand, a sensor that is too large will cause the
derived for flows with a pressure gradient by Brown [66]. thermal boundary layer to grow beyond the viscous region.
The dimensionless sensor length L+ is a crucial parameter Additionally, the spatial resolution will be adversely affected
when examining the assumptions made. if the sensor is too large, since the smallest eddies imposed
The lower limit on L+ is imposed by the boundary by the flow structures above the wall will then be integrated
layer approximation, since close to the leading and trailing along the sensor length.
edge zones of the heated strip, there is an abrupt change
of temperature where the neglected diffusive terms in 3.5. Temporal resolution
equation (12) become significant. Tardu et al [67] have
The temporal resolution of the thermal probe is affected by
conducted a numerical simulation of the heat transfer from
the different time constants of the hot-film and the substrate.
a hot-film, and found a peak of the local heat transfer at the
The hot-film usually has a much shorter time constant than
leading and trailing edges. They concluded that if the hot-film the substrate. The higher the percentage of the total heat
is too narrow, heat transfer would be completely dominated that leaks into the substrate, the lower is the sensitivity
by these edge effects. Ling [68] studied the same problem of the device to shear stress fluctuations, and this changes
in a numerical investigation and concluded that the diffusion the sensor characteristics sufficiently to invalidate the static
in the streamwise direction can be neglected if the Peclet calibration made. An example of this phenomenon was
number is larger than 5000. Pedley [69] also concluded given by Haritonidis [50] who showed that a hot-film sensor
that, provided 0.5P e0.5 < x/L < 1 0.7P e0.5 , there in a fluctuating wall shear stress environment will respond
exists a central part of the hot-film where the boundary layer quickly to the instantaneous shear stress while the substrate,
solution predicts the heat transfer within 5%. Figure 14 shows due to its much larger thermal inertia, will react slowly.
the relation proposed by Pedley [69], and it can be seen that Haritonidis also showed that the ratio of the fluctuating
the heat transfer is correctly described over a large part of sensitivity Sf to the average sensitivity Sa can be related
the hot-film area, but as the Peclet number decreases, the to the ratio of the effective lengths under dynamic and static
influence from the diffusive terms must be considered. For conditions  2/3
Peclet numbers larger than 40, the heat transfer is correctly Sf Lf
described by more than 80% of the heated area. = <1 (29)
Sa La
The upper limit of L+ is crucial since there the hot-film
thermal boundary layer may not be entirely submerged in where La is the average effective length during static
the viscous sublayer. Equation (26) has been included in calibration and Lf is the effective length during dynamic
calibration. Due to this, the hot-film becomes less sensitive
The Peclet number here is defined as the ratio of heat transported by to shear stress fluctuations at higher frequencies and the static
2
convection to that by conduction, P e = P rL+ (see Brodkey [65]). calibration in a laminar flow will not give a correct result.

679
L Lofdahl and M Gad-el-Hak

Figure 15. Flush-mounted wall shear stress sensor. (a) Schematic cross-section; (b) top view. From Kalvesten [71].

These length scales can be considerably larger than the true the flow. Haritonidis [50] computed this ratio for a number
extent of the probe. For example, Brown [66] reported that of fluids and films and the main conclusion drawn was that
the effective length scale from a static calibration was about at high frequencies, the substrate does not participate in the
twice the physical length. heat transfer process. However, this conclusion should be
Both the substrate material and the amount of heat that viewed with some caution since the frequencies studied were
is lost to the substrate are crucial when determining the the highest that could be expected in a wall flow.
temporal or frequency response of a hot-film sensor. At low The most difficult problem occurs for frequencies in the
frequencies, the thermal waves through the substrate and into intermediate range, resulting in a clear substrate influence and
the fluid are quasi-static, which means that the fluctuating an associated deviation from the static calibration. Hanratty
sensitivity of the hot-film is determined by the first derivative and Campbell [53] showed that damping by the thermal
of the static calibration curve. In this range, which is of boundary layer for pipe flow turbulence is important when
the order of a couple of cycles per second, the heat transfer
[L+ (5.65 102 )3 P r]1/2 6 1.
2
through the substrate responds without time lag to wall shear (30)
stress fluctuations. Basically, this frequency range does not
cause any major problem. For P r = 0.72, this requires the dimensionless length
For the high-frequency range at the other end of the in the streamwise direction L+ to be less than 90. For
spectrum, it is possible to estimate the substrates role turbulence applications, this is most disturbing since it is
by considering the propagation of heat waves through a in this frequency range where the most energetic eddies
semi-infinite solid slab subjected to periodic temperature are situated. The primary conclusion to be reached from
fluctuations at one end. This has been reported by this discussion is that all wall shear stress measurements in
Blackwelder [70] who compared the wavelength of the heat turbulent flows require dynamic calibration of the hot-film
wave to the hot-film length. He showed that the amplitude of sensor.
the thermal wave would attenuate to a fraction of a per cent
over a distance equal to its wavelength. A relevant quantity 3.6. MEMS thermal sensors
to consider in this context turned out to be the ratio of the
wavelength to the length of the substrate since this is an Kalvesten [71] has developed a MEMS-based, flush-mounted
indication of the extent to which the substrate will partly wall shear stress sensor with a principle of operation that
absorb heat from the heated surface and partly return it to relies on the cooling of a thermally insulated, electrically

680
MEMS-based pressure and shear stress sensors for turbulent flows

Figure 17. Cross-sectional structure of a single shear stress


sensor. From Jiang et al [72].

Figure 16. Total steady state power dissipation calibration as a


function of the wall shear stress for two different probe
orientations. From Kalvesten [71].

heated part of a chip. As depicted in figure 15, the heated


portion of the chip is relatively small, 300 60 30 m3 ,
and is thermally insulated by polyimide-filled, KOH-etched
trenches. The rectangular top area, with a side length to
side length ratio of 5:1, yields a directional sensitivity for the
measurements of the two perpendicular in-plane components
of the fluctuating wall shear stress. Due to the etch properties
of KOH, the 30 m deep, thermally insulating trenches have
sloped walls with a bottom and top width both of about
30 m. The sensitive part of the chip is electrically heated by
a polysilicon piezoresistor and its temperature is measured by Figure 18. SEM image of a single wall shear stress sensor. From
an integrated diode. For the ambient temperature, a reference Jiang et al [72].
diode is integrated on the substrate chip, far away from the
heated portion of the chip.
thickness 1.2 m and side length 200 m. The polysilicon
Kalvesten [71] performed a static wall shear stress
resistor wire, is located on the diaphragm, is 3 m wide and
calibration in the boundary layer of a flat plate. A Pitot tube
150 m long. Below the diaphragm there is a 2 m deep
and a Clauser plot were used to determine the time-averaged
vacuum cavity so that the device will have a minimal heat
wall shear stress. The power consumed to maintain the hot
conduction loss to the substrate. When the wire is heated
part of the sensor at a constant temperature was measured, and
electrically, heat is transferred to the flow by heat convection
figure 16 shows the data for two different probe orientations.
resulting in an electrically measurable power change which
For a step wise increase of electrical power, the response
is a function of the wall shear stress. Figure 18 shows
time was about 6 ms which is double the calculated value.
a photograph of a portion of the 2.85 1.00 cm2 streak-
This response was considerably shortened to 25 s when the
imaging chip, containing just one probe. The sensors were
sensor was operated in a constant-temperature mode using
calibrated in a fully developed channel flow with known
feedback electronics. Table 4 lists some calculated and
average wall shear stress values. The output of such a sensor
measured characteristics of Kalvestens MEMS-based wall
is sensitive to the fluid temperature and the measured data
shear stress sensor [71].
must be compensated for this effect.
Jiang et al [72] developed an array of wall shear stress
sensors based on the thermal principle. The primary objective
of their experiment was to map and control the low-speed 3.7. Floating element sensors
streaks in the wall region of a turbulent channel flow. The floating element technique is a direct method for sensing
Therefore, to properly capture the streaks, each sensor was skin friction, which means a direct measurement of the
made smaller than a typical streak width. For a Reynolds tangential force exerted by the fluid on a specific portion of
number (based on channel half-width and centreline velocity) the wall. The advantage of this method is that the wall shear
of 104 , the streaks were estimated to be about 1 mm in stress is determined without having to make any assumptions
width, so each sensor was designed to have a length less about either the flow field above the device or the transfer
than 300 m. Figure 17 shows a schematic diagram of function between the wall shear stress and the measured
one of these sensors [72]. It consists of a diaphragm of quantity. The sensing wall element is connected to a balance
Note that for back-up, two hot diodes and two cold diodes are fabricated that determines the magnitude of the applied force. Basically
on the same chip. two arrangements are distinguished to accomplish this: either

681
L Lofdahl and M Gad-el-Hak

Table 4. Some calculated and measured characteristics of the MEMS-based wall shear stress sensor fabricated by Kalvesten [71].
Theory Measurements
Heated chip top area, A = w ` (m ) 2
300 60 1200 600 300 60 1200 600
Heated chip thickness, h (m) 30 30 30 30
Thermal conduction conductance, Gc (W K1 ) 372 510 426 532
Thermal convection conductance, Gf (W K1 )
Perpendicular configuration at 50 m s1 6.7 207 21.0 231
Parallel configuration at 50 m s1 5.6 192 17.3 160
Thermal capacity (J K1 ) 1.2 37
Thermal time constant at zero flow (ms) 3 72 6 55
Time response (electronic feedback) (s) 25

direct measurement of the distance the wall element is moved


by the wall shear stress, so-called displacement balance;
or measurement of the force required to maintain the wall
element at its original position when actuated on by the wall
shear stress, the null balance.
The principle of a floating element balance is shown in
figure 11. In spite of the fact that the force measurement
is simple, the floating element principle is afflicted by
some severe drawbacks which strongly limit its use, as
has been summarized by Winter [51]. It is difficult to
choose the relevant size of the wall element in particular
when measuring small forces and in turbulence applications.
Misalignments and gaps around the elementespecially
when measuring small forcesare constant sources of
uncertainty and error. Effects of pressure gradients, heat
transfer and suction/blowing also cause large uncertainties
in the measurements. If the measurements are conducted Figure 19. Schematic diagram of the MEMS-based floating
element balance of Schmidt et al [73].
in a moving frame of reference, effects of gravity,
acceleration and large transients can also severely influence
the results. Haritonidis [50] discussed the mounting of
floating element balances and errors associated with the polyimide, and the sensor was designed to have a bandwidth
gaps and misalignments. In addition, floating element of 20 kHz. A static calibration of this force gauge indicated
balances fabricated with conventional techniques have in linear characteristics, and the sensor was able to measure a
general a poor frequency response and are not suitable for shear stress as low as 1 Pa. However, the sensor showed
measurements of fluctuating wall shear stress. To summarize, sensitivity to electromagnetic interference due to the high-
the idea of direct force measurements by a floating element impedance capacitance used, and drift problems attributed
balance is excellent in principle, but all the drawbacks taken to water vapour absorption by the polyimide were observed.
together make them difficult and cumbersome to work with in No measurements of fluctuating wall shear stress were made
practice. It was not until the introduction of microfabrication since the signal amplitude available from the device itself was
in the late 1980s that floating element force sensors achieved too low in spite of the fact that the first-stage amplification
a re-vitalized interest in particular for turbulence studies and was fabricated directly on the chip.
reactive flow control. Since the introduction of the Schmidt sensor [73], other
floating element sensors based on transduction, capacitive
and piezoresistive principles have been developed (Ng et al
3.8. MEMS floating element sensors
[74], Goldberg et al [75], Pan et al [76]). The sensor of
Schmidt et al [73] were the first to present a MEMS-based Ng et al [74] was small and had a floating element of size
floating element balance for operation in low-speed turbulent 120 140 m2 . It operated on a transduction scheme and
boundary layers. A schematic diagram of their sensor is was basically designed for polymer extrusion applications
shown in figure 19. A differential capacitive sensing scheme so it operated in a shear stress range of 1100 kPa. The
was used to detect the floating element movements. The area sensor of Goldberg et al [75] had a larger floating element
of the floating element used was 500 500 m2 , and it was size, 500 500 m2 . It had the same application and the
suspended by four tethers which acted both as supports and same principle of operation as that of Ng et al [74]. Neither
restoring springs. The floating element had a thickness of of these two sensors is of interest in turbulence and flow
30 m and was suspended 3 m above the silicon substrate on control applications since their sensitivity is far too low. The
which it was fabricated. The gap on either side of the tethers capacitive floating element sensor of Pan et al [76] is a force
and between the element and surrounding surface was 10 m, re-balance device designed for wind tunnel measurements,
while the element top was flush with the surrounding surface and is fabricated using a surface micromachining process.
within 1 m. The element and its tethers were made of Unfortunately, this particular fabrication technique can lead

682
MEMS-based pressure and shear stress sensors for turbulent flows

Figure 20. Schematic diagram of the floating element balance of Padmanabhan [64].

to non-planar floating element structures. The sensor has A unique future of the shear sensor described above
only been tested in laminar flow and no dynamic response of is that its dynamic response has been experimentally
this device has been reported. determined to 10 kHz. Padmanabhan [64] described
Recently, Padmanabhan et al [77] presented a floating how oscillating wall shear stress of known magnitude and
element wall shear stress sensor based on optical detection frequency can be generated using an acoustic plane wave
of instantaneous element displacement. The probe was tube. The calibration experiment was discussed previously
designed specifically for turbulent boundary layer research in section 3.3 and shown schematically in figure 13.
and has a measured resolution of 0.003 Pa and a dynamic
response of 10 kHz. A schematic diagram illustrating 3.9. Outlook for shear stress sensors
the sensing principle is shown in figure 20. The sensor
comprises a floating element which is suspended by four Since coherent structures play a significant role in the
support tethers. The element moves in the plane of the chip dynamics of turbulent shear flows, the ability to control these
under the action of wall shear stress. Two photodiodes are structures will have important technological benefits such
placed symmetrically underneath the floating element, at the as drag reduction, transition control, mixing enhancement
leading and trailing edges, and a displacement of the element and separation delay. In particular, the instantaneous wall
causes a shuttering of the photodiodes. Under uniform shear stress is of interest for reactive control of wall-bounded
illumination from above, the differential current from the flows to accomplish any of those goals (Gad-el-Hak [78]).
photodiodes is directly proportional to the magnitude and An anticipated scenario to realize this vision would be to
sign of shear stress. Analytical expressions were used to cover a fairly large portion of a surface, for instance parts
predict the static and dynamic response of the sensor and, of an aircraft wing or fuselage, with sensors and actuators.
based on the analysis, two different floating element sizes Spanwise arrays of actuators would be coupled with arrays
were fabricated, 120 120 7 m3 and 500 500 7 m3 . of wall shear stress sensors to provide a locally controlled
The device was calibrated statically in a laminar flow over region. The basic idea is that sensors upstream of the
a stress range of four orders of magnitude, 0.00310 Pa. actuators detect the passing coherent structures, and sensors
The gauge response was linear over the entire range of wall downstream of the actuators provide a performance measure
shear stress. The sensor also showed good repeatability and of the control. Fast, small and inexpensive wall shear stress
minimal drift. sensors like the microfabricated thermal or floating element

683
L Lofdahl and M Gad-el-Hak

sensors discussed in this section would be necessary to with every advance in MEMS technology, and the future
accomplish this kind of futuristic control system [78]. worldwide market for micromachines is bound to be worth
As mentioned in the introduction to section 3, fluctuating tens of billions of dollars. A sizable part of this market will be
wall shear stress is an indicator of the turbulence activity in the transducer field, and our ability to measure and control
of the flow. In the ongoing development of existing quantities of practical importance will improve dramatically.
turbulence models for application in high-Reynolds-number
flows, absolute values of the fluctuating wall shear stress
are of significant interest. To accomplish these kinds of Acknowledgments
measurements, reliable methods for conducting dynamic
calibrations of wall shear stress sensors are needed and The authors would like to acknowledge the valuable
must be developed. The recent calibration method of help of Dr Andrey Bakchinov and Mr Peter Johansson
Padmanabhan [64] is of interest and other strategies are likely for preparing the figures and commenting on this paper.
to be developed in the future. Another intriguing possibility Our sincere appreciation goes to Professor Mihir Sen for
that would be challenging is to design a microsensor where sharing his ideas regarding shear sensor calibration. The
the dynamic effects of the gauge could be controlled in such first author acknowledges the hospitality of the University
a way that just a static calibration of the sensor would be of Notre Dame and its Department of Aerospace and
adequate. Mechanical Engineering, where he spent summer 1998 as
the Melchor Chair of Engineering. He also acknowledges
4. Concluding remarks continuous financial support from the Swedish Research
Council for Engineering Sciences (TFR). The second author
In a turbulent flow, an increase in Reynolds number acknowledges the US National Science Foundation support,
automatically generates smaller length scales and shorter under the Small Grants for Exploratory Research initiative
time scales, which both in turn require small and fast sensors (SGER Grant No CTS-95-21612).
for a correct resolution of the flow field. MEMS offer a
solution to this problem since sensors with length and time
scales of the order of the relevant Kolmogorov microscales References
can now be fabricated. Additionally, these sensors can be
[1] Tennekes H and Lumley J L 1972 A First Course in
produced with high accuracy at a relatively low cost per Turbulence (Cambridge, MA: MIT Press)
unit. For instance, a MEMS pressure sensor can be used to [2] Hinze J O 1975 Turbulence Second edn (New York:
determine fluctuating pressures beneath a turbulent boundary McGraw-Hill)
layer with a spatial resolution that is about one order of [3] Petersen K E 1982 Silicon as a mechanical material Proc.
magnitude finer than what can be achieved with conventional IEEE 70 42057
[4] Bryzek J, Peterson K and McCulley W 1994 Micromachines
microphones. on the march IEEE Spectrum 31 2031
In this paper we have attempted to bridge the gap [5] Tien N C 1997 Silicon micromachined thermal sensors and
between two disciplines, namely microelectromechanical actuators Microscale Thermophys. Eng. 1 27592
systems and turbulent flows. We have reviewed the state- [6] Blake W K 1986 Mechanics of Flow-Induced Sound and
of-the-art of microsensors used to measure the instantaneous Vibration (New York: Academic)
[7] Eckelmann H 1990 A review of knowledge on pressure
pressure and wall shear stress, which we deem as quantities fluctuations Near-Wall Turbulence ed S J Kline and
of primary importance in turbulence diagnosis. For each N H Afgan (New York: Hemisphere) pp 32847
group, we gave a general background, design criteria [8] Keith W, Hurdis D and Abraham B 1992 A comparison of
and calibration procedure, and provided examples of turbulent boundary layer wall-pressure spectra J. Fluids
measurements conducted with MEMS-based sensors and, Eng. 114 33847
[9] Harrison M 1958 Pressure fluctuations on the wall adjacent
where possible, compared the results with conventional to turbulent boundary layers DTNSRDC Technical Report
measurements. No. 1260 David Taylor Naval Ship Research and
Microsensors can be fabricated at low unit cost when Development Center, Bethesda, MA, USA
mass produced, and can be spaced close together in dense [10] Willmarth W W and Wooldridge C E 1962 Measurements of
arrays. These traits are particularly useful in studies of the fluctuating pressure at the wall beneath a thick
turbulent boundary layer J. Fluid Mech. 14 187210
coherent structures in wall-bounded turbulent flows and [11] Bull M K 1967 Wall-pressure fluctuations associated with
in reactive flow control schemes. Finally, for readers subsonic turbulent boundary layer flow J. Fluid Mech. 28
interested more in the fluid mechanics aspects of the topics 71954
discussed here, two complementary papers are recommended [12] Bull M K and Thomas A S W 1976 High frequency
for further reading. The first [79] deals with flow physics wall-pressure fluctuations in turbulent boundary layers
Phys. Fluids 19 5979
in microdevices and the second [80] emphasizes MEMS [13] Schewe G 1983 On the structure and resolution of
applications in turbulence diagnosis and control. wall-pressure fluctuations associated with turbulent
In closing, microelectromechanical systems have boundary-layer flow J. Fluid Mech. 134 31128
witnessed phenomenal advances in a mere ten-year period. [14] Lauchle G C and Daniels M A 1984 Wall-pressure
The 1960s and 1970s were arguably the decades of the fluctuations in turbulent pipe flow Phys. Fluids 30
301924
transistorten years after its discoveryand it is likely that [15] Farabee T M and Casarella M 1991 Spectral features of wall
the first ten years of the third millennium will be the MEMS pressure fluctuations beneath turbulent boundary layers
decade. Medical and industrial breakthroughs are inevitable Phys. Fluids 3 241019

684
MEMS-based pressure and shear stress sensors for turbulent flows

[16] Gad-el-Hak M and Bandyopadhyay P R 1994 Reynolds [38] Aronson D, Johansson A V and Lofdahl L 1997 A shear-free
number effects in wall-bounded flows Appl. Mech. Rev. 47 turbulent boundary layerexperiments and modelling
30765 J. Fluid Mech. 338 36385
[17] Farabee T M 1986 An experimental investigation of wall [39] Kalvesten E 1994 Piezoresistive silicon microphones for
pressure fluctuations beneath non-equilibrium turbulent turbulence measurements Thesis for Technical Licentiate
flows DTNSRDC Technical Report No. 86/047 David Degree Royal Institute of Technology, TRITA-IL-9402,
Taylor Naval Ship Research and Development Center, Stockholm, Sweden
Bethesda, MA, USA [40] Guckel H 1991 Surface micromachined pressure transducers
[18] Gedney C J and Leehey P 1989 Wall pressure fluctuations Sens. Actuators A 28 13346
during transition on a flat plate Proc. Symp. on Flow [41] Rossi M 1988 Acoustics and Electroacoustics (Boston, MA:
Induced Noise due to LaminarTurbulent Transition Artech House) pp 245308
Process (ASME NCA5) (New York: ASME) [42] Kalvesten E, Lofdahl L and Stemme G 1995 Small
[19] Royer M, Holmen P, Wurm M, Aadland P and Glenn M 1983 piezoresistive silicon microphones specially designed for
ZnO on Si integrated acoustic sensor Sens. Actuators A 4 the characterization of turbulent gas flows Sens. Actuators
35762 A 4647 1515
[20] Kim E S, Kim J R and Muller R S 1991 Improved [43] Kalvesten E, Lofdahl L and Stemme G 1996 An analytical
IC-compatible piezoelectric microphone and CMOS characterization of a piezoresistive square diaphragm
process Proc. Sixth Int. Conf. on Solid-State Sensors and silicon microphone Sens. Materials 8 11336
Actuators (Transducers91) (San Francisco, California) [44] Kalvesten E, Vieider C, Lofdahl L and Stemme G 1996 An
June 2428 1991 pp 2703 integrated pressurevelocity sensor for correlation
[21] Kuhnel W 1991 Silicon condenser microphone with measurements in turbulent gas flows Sens. Actuators A 52
integrated field-effect transistor Sens. Actuators A 2527 518
5215 [45] Lofdahl L, Kalvesten E and Stemme G 1996 Small silicon
[22] Schellin R, Pedersen M, Olthuis W, Bergveld P and Hess G pressure transducers for space-time correlation
1995 A monolithically integrated silicon microphone with measurements in a flat plate boundary layer J. Fluids Eng.
piezoelectric polymer layers MME95, Micromechanics 118 45763
Europe (Copenhagen, Denmark) September 35, 1995 [46] Miller M F, Allen M G, Arkilic E, Breuer K S and Schmidt
pp 21720 M A 1997 FabryPerot pressure sensor arrays for imaging
[23] Sessler G 1991 Acoustic sensors Sens. Actuators A 2527 surface pressure distributions Int. Conf. on Solid-State
32330 Sensors and Actuators, Transducers 97 (Chicago, IL) June
[24] Schellin R and Hess G 1992 A silicon microphone based on 1619, 1997 pp 146972
piezoresistive polysilicon strain gauges Sens. Actuators A [47] Stanton T E, Marshall D and Bryant C N 1920 On the
32 5559 conditions at the boundary of a fluid in turbulent motion
[25] Guckel H 1987 Fine grained films and its application to Proc. R. Soc. A 97 41334
planar pressure transducers Proc. Fourth Int. Conf. on [48] Leveque M A 1928 Transmission de chaleur par convection
Solid-State Sensors and Actuators (Transducers 87) Ann. Mines. 13 283
(Tokyo, Japan) June 25, 1987 pp 27782 [49] Munson B H, Young D F and Okiishi T H 1990
[26] Sugiyama S, Shimaoka K and Tabata O 1993 Surface Fundamentals of Fluid Mechanics (New York: Wiley)
micromachined microdiaphragm pressure sensor Sens. [50] Haritonidis J H 1989 The measurements of wall-shear stress
Mater. 4 26575 Advances in Fluid Mechanics Measurements
[27] Liu J, Tai Y-C, Pong K and Ho C-M 1995 MEMS for ed M Gad-el-Hak (Berlin: Springer) pp 22961
pressure distribution studies of gaseous flows in [51] Winter K 1977 An outline of the techniques available for the
microchannels Proc. IEEE Eighth Int. Workshop on Micro measurement of skin friction in turbulent boundary layers
Electro Mechanical Systems (MEMS95) (Amsterdam, The Progr. Aerosp. Sci. 18 155
Netherlands) Jan 29Feb 2, 1995 (New York: IEEE) [52] Hakkinen R J 1991 Survey of skin friction measurements
[28] Scheeper P, van der Donk A, Olthuis W and Bergveld P 1994 techniques AIAA Minisymposium (Dayton, OH)
A review of silicon microphones Sens. Actuators A 44 111 [53] Hanratty T J and Campbell J A 1996 Measurement of wall
[29] Hohm D 1985 Subminiatur-silizium-kondensatormikrofon, shear stress Fluid Mechanics Measurements second edn,
fortschritte der akustik DAGA85 84750 ed R J Goldstein (Washington, DC: Taylor and Francis)
[30] Sprenkels A J 1988 A silicon subminiature electret pp 575648
capacitive microphone PhD Thesis University of Utrecht, [54] Ludwieg H 1950 Instrument for measuring the wall shearing
Utrecht, The Netherlands stress of turbulent boundary layers NACA Technical
[31] Murphy P, Hubschi K, De Rooij N and Racine C 1989 Memorandum No. 1284 NACA, Washington, DC, USA
Subminiature silicon integrated electret capacitive [55] Liepmann H W and Skinner G T 1954 Shearing-stress
microphone IEEE Trans. Electric Insul. 24 4958 measurements by use of a heated element NACA Technical
[32] Bergqvist J 1994 Modelling and micromachining of Note No. 3268 NACA, Washington, DC, USA
capacitive microphones PhD Thesis Uppsala University, [56] Bellhouse B J and Schultz L 1966 Determination of mean
Uppsala, Sweden and dynamic skin friction separation in low-speed flow
[33] Bergqvist J and Rudolf F 1990 A new condenser microphone with a thin-film heated element J. Fluid Mech. 24
in silicon Sens. Actuators A 21 1235 379400
[34] Kolmogorov A N 1941 The local structure of turbulence in [57] Menendez A N and Ramaprian B R 1985 The use of
incompressible viscous fluid for very large Reynolds flush-mounted hot-film gauges to measure skin friction
numbers Comp. Rend. Acad. Sci. URSS 30 3015 in unsteady boundary layers J. Fluid Mech. 161
[35] Kolmogorov A N 1941 On logarithmicnormal law of 13959
distribution Comp. Rend. Acad. Sci. URSS 31 99 [58] King L V 1914 On the convection of heat from small
[36] Aronson D and Lofdahl L 1994 The plane wake of a cylinders in a stream of fluid: determination of the
cylinder: an estimate of the pressure strain rate tensor convection constants of small platinum wires, with
Phys. Fluids 6 271621 applications to hot-wire anemometry Proc. R. Soc. A 90
[37] Ligrani P M and Bradshaw P 1987 Spatial resolution and 56370
measurements of turbulence in the viscous sublayer using [59] Ramaprian B R and Tu S W 1983 Calibration of a heat flux
subminiature hot-wire probes Experiments in Fluids 5 gauge for skin friction measurement J. Fluids Eng. 105
40717 4557

685
L Lofdahl and M Gad-el-Hak

[60] Patel V C 1965 Calibration of the preston tube and limitations [76] Pan T, Hyman D, Mehregany M, Reshotko E and Willis B
on its use in pressure gradients J. Fluid Mech. 23 1995 Calibration of microfabricated shear stress sensors
185208 Proc. Eighth Int. Conf. on Solid-State Sensors and
[61] Breuer K S 1995 Stochastic calibration of sensors in turbulent Actuators (Transducers95) (Stockholm, Sweden) June
flow fields Experiments in Fluids vol 19, pp 13841 2529 pp 4436
[62] Bremhorst K and Gilmore D B 1976 Comparison of dynamic [77] Padmanabhan A, Goldberg H, Breuer K and Schmidt M
and static hot-wire anemometer calibrations for velocity 1995 A silicon micromachined floating-element
perturbation measurements J. Phys. E: Sci. Instrum. 9 shear-stress sensor with optical position sensing by
1097100 photodiodes Proc. Eighth Int. Conf. on Solid-State
[63] Bellhouse B J and Rasmussen C G 1968 Low-frequency Sensors and Actuators (Transducers95) (Stockholm,
characteristics of hot-film anemometers DISA Info. 6 310 Sweden) June 2529 pp 4369
[64] Padmanabhan A 1997 Silicon micromachined sensors and [78] Gad-el-Hak M 1996 Modern developments in flow control
sensor arrays for shear-stress measurements in Appl. Mech. Rev. 49 36579
aerodynamic flows PhD Thesis Department of [79] Gad-el-Hak M 1999 The fluid mechanics of
Aeronautics and Astronautics, Massachusetts Institute of microdevicesthe Freeman Scholar Lecture J. Fluids
Technology, Cambridge, MA, USA Eng. 121 129
[65] Brodkey R S 1967 The Phenomena of Fluid Motions (New [80] Lofdahl L and Gad-el-Hak M 1999 MEMS applications in
York: Addison-Wesley) turbulence and flow control Progr. Aerospace Sci. 35
[66] Brown G L 1967 Theory and application of heated films for 101203
skin friction measurements Proc. Heat Transfer Fluid [81] Muller R S 1987 Strategies for sensor research Proc. Fourth
Mech. Inst. pp 36181 Int. Conf. Solid-State Sensors and Actuators
[67] Tardu S, Pham C T and Binder G 1991 Effects of (Transducers87) (Tokyo, Japan) 25 June pp 10711
longitudinal diffusion in the fluid and heat conduction to [82] Franz J 1988 Aufbau, funktionsweise und technische
the substrate on response of wall hot-film gauges Realisierung eines piezoelectrischen Siliziumsensors fur
Advances in Turbulence vol 3, ed A V Johansson and akustische Grossen VDI-Berichre No. 677, VDI, Germany
P H Alfredsson (Berlin: Springer) pp 50613 pp 299302
[68] Ling S C 1963 Heat transfer from a small isothermal [83] Sprenkels A J, Groothengel R A, Verloop A J and Bergveld P
spanwise strip on an insulated boundary J. Heat Trans. C 1989 Development of an electret microphone in silicon
85 2306 Sens. Actuators A 17 50912
[69] Pedley T J 1972 On the forced heat transfer from a hot-film [84] Hohm D and Hess G 1989 A subminiature condenser
embedded in the wall in two-dimensional unsteady flow microphone with silicon-nitride membrane and silicon
J. Fluid Mech. 55 32957 backplate J. Acoust. Soc. Am. 85 47680
[70] Blackwelder R F 1981 Hot-wire and hot-film anemometers [85] Voorthuysen J A, Bergveld P and Sprenkels A J 1989
Methods of Experimental PhysicsFluid Dynamics Semiconductor-based electret sensors for sound and
vol 18A, ed R J Emrich (New York: Academic) pressure IEEE Trans. Electric. Insul. 24 2676
pp 259314 [86] Kuhnel W and Hess G 1992 Micromachined subminiature
[71] Kalvesten E 1996 Pressure and wall shear stress sensors for condenser microphones in silicon Sens. Actuators A 32
turbulence measurements PhD Thesis Royal Institute of 5604
Technology, TRITA-ILA-9601, Stockholm, Sweden [87] Scheeper P, van der Donk A, Olthuis W and Bergveld P 1992
[72] Jiang F-K, Tai Y-C, Gupta B, Goodman R, Tung S, Huang J Fabrication of silicon condenser microphones using
and Ho C-M 1996 A surface micromachined shear-stress silicon wafer technology J. Micro. Syst. 1 14754
imager Proc. IEEE Ninth Int. Workshop on Micro Electro [88] Bourouina T, Spirkovitch S, Baillieu F and Vauge C 1992 A
Mechanical Systems (Amsterdam, The Netherlands) new condenser microphone with a p+ silicon membrane
February 1115 (New York: IEEE) pp 11015 Sens. Actuators A 31 14952
[73] Schmidt M, Howe R, Senturia S and Haritonidis J 1988 [89] Ried R, Kim E, Hong D and Muller R 1993 Piezoelectric
Design and calibration of a microfabricated microphone with on-chip CMOS circuits J. Micro. Syst. 2
floating-element shear-stress sensor IEEE Trans. Electron. 11120
Dev. 35 7507 [90] Kovacs A and Stoffel A 1995 Integrated condenser
[74] Ng K, Shajii K and Schmidt M 1991 A liquid shear-stress microphone with polysilicon electrodes MME95,
sensor fabricated using wafer bonding technology Proc. Micromechanics Europe (Copenhagen, Denmark) 35
Sixth Int. Conf. on Solid-State Sensors and Actuators September pp 13235
(Transducers 91) (San Francisco, CA) June 2427 1991 [91] Kronast W, Muller B and Stoffel A 1995 Miniaturized
pp 9314 single-chip silicon membrane microphone with integrated
[75] Goldberg H D, Breuer K S and Schmidt M A 1994 A silicon field-effect transistor MME95, Micromechanics Europe
wafer-bonding technology for microfabricated shear-stress (Copenhagen, Denmark) 35 September pp 1369
sensors with backside contacts Technical Digest, [92] Oudheusden B and Huijsing J 1988 Integrated flow friction
Solid-State Sensor and Actuator Workshop pp 11115 sensor Sens. Actuators A 15 13544

686

View publication stats