ARTICLE IN PRESS

Progress in Aerospace Sciences 44 (2008) 479–502

Contents lists available at ScienceDirect

Progress in Aerospace Sciences

journal homepage: www.elsevier.com/locate/paerosci

Active control of flow-induced cavity oscillations

Louis N. Cattafesta IIIa,, Qi Song a, David R. Williams b, Clarence W. Rowley c, Farrukh S. Alvi d
a
Mechanical and Aerospace Engineering, University of Florida, P.O. Box 116250, Gainesville, FL 32611-6250, USA
b
MMAE Department, Illinois Institute of Technology, E-1 Bldg., 3110 S. State St., Chicago, IL 60616, USA
c
Mechanical and Aerospace Engineering, Princeton University, D232 E-quad, Princeton, NJ 08544, USA
d
Mechanical Engineering, Florida A&M and Florida State University, 2525 Pottsdamer St., Tallahassee, FL 32310, USA

a r t i c l e in fo abstract

Available online 11 September 2008 A review of active control of flow-induced cavity oscillations is motivated by two factors. First, the
Keywords: search for solutions to the practical problem of suppressing oscillations caused by flow over open
Active flow control cavities has generated significant interest in this area. Second, cavity oscillation control serves as a
Cavity oscillations model problem in the growing multidisciplinary field of flow control. As such, we attempt to summarize
Cavity resonance recent activities in this area, with emphasis on experimental implementation of open- and closed-loop
Flow-induced oscillations control approaches. In addition to describing successes, failures, and outstanding issues relevant to
Actuators cavity oscillations, we highlight the characteristics of the various actuators, flow sensing and
Feedback control measurement, and control methodologies employed to date in order to emphasize the choices,
challenges, and potential of flow control in this and other applications, such as impact on store
trajectory.
& 2008 Elsevier Ltd. All rights reserved.

Contents

1. Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 479
2. Suppression of cavity oscillations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 481
2.1. Flow-control classifications . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 481
2.2. Passive/active and open-loop suppression studies . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 482
2.3. Effect of control on the flow field and store trajectory . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 487
3. Sensors and actuators . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 488
3.1. Passive actuators . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 488
3.2. Active open-loop actuators . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 489
3.3. Active closed-loop actuators . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 490
3.4. Sensors and flow measurements . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 493
4. Closed-loop control methodologies. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 493
4.1. Quasi-static vs. dynamic controllers . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 496
4.2. Models . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 496
4.2.1. System identification . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 496
4.2.2. POD/Galerkin models . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 497
4.2.3. Rossiter-type models . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 497
4.3. Control algorithms . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 497
4.3.1. State estimation: observers and static estimators . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 499
4.4. Fundamental limits on achievable performance. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 499
5. Summary and outlook . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 500
Acknowledgements . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 500
References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 500

1. Introduction

 Corresponding author. Tel.: +1 352 846 3017; fax: +1 352 392 7303. Flow over cavities has received a great deal of interest over the
E-mail address: cattafes@ufl.edu (L.N. Cattafesta III). last several years because of practical and academic interests

0376-0421/$ - see front matter & 2008 Elsevier Ltd. All rights reserved.
doi:10.1016/j.paerosci.2008.07.002
 ARTICLE IN PRESS

480 L.N. Cattafesta III et al. / Progress in Aerospace Sciences 44 (2008) 479–502

λ
Turbulent
Boundary Layer y

δ
x
Source
Receptivity Characteristics

D
Feedback

Fig. 1. Schematic illustrating flow-induced cavity resonance for an upstream turbulent boundary layer.

associated with controlling such flows. The problem of accurate

prediction and control over a wide range of flow conditions is
not solved. The need to accurately model the disparate scales
of acoustic and vortical disturbances driving the oscillations is a
difficult task for fluid dynamicists and aeroacousticians, while
control theorists are challenged by the multiple competing modes
of oscillation that must be controlled to achieve suppression.
These and a number of other issues have established flow-induced
cavity oscillations as a canonical problem in flow control.
The nature of flow-induced oscillations in an open cavity is
illustrated in Fig. 1. A boundary layer of thickness d and
momentum thickness y separates at the upstream edge of the
cavity of length L, depth D, and width W. The resulting shear layer
develops based on its initial conditions (imposed by the upstream
boundary layer and cavity acoustic field) and the instability
characteristics of the mean shear-layer profile. The shear layer
spans the length of the cavity and ultimately reattaches near
the trailing edge of the cavity in an ‘‘open’’ cavity flow. The
reattachment region acts as the primary acoustic source. Acoustic
waves travel inside the cavity (and outside for subsonic flow), Fig. 2. Pressure spectra for different L/D cavities at Mp ¼ 2. Unsteady pressure
towards the cavity leading edge. The incident acoustic waves force spectra measured using a transducer located at the upstream wall, as shown in the
inset (from Zhuang et al. [1]). The dimensionless frequency is the Strouhal number
the shear layer, setting the initial amplitude and phase of the St ¼ fL/UN.
instability waves through a receptivity process. These instabilities
grow to form large-scale vortical structures that convect down-
stream at a fraction of the free stream speed before impinging
near the trailing edge. sive review of the subject here. The interested reader is referred
The overall process produces resonant frequencies, which are to several reviews spanning three decades of research [6–14].
referred to as cavity tones. Fig. 2 illustrates representative, In particular, the review by Colonius [12] provides a summary
unsteady pressure spectra (dB re 20 mPa) at Mach 2 for a range of numerical simulations and flow-physics modeling, permitting
of cavity length/depth L/D ratios [1]. The spectra are dominated by us to largely ignore these relevant topics.
high-amplitude, discrete-frequency tones and large broadband Control of grazing flow over cavities is pertinent to a wide
levels. In many cases, multiple tones are observed, and these are range of real-world applications, ranging from landing-gear and
often accompanied by their harmonics. In the context of cavity weapons bays in aircraft to flow in gas transport systems [15],
flows, the flow–acoustic coupling which leads to resonance is over sunroofs and windows in automobiles [16], and in instru-
commonly called the Rossiter mechanism [2], although a similar ment or telescope bays [17]. The high dynamic loads illustrated in
phenomenological model was proposed for the edge-tone pro- Fig. 2 are generally present in all of these applications and can
blem more than a decade earlier by Powell [3]. The relevant lead to structural fatigue of the cavity and its contents or, in the
dimensionless parameters are L/D, L/W, and L/y, as well as the flow case of compressible flow, aero-optic distortion [18]. In addition,
parameters prms/qN, Reynolds number Rey, Mach number MN, and the highly oscillatory flow field generated by cavity flows can
shape factor H ¼ d*/y, where prms is the rms pressure fluctuation, adversely affect the safe departure and accurate delivery of
qN is the free stream dynamic pressure, and d* and y are the munitions stored in the weapons bay. This problem has become
displacement and momentum thicknesses, respectively, of the more acute with the recent emphasis on store separation of
upstream boundary layer. ‘‘smart’’ weapons that are lighter and more compact [19].
Cavity flows have been the subject of numerous studies since Although the overall goal of control is usually to reduce the
the 1950s [4,5], and we do not attempt to provide a comprehen- flow unsteadiness in some form, the specific objective of what
 ARTICLE IN PRESS

L.N. Cattafesta III et al. / Progress in Aerospace Sciences 44 (2008) 479–502 481

constitutes ‘‘effective control’’ is largely dependent on the applica- modeling/design approaches are summarized in Section 4. Finally,
tion. In strongly resonant cases, where one or more discrete tones Section 5 provides a summary and offers our perspective on
are dominant compared to the background broadband level, outstanding issues and future directions.
attenuation of the tones is generally sufficient. However, in
situations where the broadband levels may be comparable to
the tonal amplitudes, a reduction in the overall (integrated) noise 2. Suppression of cavity oscillations
levels may be required. In still other applications, such as aero-
optics, the objective may be to reduce optical wavefront aberra- 2.1. Flow-control classifications
tions caused by flow-induced, index-of-refraction variations. In
any case, the key is to translate the application objective to an Techniques to suppress cavity oscillations can be classified in
effective flow-control strategy and, in the case of closed-loop several ways. In this paper, we choose the classification shown in
control, to a mathematical statement of the control objective. Fig. 3, to be consistent with the terminology used in active noise
Therefore, the primary purpose of this review is to summarize and vibration control. Active control provides external energy (e.g.,
recent activities in active control of flow-induced cavity oscilla- mechanical or electrical) input to an adjustable actuator to control
tions, with particular emphasis on implementation of open- and the flow, while passive control techniques do not. Passive control
closed-loop control approaches. In addition to describing suc- of cavity oscillations has been successfully implemented via
cesses, failures, and outstanding issues relevant to cavity oscilla- geometric modifications using, for example, fixed fences, spoilers,
tions, we highlight the characteristics of various actuators, flow ramps [20,21], and a passive bleed system [22]. Note that some
sensing and measurement, and control methodologies in order to control devices considered passive by this classification extract
emphasize the choices, challenges, and potential of flow control. energy from the flow itself and have been called ‘‘active’’ by other
The paper is organized as follows. Section 2 provides a brief researchers. Pertinent examples include unpowered or passive
overview of passive and open-loop suppression studies and resonance tubes [19,23] and cylinders or rods placed in the
discusses the application to store separation. While recent studies boundary layer near the leading edge of the cavity [24,25]. These
are emphasized, some classical studies are reviewed to place devices, described further in Section 3, are sometimes referred to
these new results in proper context. Section 3 describes actuators as active because they provide an oscillatory input to the flow, but
and flow measurements, while closed-loop control and pertinent their effect on the flow cannot be adjusted without either
changing the flow conditions or changing the device itself.
Active control is further divided into open- and closed-loop
Flow Control approaches. As shown in Fig. 4, by its very definition, closed-loop
Approaches control implies a feedback loop, in which some flow quantity is
directly sensed or estimated and fed back to modify the control
signal [26]. Open loop corresponds to the case when there is no
Passive Control Active Control feedback loop.
A further non-standard but useful classification of closed-loop
flow control is that of quasi-static vs. dynamic feedback control.
The quasi-static case corresponds to slow tuning of an open-loop
Open-Loop Closed-Loop
control approach and occurs when the time scales of feedback are
large compared to the time scales of the plant (i.e., flow). This
approach is particularly relevant in nonlinear fluid dynamic
Quasi-static Dynamic systems, where the fundamental notion of frequency preservation
in a linear system does not hold. As discussed in Section 5,
Fig. 3. Possible classification of flow-control approaches. the quasi-static approach was successfully used by Shaw and

Disturbance

d
Control
Signal
Actuating or
(Error) Manipulated Controlled
Reference
Signal Variable Output
Input Control Plant
r + e C u P y
−

FORWARD PATH →
Primary
Feedback
Signal

Feedback
H

← FEEDBACK PATH

Fig. 4. Components of a feedback control system [26]. The various time-domain signals are in lower case, while the transfer functions of the blocks are in upper case
(Laplace domain).
 ARTICLE IN PRESS

482 L.N. Cattafesta III et al. / Progress in Aerospace Sciences 44 (2008) 479–502

Northcraft [27]. The usual dynamic compensation case corre- the potential mechanism of the actuator to the interaction of
sponds to the situation when the above time scales are the shed vortices with the shear-layer instabilities. As will be
commensurate. This can be implemented using an analog (see, discussed later, however, there are additional possible mechan-
for example, Williams et al. [28]) or ‘‘real-time’’ digital control isms noted by other investigators that influence the suppression
systems [29]. In this context, ‘‘real time’’ refers to the situation in effectiveness of the cylinder.
which the control signal is updated at the sampling rate of the At the same time in the mid-1990s, Ahuja and his colleagues
data system, and the actuator response to the flow state changes were investigating other novel control strategies. For example,
at the time scales of the dynamics. Mendoza and Ahuja [33] studied the effect of a steady wall jet on
the tone production mechanism, using a Coanda surface. Although
no mass flow measurements were obtained, upstream boundary
2.2. Passive/active and open-loop suppression studies layer profiles showed an increase in d with blowing, thereby
leading to the hypothesis that the suppression was due to reduced
Table 1 provides a detailed list of passive/active and open-loop amplification of the shear-layer instabilities.
cavity suppression studies. Where possible, the cavity dimensions Hsu and Ahuja [34] studied the effect of a trailing-edge array of
and flow conditions are summarized. The methods and key results Helmholtz resonators (i.e., commercial syringes) on cavity noise,
are also included. The investigations are listed in chronological and they obtained some suppression at lower Mach numbers. At
order for historical purposes. It is impossible to include all prior intermediate Mach numbers, the resonators reduced the magni-
research; there are numerous other (mostly passive) studies that tude of the targeted tone, but new tones appeared at other
have not been included here because most of these have been frequencies—a phenomenon that has been observed by many
summarized in the other reviews cited earlier. Some key researchers. At high Mach numbers, no suppression was obtained,
observations are discussed here. but the authors believed that the reason was likely due to the
Sarohia and Massier [30] studied the efficacy of steady mass difficulty in setting the small resonator volume accurately. This
injection at the base or floor of two different axisymmetric cavity study is noteworthy for its attempt to control cavity oscillations in
models for both laminar and turbulent boundary layers. While the vicinity of the acoustic source origin near the trailing-edge
base injection was effective at suppressing cavity tones, large impingement region. Generally, actuators are placed at the
mass flow rates were required—Bc ¼ 5–15%. The mass flow rate leading edge of the cavity to leverage the growth of instabilities
parameter used by the authors is equivalent to that proposed later in the shear layer.
by Vakili and Gauthier [31] for rectangular cavities: In 1997, Cattafesta et al. [35] presented the use of a six-element
  piezoelectric flap array flush mounted at the leading edge of
rw V w Ainj m_
Bc ¼ ¼ , (1) the cavity. The bandwidth of the actuators was large enough
re V e Acavity re V e Acavity
(300 Hz) to provide forcing at frequencies comparable to that of
where Ainj is the total injector area, and Acavity ¼ LW is the area of the cavity tones. Open-loop sinusoidal forcing at a sufficient
the cavity exposed to the free stream flow. amplitude and appropriate detuned frequency was capable of
Sarno and Franke [32] studied static and oscillating fences, and suppressing the cavity tone. Shear-layer velocity measurements
also steady and pulsed injection (at 01 or 451 with respect to the indicated that the actuators seeded the shear layer with a
free stream direction) at transonic speeds near the cavity leading disturbance that was large enough to prevent the growth of the
edge. Blowing coefficients Bc of up to 7% were used. While the natural cavity disturbances. However, the possibility of starving
static fences provided the best suppression, the bandwidth of the the growth of natural instabilities at high subsonic and supersonic
mechanical fences was limited to o220 Hz, while the pulsed Mach numbers is questionable, because of the large amount of
injection was o80 Hz. These frequencies were at least an order- mean-flow energy available (M 21 ) for disturbance amplification.
of-magnitude lower than the frequencies of the cavity tones, In 1998, Shaw continued his study of leading-edge HFTGs, low-
and thereby constituted a quasi-static or low-frequency forcing. frequency pulsed-fluidic injection, and oscillating flaps [36].
Nonetheless, they represented an important step in the evolution While various diameter HFTGs mounted at a fixed height were
of active control of cavity oscillations, both in terms of approach shown to be effective, the suppression improved as the diameter
and also the introduction of scaling laws for such actuators. was increased. However, the relative height of the cylinder in the
Vakili and Gauthier [31] obtained significant acoustic tone boundary layer was not reported, which is now known to be an
attenuation with steady normal mass injection through variable- important parameter. Shaw also discussed two potential mechan-
density porous plates upstream of the cavity leading edge at Mach isms of the HFTG: (1) high-frequency forcing due to shedding and
1.8 using Bc4%. They attributed the attenuation to a thickening of (2) reduced shear-layer growth rates due to boundary layer
the cavity shear layer and a corresponding alteration of its thickening.
instability characteristics. Rescaling Shaw’s pulsed-blowing results shows that they are
McGrath and Shaw [24] subsequently studied mechanical consistent with prior results, since Bc values of a few percent were
oscillations of hinged flaps at frequencies up to 35 Hz over a range required to suppress the tones. However, no spectra were reported
of subsonic and supersonic Mach numbers. Similar to the Sarno to assess the impact of blowing on the broadband noise.
and Franke experiments, the forcing frequency was an order-of- Interestingly, the tone amplitude continued to decrease as the
magnitude lower than the resonant tone frequency. The static and pulse frequency of the injector reached its upper limit of 100 Hz.
oscillatory deflections were on the order of the boundary layer Furthermore, normal injection was shown to be superior to
thickness d and were shown to be effective despite their limited tangential blowing.
bandwidth. The oscillatory flap frequency in Shaw’s experiment was varied
McGrath and Shaw were the first to study the effect of a from 0 to 100 Hz and provided maximum suppression at 5 Hz. A
cylinder placed in the upstream subsonic boundary layer. Because monotonic improvement (reduction) in unsteady pressure level
of the well-known shedding characteristics of a circular cylinder occurred as the dynamic deflection angle increased to its upper
of diameter d, St ¼ fd/UE0.2, over a wide Reynolds number range, limit, corresponding to a deflection of order d. However, the
this device was called a high-frequency tone generator (HFTG). increase in d for a full-scale aircraft led Shaw to conclude that this
The cylinder was capable of producing substantial reductions of approach (low-frequency, large-amplitude, open-loop forcing)
both the cavity tones and the broadband. The authors attributed was not feasible for a full-scale aircraft.
Table 1
Summary of selected passive and active and open-loop cavity suppression studies

Study Conditions Method Comments

Sarohia and  Two axisymmetric cavity models  Steady injection at base of the cavity Bc5–15%  Large mass flow rates were required to stabilize
Massier  L/D ¼ 0–1.5 cavity oscillations
[30]  MN ¼ 0–0.5
 Laminar and turbulent BL

Sarno and  L/D ¼ 2, L/W ¼ 6.4 at MN ¼ 0.6, 0.7, 0.9, 1.1, 1.3, 1.5  Static and oscillating mechanical fences (o220 Hz)  Insufficient actuator bandwidth
Franke  Calculated d ¼ 0.048 in (at MN ¼ 1.53) to 0.062 in (at  Steady and pulsed flow injection (o80 Hz) at either 01 or 451 w.r.t. free stream at leading  Low-frequency forcing ineffective
[32] MN ¼ 0.62) edge  Significant 3-D effects possible
 Bc up to 7%

L.N. Cattafesta III et al. / Progress in Aerospace Sciences 44 (2008) 479–502

Vakili and  L/D ¼ L/W ¼ 2.54, MN ¼ 1.8  Obtained significant attenuation with steady mass injection through porous plates upstream  Proposed blowing coefficient
Gauthier  Red ¼ 3.68  105 of cavity leading edge Bc4%
 
[31] rw V w Ainj m_
Bc ¼ ¼
re V e Acavity r1 U 1 Acavity
 Attenuation attributed to thickening of cavity

ARTICLE IN PRESS
shear layer to alter its instability characteristics

McGrath and  L/D ¼ 2.56, 3.73, 6.83  Mechanical oscillations of hinged flap up to 35 Hz at MN ¼ 0.6, 0.8, 1.5, 1.89  Oscillating flaps show effective reduction of
Shaw  Re/m ¼ 6.6  106  Static and oscillatory deflections of up to 1d tones at subsonic and supersonic flow
[24]  d ¼ 0.13 in at MN ¼ 0.85  Cylinder (d ¼ 0.062 in) placed in upstream cavity BL at MN ¼ 0.6, 0.8-called high-frequency conditions
tone generator HFTG  Limited bandwidth of actuator
 Based on shedding concept St ¼ fd/U ¼ 0.2 for Red ¼ 103–105  HFTG shows effective reduction of tones and
broadband
 Proposed mechanism is the interaction of shed
vortices with shear-layer instabilities
 Resonance effects?

Mendoza  L/D ¼ 3.75, L/W ¼ 0.47  Studied effects of steady wall jet on cavity noise (normal via 0.2 mm slot and upstream via a  Upstream BL profiles showed an increase in d
and  MN ¼ 0.36, 0.44, 0.55, 0.9, 1.05 1/4 in radius of curvature Coanda surface 0.5625 in upstream of cavity leading edge) with upstream blowing from Coanda surface
Ahuja  d ¼ 0.057, 0.059, 0.062 in at MN ¼ 0.36, 0.44, 0.55,  Proposed d/L40.07 for elimination of cavity tones  Hypothesized that suppression is due to
[33] respectively reduced shear-layer growth rate
 No injected mass flow measurements provided

Hsu and  L/D ¼ 2.5, L/W ¼ 0.47  Studied effect of trailing-edge array of Helmholtz resonators (commercial syringes) on cavity  No effect of resonators when in their closed
Ahuja  MN ¼ 0.34, 0.53, 0.9 noise position
[34]  Significant reductions observed at low Mach
numbers, but new tones can appear
 Negligible suppression at M ¼ 0.9
 Reason may be due to difficulty in accurately
setting correct volume of all resonators

Cattafesta et  L/D ¼ 0.5, L/W ¼ 0.5, UN ¼ 40 m/s, Rey ¼ 4750, L/y ¼ 81  Used piezoelectric flaps flush mounted at the leading edge of the cavity to suppress low-  OL forcing at detuned frequency suppresses
al. [35]  L/D ¼ 2.0, L/W ¼ 2.0, UN ¼ 45 m/s, Rey ¼ 5210, L/ speed cavity oscillations oscillations
y ¼ 328  Shear-layer meas. w/ & w/o control. Also CL
results

Lamp and  L/D ¼ 4, L/W ¼ 3  Used rotary valve actuator to provide steady and/or oscillatory blowing 0.1 in upstream of  Configuration emphasized 3-D effects
Chokani  MN ¼ 0.15, 0.23 cavity at frequencies up to 750 Hz  Showed that oscillatory blowing can reduce

[38]  Re/m ¼ 5.9  106  Used hcm i ¼ ru2rms Aj qAr up to 0.16% where Ar ¼ cavity front wall area (2 by 1.5 in) up to 0.16% tone amplitude significantly (up to 10 dB)
provided frequency of forcing is not a harmonic
of the cavity resonance, but new tones appear at

483
the forcing frequency
 484
Table 1 (continued )

Study Conditions Method Comments

Shaw [36]  L/D ¼ 6.5, L/W ¼ 3.67  Studied leading edge oscillating flaps, pulsed fluidic actuation, and HFTG using 1/16, 1/8, and  Discussed possible mechanisms of HFTG (mode
 MN ¼ 0.6, 0.85, 0.95, 1.05 3/16 in diameter cylindrical rods located at a height of 0.3 in competition, shear-layer instability)
 d0.38 in  Low-freq. notched flaps effective when

L.N. Cattafesta III et al. / Progress in Aerospace Sciences 44 (2008) 479–502

 Re/m ¼ 6.6  106 oscillating flap reaches BL edge
 Showed normal injection (Bco4.5%) superior to
tangential

Fabris and  L/D ¼ 4, L/W ¼ 0.49  Unsteady bleed forcing (second mode) through a 12.7 mm slot located 12.7 mm ahead of  Shear-layer receptive to unsteady bleed forcing
Williams  MN ¼ 0.15, 0.23 (y ¼ 3.1 mm) cavity leading edge via speakers

ARTICLE IN PRESS
[37]
Raman et al.  L/D ¼ 6, L/W ¼ 1.7  Used upstream miniature fluidic oscillators to suppress cavity oscillations via sine, square,  Comparable steady injection and upstream
[39,40]  MN ¼ 0.4-0.7 and triangular waveforms up to 3 kHz with mass flow rates of only 0.12% of main jet flow acoustic excitation do not suppress the cavity
 Jet cavity configuration resonance
 Hypothesized that periodic sweeping motion in
spanwise direction destroys the spanwise
coherence of shear layer

Stanek et al.  L/D ¼ L/W ¼ 5  Investigated ‘‘high-frequency’’ powered resonance tubes, protruding piezoceramic driven  Successful hifex forcing for powered resonance
[19]  MN ¼ 0.4, 0.6, 0.85, 0.95, 1.19, 1.35 wedges, a cylindrical rod, and passive resonance tubes vs. conventional ‘‘low-frequency’’ 1 d tube at Bc ¼ 1.6%
 Calculated d ¼ 10.5 mm @ MN ¼ 0.6, Re/m ¼ 11 106 spoiler  Hypothesize that mechanism is ‘‘accelerated
energy cascade’’

Bueno et al.  MN ¼ 2, d ¼ 0.53 in  Six fast-response (3 ms) jets used to study effects of upstream pulsed and steady mass  Bc ¼ 0.28%, 0.24%, 0.18%, 0.16% at L/D ¼ 5, 6, 8, 9,
[48]  Re/m ¼ 3.0  107 injection via a staggered configuration respectively, for continuous injection with all
 L/D ¼ 5, 6, 8, 9, W/D ¼ 3  Studied single short duration (1–2 ms) and long duration (50% duty cycle of forcing frequency six valves
of 50 or 80 Hz) pulses  Found that continuous mass injection was more
effective than pulsed injection in suppressing
tones and overall noise level

Ukeiley et al.  At MN ¼ 0.6, 0.75, d ¼ 0.1, 0.08 in, Re/m ¼ 3.6, 4.9  107,  Used leading edge fence and 0.03 in diameter cylindrical rod at various locations in BL for  Shear-layer profiles suggest that rod works by
[25] respectively, L/D ¼ 5.6, 9 suppression ‘‘lifting’’ the shear layer and also by altering the
 W/D ¼ 2 mean shear layer
 At MN ¼ 0.8 & 1.4, Re/m ¼ 3.15  107, cavity dimensions  Cross-correlations of the sensors in the cavity
doubled suggest that the amplitude of the upstream
traveling wave along the floor of the cavity was
reduced slightly

Stanek et al.  Similar conditions as in Stanek et al. [19]  Investigated powered and unpowered resonance tubes, microjets vs. various other devices  Proposed that hifex forcing alters the mean flow
[41]  Attempted to reduce mass flow requirements of mass flow devices and its inviscid stability characteristics such
that the growth of large-scale disturbances is
prevented
 Provides evidence that substantial amount of
suppression in powered resonance tubes may
be due to steady blowing effects (optimal
Bc0.6%)
Stanek et al.  Similar conditions as in Stanek et al. [19]  Investigated cylindrical rod in crossflow  Recommended optimal location is rod center at
[42]  Investigated endcaps and fences BL edge, optimal diameter is 2d/3
 Concluded that acoustic suppression is due to
high-frequency rod shedding

Ukeiley et al.  Similar conditions as in Ukeiley et al. [25]  Investigated powered whistles (8) to superimpose hifex perturbations on horizontal steady  Used Bc up to 0.4% but best results (factor of
[56] blowing at the leading edge with heated air, nitrogen (28 kHz), and helium (78 kHz) 2–4 reduction in aft wall OASPL) were with
lowest steady helium injection rate (Bc0.09%)
 Suppression mechanism unclear; limited PIV
images, and cross-correlation data suggest that
injection alters impingement region and
disrupts acoustic feedback loop

Zhuang et al.  MN ¼ 2, L/D ¼ 5.1, L/W ¼ 5.8, ReL ¼ 2.8  106  Investigated 400 mm diameter microjet array (12) with vertical injection  Found Bc ¼ 0.1–0.5% was sufficient for 5–11 dB
[1,50] reduction in OASPL and 13–28 dB peak
reduction
 Saturation noted as Bc increased
 High reversed flow velocities (40–50% of the

L.N. Cattafesta III et al. / Progress in Aerospace Sciences 44 (2008) 479–502

free stream) were observed in the middle and
near the bottom half of the cavity

Sahoo et al.  Similar conditions as in Zhuang et al. [50]  Microjet actuators  Developed physics-based model explanation of
[53] the effects of microjet blowing on supersonic

ARTICLE IN PRESS
cavity control

Schmit et al.  MN ¼ 0.85, 1.19  Powered resonance tube  Direct comparisons of low- and high-frequency
[55]  L/D ¼ 5  Rod wrapped with wire forcing did not show one method to be superior
 W/D ¼ 1  Rotary pulse-blowing to the other
 Sawtooth spoiler  Passive methods were ineffective at supersonic
speeds

Williams et  MN ¼ 1.86  Pulsed-blowing type of actuator with compressed air from siren valves  The overall supersonic cavity system performs
al. [66]  Re/m ¼ 4.9  107  Bandwidth of the device 1.5 kHz as a linear system.
 L/D ¼ 5  2-D slot orifice with height 3.2 mm  Input disturbances are amplified when the
 L/W ¼ 1.5  Bc ¼ 0.13% external forcing frequency is close to that of a
 d ¼ 8 mm Rossiter mode, and attenuated when the
external forcing frequency is between those of
Rossiter modes.

Ukeiley et al.  3-D cavity with diverging side walls  Eight 400 mm diameter jets evenly spaced along the leading edge of the cavity  For microjet case, Bc ¼ 0.1–0.55% was sufficient
[49]  MN ¼ 1.5  Different number of slots with different lengths and fixed width 0.254 mm for 1–6 dB reduction in OASPL and 5–8 dB in
 L/D ¼ 5.5  Bc ¼ 0.1–0.55% peak reduction.
 For 2-D slot case, up to 7 dB reduction in OASPL

485
 ARTICLE IN PRESS

486 L.N. Cattafesta III et al. / Progress in Aerospace Sciences 44 (2008) 479–502

In 1999, three new approaches were reported. Fabris and et al. [25] and the numerical simulations of Arunajatesan et al.
Williams [37] used unsteady bleed (zero-net mass-flux) forcing to [45], which indicate that the cylinder can also lift the shear layer
produce a broadband actuator capable of producing a complex and cause the impingement region to be altered. If the shear-layer
input disturbance comprised of multiple frequency components. impingement location is indeed altered, then the source strength
They demonstrated that the shear layer was most receptive to is presumably affected.
horizontal or tangential forcing via shear-layer velocity measure- It is clear that these studies provide insufficient information to
ments, in contrast to the results of Shaw [36]. sort out these different physical mechanisms. To do so requires
Lamp and Chokani [38] used a rotary-valve actuator to confirmation in the form of detailed experiments, analysis, and
provide steady and/or oscillatory blowing upstream of the cavity validated simulations to determine the mean velocity profile and
leading edge at a particular pulsing frequency. Their actuator subsequent shear-layer instability characteristics for various high-
configuration emphasized three-dimensional effects and showed frequency devices.
that oscillatory blowing can reduce tone amplitudes provided the One other mechanism worth mentioning is the possibility
forcing frequency is not a harmonic of the cavity resonance. of rod or wire resonance. A cylinder in crossflow can vary in
Raman et al. [39] and Raman and Raghu [40] used novel complexity from a resonating string in tension [46] to a pinned or
miniature fluidic oscillators to suppress cavity oscillations. These clamped rod in tension or compression (that can sustain bending)
devices were capable of producing up to 3 kHz oscillations with depending on its diameter, length, material, and boundary
mass flow rates of less than 0.12% of the main jet flow and condition [47]. Sample calculations reveal fundamental resonant
produced significant tonal reductions. However, the mass flow frequencies which can vary from several hundred hertz to several
rate and frequency of oscillations are coupled (albeit in a kilohertz, depending on the mounting configuration. If the wire/
predictable fashion). Whether the steady mass additions or the rod resonates due to the broadband excitation of the turbulent
unsteady oscillations were responsible for the sound suppression boundary layer and/or fluctuating pressure field, fluid/structure
could not be determined. This is a key unresolved issue and, as interaction effects, which have been ignored to date, may be
emphasized in Ref. [13], independent control of the mean and significant.
oscillatory components is required. There are a few other studies involving steady and/or pulsed
Stanek et al. [19,41,42] reported on a series of larger-scale blowing that have provided physical insight or have shown
experiments conducted in the United Kingdom over the past promising results. Bueno et al. [48] used an array of six fast-
few years. In the first experiment reported in 2000 [19], they response (3 ms) miniature jets mounted upstream of the leading
investigated powered resonance tubes, protruding piezoceramic edge to study the effects of normal injection on a Mach 2 cavity
driven wedges, a cylindrical rod, and passive resonance tubes vs. a flow. They used instantaneous and ensemble-averaged pressure–
conventional spoiler. An interesting result was that the powered time histories and cross-correlations to study the effects of single
resonance tubes (see [23] and Section 3 below) demonstrated short and long cyclical pulses (50% duty cycle), the latter with
significant tonal and broadband reduction when Bc1.6%. The relatively low forcing frequencies (50 or 80 Hz) compared to
result was termed a successful demonstration of high-frequency that of the tones. They compared their pulsed results to steady
forcing (defined as a frequency that is very large compared to that blowing with Bc ¼ 0.28%, 0.24%, 0.18%, and 0.16% at L/D ¼ 5, 6, 8,
of the cavity tones). and 9, respectively, and concluded that continuous mass injection
A follow-on study [41] reported in 2002 investigated powered is more effective for suppression than low-frequency pulsed
and unpowered resonance tubes (in which the resonator tubes blowing.
were blocked to inhibit high-frequency excitation), and microjets Ukeiley et al. [49] used an array of eight powered whistles
vs. various other devices. While the powered resonance tubes mounted in the forward cavity wall as flow-control actuators.
were redesigned to reduce their mass flow requirements, optimal These devices essentially produce a high-frequency tonal oscilla-
suppression still required Bc0.6%. The unpowered resonance tion superimposed on a steady jet. The jet is directed in the
tubes consistently provided the best suppression, indicating that downstream direction but has a slight vertical velocity compo-
the primary suppression mechanism of these devices is not just nent. The authors studied the novel use of different injection gases
due to high-frequency forcing but is also influenced by the steady (heated air, nitrogen, and helium) with and without the high-
blowing component. The results also introduced microjet blowing, frequency ‘‘whistle’’ component. The best sound suppression
and showed that vertical blowing is required for these devices to results were obtained using steady helium blowing (no high-
be successful in this application. frequency component) with very low Bc ¼ 0.09%. The suppression
Stanek et al. [41] offered a new explanation for the high- mechanism requires further study, but sample Particle Image
frequency forcing effect. The intrinsic idea was that high- Velocimetry (PIV) images and cross-correlations of pressure–time
frequency forcing alters the instability characteristics such that histories suggest that the injection alters the impingement
the growth of large-scale disturbances is inhibited or prevented. region and disrupts the acoustic feedback loop. Their results
They hypothesized that the mechanism was a decelerated energy also highlight the need to rigorously study high-frequency
cascade in contrast to the findings of Wiltse and Glezer [43]. forcing effects isolated from the mass flow injection from the
In 2003, Stanek et al. [42] reported various aspects of the actuator.
cylindrical rod in crossflow. They studied the vertical position Zhuang et al. [50] investigated the use of a vertically directed
of the rod H/d in the boundary layer, its relative size d/d, microjet array mounted upstream of the cavity leading edge. The
installation issues, and end conditions. They recommended an microjets had a 400 micron diameter and produced sonic jets that
optimal location as centered at the edge of the boundary layer and interact with the upstream boundary layer. The authors show
an optimal size of d/d ¼ 2/3. They argued that their results how, at Mach 2, the microjets alter the cavity shear-layer
conclusively demonstrate that the suppression is due to high- thickness and the receptivity process, introduce streamwise
frequency forcing via vortex shedding from the cylinder. Addi- vorticity, and alter the shear-layer trajectory and the resulting
tional arguments are provided in Ref. [44], which discusses the impingement region. Significant tonal and broadband suppression
local stabilization of the shear layer via high-frequency forcing. levels were achieved with Bc as low as 0.15%, which is significantly
While the cylinder clearly affects the mean flow and its stability lower than the mass injection required in other studies using
characteristics, there are other important factors that cannot be steady blowing. Higher levels of Bc produced no significant
ignored, including experimental evidence presented by Ukeiley improvement.
 ARTICLE IN PRESS

L.N. Cattafesta III et al. / Progress in Aerospace Sciences 44 (2008) 479–502 487

Collectively, the blowing results described above indicate that concern for stores released from internal bays at high subsonic
manually optimized steady blowing configurations with Bcp0.2% speeds [52].
can be effective suppression devices. At subsonic speeds the It is reasonable to assume that reducing cavity resonance, as
primary mechanisms appear to be an alteration of the shear-layer measured by the reduction in prms inside and near the cavity,
stability characteristics, the introduction of streamwise vorticity, would have a favorable effect on the release of stores from internal
and modification of the shear-layer impingement location. While bays. This is because the flow–acoustic coupling that governs
these are also important at supersonic speeds, shock wave/ cavity resonance (see Section 1) implies that a reduction in the
boundary layer interactions at the upstream cavity edge and the dynamic pressures should be concomitant with a decline in the
ensuing shear-layer trajectory alteration appear to be additional flow-field unsteadiness in terms of the fluctuating velocities.
factors that should be considered. Recent studies, by Zhuang et al. [1] of supersonic flow over
It is interesting that when all of the available blowing data is rectangular, three-dimensional cavities (i.e., not spanning the
expressed using the blowing coefficient definition of Vakili and width of the test section), and that of Ukeiley et al. [49] using non-
Gauthier [31], one finds that the evolution of steady blowing rectangular cavity configurations, provide direct evidence of this
techniques has reduced the effective (not necessarily optimal) Bc correlation between the unsteady pressures and the global
from O(10%) by two orders of magnitude down to O(0.1%). Note velocity field. The use of normally oriented, microjet arrays by
that the definition of Bc accounts for the cavity area but does not Zhuang et al., which led to reductions of up to 10 dB in the overall
directly incorporate the scaling effects of the boundary layer. This unsteady pressure level (OASPL) and more than 20 dB in the tonal
has important implications for full-scale applications and is amplitudes, also dramatically reduced the fluctuating velocities
addressed further in Section 3. In comparison, low-frequency inside the cavity.
forcing does not appear to be very attractive when the actuator Fig. 5 from Zhuang et al. [50] illustrates the dramatic effect of
bandwidth is insufficient in comparison to the frequencies of the flow control on the unsteadiness of the global flow field. These
cavity tones. This is discussed further in Sections 3 and 5. measurements were obtained using planar two-component PIV
High-frequency excitation, whether it is passive or active, for Mach 2 flow over an L/D5 cavity. Here, Fig. 5a shows the
appears promising for both tonal and broadband suppression. contour plot of vrms, the unsteady vertical component of the
However, the responsible mechanisms require further study. velocity for the baseline uncontrolled case, while Fig. 5b
There is ample evidence that high-frequency forcing alters the corresponds to the case with microjet control. As seen in Fig. 5a,
mean flow. As a result, the shear-layer stability characteristics are very high rms intensities, as much as 10% of the free stream
altered and, in some cases, the trajectory of the shear layer is velocity, are observed inside baseline cavity. However, the
modified. When the impingement location is altered, the strength microjets reduce the fluctuation intensities in the velocity field.
of the acoustic source is reduced and the broadband noise level Flow control not only reduces the magnitude of fluctuating
decreases. As will be discussed in Section 5, to date closed-loop intensities, by a factor of two or more in some regions, but also
control produces comparatively little change in the mean-flow reduces the spatial extent of the region where the flow is highly
properties and, as such, has only been shown to be effective for unsteady. As shown by Ukeiley et al. [25] and Zhuang et al. [1,50]
tonal control. among others, various open-loop control approaches modify the
mean-flow properties. These include (1) significant changes in the
flow field inside the cavity, especially the reversed flow region;
2.3. Effect of control on the flow field and store trajectory
(2) modification of the cavity shear-layer thickness, especially in
the initial, highly receptive region; and (3) a change in the shear-
As discussed in the introduction, in addition to the increase in layer growth rate and/or its position relative to the trailing edge.
unsteady loads on the various aircraft structures, the highly Although it is a rather intuitive inference that a reduction in
oscillatory flow field generated by cavity flows can also adversely the velocity fluctuations inside the cavity (due to flow control)
affect the safe departure and accurate delivery of stores from should enhance store separation characteristics, the details of the
internal weapons bays. Such internal weapon bays are becoming fundamental mechanisms behind this relationship—including
more common in modern aircraft in large part due to the the influence of changes in the mean-flow field—are still not
decreased aerodynamic drag and the lower radar signatures when completely clear. Some factors to be considered include:
compared to the external storage of weapons in earlier aircraft.
Also, as mentioned earlier, this issue is becoming more pressing
due to the increased reliance on smaller, and thereby lighter, 1. a reduction in the flow unsteadiness provides the store with
munitions that are more susceptible to the large-scale oscillations more repeatable and stable initial conditions compared to the
in cavity flows that can lead to undesirable store separation [19]. uncontrolled case and reduces disturbances that affect the
Although more acute at supersonic speeds [51], this issue is still of store trajectory after it is released;

Fig. 5. Velocity-field measurements showing the effect of control on the flow unsteadiness, from Zhuang et al. [50]. Contours of vrms with (a) no control and (b) microjet
control.
 ARTICLE IN PRESS

488 L.N. Cattafesta III et al. / Progress in Aerospace Sciences 44 (2008) 479–502

2. a change in the mean profile of the shear layer can alter the
forces and moments encountered by the store, especially as it
traverses a finite-thickness shear layer.

Clearly, more fundamental and comprehensive studies are

needed that correlate the flow-field properties—and its response
to control—to store dynamics. These can then be used to gain a
better understanding of the flow physics governing this complex
phenomenon and to develop more robust models that predict
store trajectory. Such an attempt was recently made by Sahoo
et al. [53] who built upon the extensive dynamic models of store
release trajectory developed by Shalev et al. [52]. In particular,
Sahoo et al. [53] developed a low-order model to predict store
dynamics for stores released from internal bays exposed to
supersonic flow. Using slender-body dynamics, their model only
considers pitch and plunge motion of the store. It also contains
several other simplifications—in large part to make the analysis
more tractable—such as assuming the cavity to be quiescent,
which is not always the case as documented in numerous studies
cited in this article.
A noteworthy aspect of their model is that, where available,
experimental data are incorporated to better capture store
dynamics. For example, the effect of shear-layer profile/thickness
is evaluated using velocity-field measurements from Zhuang
et al. [50]. Furthermore, their dynamic model also attempts to
capture the effect of the primary parameters that influence
control efficacy on store trajectory in experimental studies.
For example, the impact of varying Cm, the steady momentum
coefficient, has been incorporated. The model predictions
are finally compared to store trajectory data obtained from
supersonic free-drop tests conducted at the Boeing Polysonic
Wind Tunnel [51]. Although the predictions do not agree with
Fig. 6. High-speed instantaneous shadowgraphs, showing the favorable effect of
the details of the measured store trajectory; the model captures
control on store trajectory (from Bower et al. [51]).
some of the overall trends. More importantly, it does accurately
predict the success or failure of control in achieving a safe
3. Sensors and actuators
store departure from the weapons bay. Considering the many
simplifications inherent in Sahoo’s model, such agreement
implies that there is hope for developing more sophisticated In general, actuators for cavity tone suppression are devices
models that more accurately predict store dynamics in a that act to disrupt some element of the acoustic resonance
quantitative sense. mechanism. Actuators may be a component of a closed-loop
Even though the details of the relationship between cavity flow control system or act independently in an open-loop or passive
oscillations and store trajectory are not well understood, the mode. The cavity tones are suppressed when successful, but in
favorable impact of effective cavity/noise control on trajectory has some cases the broadband noise level is reduced as well [24].
been demonstrated in the free-drop tests as Boeing Polysonic Examples of actuators, their strengths and weaknesses, and the
Wind Tunnel [51]. An example of this can be seen in the sequence issues associated with their application will be discussed in this
of high-speed, instantaneous Schlieren images shown in Fig. 6 section.
[51]. The cavity is located on the top, and the store is ejected
out of the cavity in the downward direction. The images on
3.1. Passive actuators
the left correspond to the baseline or no-control case, while those
on the right correspond to a case where cavity oscillations
(quantified via pressure spectra inside the cavity) have been Passive actuators attenuate tones by changing the character-
significantly reduced using active flow control. Repeated free-drop istics of the shear flow over the cavity. Passive actuators require
tests demonstrate that without flow control, the store develops an few, if any, moving parts and tend to be inexpensive and simple
undesirable pitch-up moment and returns to the cavity (bay) devices. However, they often do not work well at off-design
for most cases. However, the activation of flow control leads to conditions [54,55]. When effective, passive actuators disrupt the
an overall pitch-down moment on the store, leading to a safe resonance through one of at least three mechanisms:
departure.
A better understanding of the store dynamics can be achieved (1) the trajectory of the mean shear layer is changed such that the
only through additional, careful experimental studies that provide reattachment point is shifted downstream of the cavity edge
high-fidelity, time-resolved measurements, such as synchronized [25,45,48],
velocity-field and pressure data or velocity combined with (2) the stability characteristics are modified by changes in the
store trajectory measurements. Such data would enable the shear-layer velocity profiles and/or gas properties [12,56], so
incorporation of more relevant physics into store dynamics that the resonant modes are not amplified, and
and would also serve as a benchmark for validation of CFD (3) the spanwise coherence of the shear layer and corresponding
simulations. Rossiter mode is disrupted [45].
 ARTICLE IN PRESS

L.N. Cattafesta III et al. / Progress in Aerospace Sciences 44 (2008) 479–502 489

Examples of passive actuators include leading-edge ramps It is often a challenge to the actuator designer to maintain, for
[32], spoilers [57], fences [25], steady gas injection [31], or example, velocity fluctuation amplitudes at the same order-of-
contouring of the trailing edge of the cavity [20]. Spoilers and magnitude as the mean flow as the frequency increases. Because
fences are commonly installed on production aircraft to reduce these devices have a strong mean component, it is often difficult
the resonant tones in weapons bays and deploy when the bay to separate which component (i.e., mean or oscillatory) of
doors open. The fences act to increase the shear-layer thickness, actuation is responsible for the flow control.
which shifts the most unstable shear-layer frequencies to lower As suggested by Stanek et al. [19] actuators can be further
values. Spoilers and ramps deflect the mean separation streamline categorized into low-frequency excitation and high-frequency
higher into the flow so that reattachment occurs downstream of excitation (‘‘hifex’’). Hifex corresponds to forcing at an order-of-
the cavity trailing edge. This weakens the feedback acoustic wave magnitude larger than the frequencies of the resonant tones,
and the resulting strength of the Rossiter modes. while low-frequency excitation corresponds to frequencies that
Similarly, as shown in Fig. 7, rods placed in the upstream are an order-of-magnitude or more less than the resonant tones.
boundary layer produce a mean wake (momentum deficit), which Sarno and Franke [32] proposed the concept of forcing the shear
modifies the mean shear-layer development [25,42,45]. Ukeiley et layer at a frequency different from that of the Rossiter mode as a
al. [25] studied both rods and variable-height fences and found way to suppress resonance. Because their actuators were limited
the effect of the device on the mean gradient of the shear layer to to frequencies an order-of-magnitude lower than that of the first
be important in determining the level of attenuation. Rossiter mode, the results did not provide convincing evidence
that the low-frequency forcing approach could be effective. By
using a piezoelectric flap, Cattafesta et al. [35] were the first to
3.2. Active open-loop actuators
clearly demonstrate that forcing the shear layer to oscillate at a
frequency different from that of the Rossiter modes could result in
Actuators that require energy to operate and, in turn, add
noise attenuation in a low-speed cavity flow. Provided the
energy to the flow are defined as active control devices. The term
excitation frequency was not in a narrow band near a Rossiter
‘‘open loop’’ emphasizes that a feedback signal is not used to
mode, in which case lock-on resonance occurred, the piezoelectric
control the actuator output. Examples include oscillating electro-
flaps were able to attenuate the mode by exciting shear-layer
mechanical [32,57,58] and piezoelectric flaps [19,35,59–62],
instabilities incommensurate with the Rossiter resonance me-
steady blowing [32,48,63,64], and pulsed blowing [27,36,65,66],
chanism. The non-Rossiter shear-layer modes grow at the expense
voice-coil drivers [67,68], powered resonance tubes [19,23,69,70],
of the natural Rossiter modes, indicating a nonlinear process. The
fluidic oscillating jets [39], and plasma actuators [71]. All of these
effectiveness of this approach at higher Mach numbers is
open-loop actuators have demonstrated control of cavity reso-
questionable due to the increasingly larger mean-flow energy
nance at subsonic flow conditions, but only the powered
available [66]. Wider bandwidth voice-coil type actuators have
resonance tube and the steady and pulsed jets and microjets
been used by Little et al. [73] to explore the effects of open-loop
have been successful at supersonic free stream conditions.
sinusoidal forcing at Mach 0.3.
It is important to recognize that a mean component of forcing
Actuation at frequencies an order-of-magnitude larger than the
is associated with most unsteady actuators. Even in the case of
Rossiter modes can also lead to suppression of the cavity tones via
actuators that have no net mass addition, there will be a net
hifex. In particular, the technique pioneered by McGrath and Shaw
momentum flux associated with the second-order streaming
[24] and revisited by others has suppressed tones at supersonic
effect [72]. Streaming is the steady, secondary flow component
flow conditions. One hifex hypothesis [19] argues that energy
resulting from the quadratic nonlinear interaction of the unsteady
addition to the shear layer at length scales much smaller than the
flow components. Whether this effect is significant or not depends
coherent shear-layer vortices will directly increase dissipation and
on the amplitude of the oscillations. Actuators with a nonzero
accelerates the turbulent energy cascade. Based on the results
mass addition, such as pulsed jets, siren valves, powered
of numerical simulations of oscillatory (i.e., ac only) and pulsed
resonance tubes, fluidics, and whistles will have first-order
(i.e., dc and ac) jets, Stanek [44] argues that hifex results in a local
mean-flow components with disturbance magnitudes that are
stabilization of the cavity shear layer to low-frequency perturba-
comparable to or exceed the amplitude of the unsteady component.
tions and, in turn, cavity oscillations. Stabilization is observed
to occur only when the pulsing frequency is above a critical
threshold, in which case the input pulse rapidly decays.
Note that hifex can be achieved via passive control with rods
mounted in the upstream boundary layer, as shown in Fig. 7, due
to passive high-frequency vortex shedding or via active control
(with, for example, powered whistles and powered resonant tubes
shown in Fig. 8 and described in Refs. [56,74]). A typical hifex
experiment will produce disturbances in the range of 5 kHz or
higher in order to suppress tones in the 500 Hz range or lower.
However, the physics of the hifex effect are difficult to sort out,
because all hifex actuators realized to date also have a substantial
effect on the mean flow that can influence the shear-layer
development, its impingement location, and the acoustic source.
The mean-flow-modification effect using rods in crossflow has
been demonstrated by Ukeiley et al. [25] and Arunajatesan et al.
[45]. As mentioned above, Stanek [44] and, even earlier, Rizzetta
and Visbal [75] have shown, via large-eddy simulations in a Mach
1.19 cavity, that pure high-frequency oscillatory forcing can
suppress cavity oscillations, but no reported experiments have
Fig. 7. Photograph of ‘‘CALSPAN95’’ cylindrical rod installed at the cavity leading been able to corroborate this hypothesis. Hence, this remains an
edge (from Stanek et al. [42]). The vertical bay doors are deployed. open question and warrants additional research.
 ARTICLE IN PRESS

490 L.N. Cattafesta III et al. / Progress in Aerospace Sciences 44 (2008) 479–502

feedback is used to control the flow, then the dynamics of the flow
system are changed. Unlike open-loop control, the transfer
function of the actuator plays a major role in how the overall
system will behave. For example, the bandwidth of the actuator
plays a crucial role in determining the maximum achievable
performance from the closed-loop system, as a consequence of the
area rule discussed in Section 4.4. The bandwidth should ideally
be large enough to suppress more than one Rossiter mode.
Otherwise an actuator with small bandwidth cannot attenuate a
resonant peak without amplifying undesirable neighboring modes
[81]. For this reason, it is important to know the actuator transfer
function when designing a closed-loop control algorithm. More
details will be given in Section 4, and techniques to estimate
the transfer function for piezoelectric flaps and synthetic jets
Fig. 8. Photograph of an array of powered resonance tubes (PRT) installed at the
leading edge of the cavity in a similar installation as in Fig. 7. The inset shows a are described, for example, in Refs. [59,91], respectively.
schematic of a single actuator. Adapted from Stanek et al. [74]. Initially the closed-loop actuator must have enough power to
produce a disturbance that exceeds the receptivity-induced
Regardless of the type used, active open-loop actuators are perturbation at the upstream end of the cavity. After the system
attractive because of their relative simplicity and ability to be responds to the actuation, the actuator power requirements
activated when needed. However, Shaw and Northcraft [27] decrease with the decreasing tone amplitudes. Cattafesta et al.
demonstrated the sensitivity of the control effect to forcing [35] demonstrated the initial high-amplitude piezoelectric flap
frequency and changing flow conditions. Open-loop actuators, like actuator output (see Table 2) reduced by an order-of-magnitude
passive actuators, must be optimized for each flow condition. after control was established.
Furthermore, Cattafesta et al. [35] demonstrated with piezo- To date there has not been a demonstration of closed-loop
electric flap actuators at the cavity leading edge that an order-of- control (Type A) at supersonic speeds. Shaw and Northcraft [27]
magnitude higher power is required to drive open-loop actuators demonstrated that both open-loop and Type B closed-loop control
compared to closed-loop systems for equivalent reductions. were capable of controlling the tones and reducing the broadband
Similar power penalties with other open-loop actuators are noise levels with pulsed-fluidic injection at supersonic speeds.
expected. The challenge associated with closed-loop control is to achieve
similar noise reductions with an order-of-magnitude lesser input
power by using Type A actuators. Actuators with large amplitudes,
3.3. Active closed-loop actuators high bandwidth, and fast time response are therefore required,
which raises the issue of actuator scaling. Shaw [92] studied the
Actuators for closed-loop control form part of a system that mean mass flow rate requirements for a pulsed injection system
includes at least one flow state sensor and a feedback control directed perpendicular to the free stream and determined that the
algorithm. This approach is the most expensive in terms of steady momentum coefficient C m ¼ mU _ jet =ðq1 W dÞ collapsed two
hardware and complexity, but it offers the greatest adaptability sets of data at Mach 0.95, with frequencies ranging from 100 to
to changing flow conditions and potentially the lowest power 600 Hz and widely varying d and qN. Note that the area in the
consumption. A summary of some actuators used in closed-loop denominator is the width of the cavity multiplied by the boundary
cavity control experiments is provided in Table 2. Again, the layer thickness at the leading edge of the cavity. This scaling
investigations are listed in chronological order for historical parameter was used successfully for weapons bay cavity flow
purposes. control in the flight test of an F-111 [93]. More recently, in
Closed-loop actuators for active control of cavity oscillations experiments ranging from small (model scale o3%) laboratory
also fall into two types. A ‘‘Type A’’ actuator is a device with scale tests to large industrial scale models, where model scale is
sufficient bandwidth that is capable of producing, at any instant in 410% of full scale, Zhuang et al. [50], Ukeiley et al. [49], and
time, a control input consisting of several frequencies, each with Bower et al. [51] have determined that the steady momentum
its own amplitude and phase. This type of actuator has a time coefficient is the most relevant scaling parameter for fluidic
response commensurate with the time scales of the cavity flow actuator design. It appears to capture the fundamental parameters
dynamics so that it can be used in a dynamic feedback governing the efficacy of actuators in reducing the cavity dynamic
compensation scheme. A voice-coil actuator is an example of loads, as well as the flow unsteadiness.
a Type A actuator (see Table 2). A ‘‘Type B’’ actuator is also a For the purposes of Type A flow-control actuation with
broadband device, but it produces a control signal of prescribed piezoelectric flaps or with zero-net mass-flux actuators, the
amplitude and primarily one fundamental frequency (with momentum coefficient will need to include the fluctuating
perhaps harmonics and a mean-flow component) at any instant component of velocity, presumably as an oscillatory momentum
in time. The actuator frequency can change only on a time coefficient cm ¼ rjet u2rms Ajet =ðq1 W dÞ. But the problem of designing
scale that is large compared to the time scales of the cavity a wide bandwidth actuator that works effectively at high subsonic
flow dynamics. Shaw and Northcraft [27] successfully used this Mach numbers is exacerbated by the quadratic dependence of the
approach with a rotary-valve, pulsed-blowing actuator. At any dynamic pressure q1 ¼ 0:5gp1 M 21 on Mach number. In conven-
instant in time, the rotor spins at a particular rotation rate (rpm), tional pressure-driven tunnels, where increasing subsonic Mach
and the supply pressure has a certain value. These can both be numbers are achieved by increasing the stagnation pressure, pN is
changed via, for example, a control voltage to a dc motor and approximately constant and is on the order of the atmospheric
servo valve, but the slow time response due to its finite inertia pressure. Because the tones are stronger than the broadband
precludes a rapid change in the actuator output or the ability to noise, the actuator authority must also increase with Mach
phase lock with the acoustic field. number in a similar manner. For example, using cm-based scaling,
The majority of closed-loop, flow-control experiments sum- the dynamic pressure increases by a factor of nine as the Mach
marized in Table 2 use Type A actuators [35,68,76–90]. When number increases by a factor of three for a fixed static pressure.
Table 2
Summary of selected actuators used for closed-loop control

Study Conditions Method of actuation Control approach Power (W) Comments

Cattafesta et al.  UN ¼ 40m/s, L/D ¼ 0.5, L/  Unimorph piezoelectric flaps at leading edge  Type A  Comparison of open-loop voltage and
[35] W ¼ 0.5, Rey ¼ 4750, L/y ¼ 81  ftypE300 Hz closed-loop voltage requirements
 UN ¼ 45 m/s, L/D ¼ 2.0, L/  23 mm/V @ 300 Hz
W ¼ 2.0, Rey ¼ 5210, L/y ¼ 328 spring piezoceramic
 W ¼ 12 in, D ¼ 6, 12 in metal shim

base
screw/post

L.N. Cattafesta III et al. / Progress in Aerospace Sciences 44 (2008) 479–502

Shaw and  MN ¼ 0.6–1.05  Pulsed fluidic at leading edge, 901 w.r.t. flow  Type B
Northcraft  L/D ¼ 6.46, L/W ¼ 3.67  Frequencyo650 Hz  Mass flow and frequency adjusted based on
[27] rms levels within 3 frequency bands

ARTICLE IN PRESS
Cattafesta et al.  MN ¼ 0.4–1.35  Piezoelectric flaps at cavity leading edge  Type A  Feedback signal is pressure near leading
[29]  L/D ¼ 5  fresE475 Hz  Adaptive disturbance rejection controller edge of cavity
 2.5 mm/V @ 500 Hz
 1.875 mm max disp.

Mongeau et al.  UN ¼ 14–29 m/s  Oscillating spoiler hinged at leading edge of  Type A o100  Digital control using Simulink& Real-
[57]  Partially closed cavity, 1/5 scale cavity opening  Loop shaping of open-loop transfer functions Time Workshop
car model  Voicecoil driver measured on system components
 Max. 1 mm displacement @ 120 Hz

Williams et al.  MN ¼ 0.25–0.55  Zero-net-mass flow oscillations through  Type A 0.33–53  Feedback signal from cavity floor at x/
[28]  L/D ¼ 5 leading edge slot, 01 w.r.t. flow  Model-based control with 2nd-order bandpass depending on L ¼ 0.8, 0.875
 Two 500 W, 8 in speakers filters control
 ftypE340 Hz

Ziada [77]  MN ¼ 0.3  Synthetic jet through leading edge slot, 451  Type A
 L/D ¼ 2.5, 4 w.r.t. flow  Time delay and gain applied to feedback signal
 L ¼ 127, 203 mm  Two 50 W, 100 mm diameter speakers
 D ¼ 50.8 mm, W ¼ 76 mm

491
 492
Table 2 (continued )

Study Conditions Method of actuation Control approach Power (W) Comments

Kegerise et al.  MN ¼ 0.275  Bimorph piezoelectric flap at cavity leading  Type A 1

[84–86]  L/D ¼ 5 edge  Generalized predictive control
 fn1200 Hz
 DC gain0.25 mm/V

L.N. Cattafesta III et al. / Progress in Aerospace Sciences 44 (2008) 479–502

Cabell et al.  MN ¼ 0.275, 0.35, 0.45  Piezoelectric-driven type of synthetic jet  Type A
[83]  L/D ¼ 5 actuator with 44.5 mm  0.5 mm 2-D slot  Linear Quadratic control design using the
 L/W ¼ 3 orifice. state-space models computed from
 Peak velocity at slot exit is X15 m/s over experimental data.
60–1500 Hz.

ARTICLE IN PRESS
Debiasi and  L ¼ 50.8 mm, W ¼ 50.8 mm  Zero-net mass-flux forcing of the shear layer at  Type A
Samimy  L/D ¼ 4 the leading edge through a slot  Logic-based controller with forcing frequency
[76,87]  d ¼ 2.5 mm (50.8 mm  1 mm) at 301 w.r.t. flow between 2 and 5 kHz and fixed amplitude
 Red ¼ 2  104, MN ¼ 0.25–0.5  A compression driver with bandwidth 1–20 kHz

main flow
control flow

kulite
transducer

compression
driver

Micheau et al.  L ¼ 800 mm, W ¼ 150 mm  A vibrating surface (3 cm by 3 cm) with a shaker  Type A
[107]  L/D ¼ 5.3 located at the trailing edge  Envelope controller
 ReL ¼ 3  105, MN ¼ 0.11

Rowley et al.  L ¼ 152.4 mm, W ¼ 76.2 mm  Zero-net mass-flux forcing at the leading edge  Type A
[129]  L/D ¼ 2  200 W compression driver  Heuristic feedback control law
 MN ¼ 0.6, L/y ¼ 52.8  Exponential horn design
 ARTICLE IN PRESS

L.N. Cattafesta III et al. / Progress in Aerospace Sciences 44 (2008) 479–502 493

This implies that the actuator rms velocity fluctuation level will sonic, cavity flows using a combination of fluctuating pressure
need to increase by a factor of three to be effective. Perhaps a measurements and phase-conditioned measurements of the
more realistic scenario for actuator development would employ a density field inside the cavity using a quantitative, non-intrusive
wind tunnel that allows independent control of Mach number and Schlieren instrument. The quantitative Schlieren technique (and
qN to simulate the actual flight environment. Until such actuators hot wires) was used to study nonlinear mode interactions
and experiments are available, numerical simulations can and and mode switching [28,102]. Zhuang et al. [50] also used a
should be used to study virtual actuator scaling requirements. combination of Schlieren and shadowgraph methods along with
Finally, we note a reported discrepancy for the optimal unsteady surface pressure measurements to study supersonic
injection angle. As summarized in Table 1, Shaw [36] has found cavity flows. Such data provide invaluable information regarding
that vertical injection is superior for pulsed blowing at low the spatio-temporal nature of the events that dominate the cavity
frequencies, while Williams et al. [28] and Kegerise et al. [84] find dynamics and can provide insight concerning nonlinear interac-
horizontal injection is superior for Type A zero-net mass-flux tions and model switching.
actuators. Stanek et al. [41] have found that steady microjets have Similarly, Forestier et al. [103] have studied transonic
been effective only in vertical injection configurations. The reason (Mach 0.8) flow over deep cavities (L/D0.4) using high-speed
for this observed discrepancy is unclear and again warrants Schlieren photography to visually examine the periodic compo-
further careful studies. nent of the cavity shear-layer oscillations. In addition, they
obtained phase-locked laser Doppler velocimetry measurements
to examine the evolution of periodic and spatially coherent
3.4. Sensors and flow measurements structures (or vortices) extracted from the velocity-field data.
Murray and Elliott [104] have used Schlieren photography and
Since the aim of most cavity control experiments is to reduce planar laser imaging of supersonic cavity flows (1.8–3.5) to study
the pressure fluctuations, unsteady pressure sensors are com- the characteristics of cavity shear-layer structures. Little et al. [73]
monly used in laboratory control applications. Lou et al. [94] used and Samimy et al. [89] have used phase-locked planar laser
microphones, with their lower noise floor and reduced full-scale imaging to visualize the organized structure of various baseline
range, to measure (and control) the flow-induced pressure and open-loop controlled cavity flows at Mach 0.3.
oscillations due to an impinging jet. But the unsteady pressure Some investigators have used PIV with varying degrees of
levels inside a cavity often exceed the linear full-scale range success to study high-speed cavity flows [50,56,89,105]. Most of
of microphones, forcing the use of unsteady pressure transducers. the difficulties are associated with proper seeding of the cavity
By far the most common sensors used to examine cavity flows flow. Zhuang et al. [50] have obtained velocity and vorticity fields
are miniature, high-frequency-response piezoresistive pressure in a supersonic flow and have examined the effects of microjet
transducers, such as those made by Kulite, Endevco, and PCB control (see Fig. 5).
Piezotronics. With regard to safe separation of stores from military aircraft,
These transducers are usually flush mounted or recessed one wishes to predict the trajectory, based on a simple measure-
beneath a pinhole—to achieve better spatial resolution at ment, such as the fluctuating pressures inside the cavity. To
frequencies well below the Helmholtz frequency of the recessed accomplish this, correlations between measurements made inside
cavity/pinhole combination. They are usually mounted on one of the cavity and on the store are required. Such a database can
the cavity surfaces, generally the floor and/or the leading and be created through captive trajectory tests, where forces and
trailing surfaces. Alternatively, they are placed in the tunnel walls moments on the store are measured simultaneously with the
[89] near the cavity to measure the near acoustic field (o one unsteady pressures inside the cavity. Another approach is the use
acoustic wavelength from the trailing edge). The small size, of sensors, such as accelerometers, gyros, and inclinometers,
linearity (output voltage linearly proportional to input pressure), which measure the pertinent store trajectory parameters. Stores
their flat-frequency response over a large frequency range, and embedded with such sensors combined with telemetry data
their high dynamic range (ratio of maximum-to-minimum systems can then be used to obtain and correlate the signals.
detectable pressure) make them an excellent tool for characteriz- Further discussion on the effect of control on store trajectory
ing the cavity dynamics. Such transducers have been used to and recent efforts on modeling this behavior were discussed in
examine subsonic [95], supersonic [50], and hypersonic [96] Section 2.3.
cavity flows. They have been used in small-scale, laboratory
facilities and in larger commercial testing facilities [41,94].
In order to better understand the global flow behavior, one 4. Closed-loop control methodologies
must look beyond the surface pressure or acoustic field. The
dynamics of cavity flows have been examined using a number of The active control suppression studies described in Section 2
other diagnostic tools. For example, relatively low Mach number used open-loop techniques. Here, we summarize several studies
cavity flows have also been examined using constant-temperature (see Table 3) that used closed-loop control, which incorporates
hot-wire anemometry. Shear-layer profiles (mean and fluctuating) some type of feedback from a sensor placed in the flow. As
and, in some cases, instability growth rates have been obtained. categorized in the diagram in Fig. 3, one way to use this feedback
Examples can be found in Mendoza and Ahuja [33], Hsu and Ahuja is to ‘‘tune’’ an inherently open-loop approach, for instance by
[34], Cattafesta et al. [35], Garg and Cattafesta [97], Kegerise [98], slowly modulating the frequency and amplitude of open-loop
and Williams and coworkers [28,37,66,67,80,95]. forcing (labeled quasi-static) to achieve the best suppression in a
Hot-wire anemometry becomes increasingly problematic in quasi-static or time-averaged sense. A different way is to use
high-speed flows due to wire breakage and compressibility effects feedback at the time scale of the dynamics (labeled dynamic). This
[99]. Consequently, other techniques have been used. For example, latter approach can have even more beneficial effects, such as
Little et al. [73] and Samimy et al. [89] have used hot-film probes. even lower power required for control.
Vakili et al. [100] used a multi-hole probe to obtain upstream Open-loop control cannot alter the dynamics of a system
boundary layer profiles as a function of mass injection. Kegerise (e.g., stabilize an instability) except by exciting nonlinearities
[98], Kegerise et al. [101], and Garg and Cattafesta [97] (e.g., by modifying the mean flow). This implies a large power
characterized the spatio-temporal behavior of high-speed, sub- requirement, either as power explicitly supplied to an actuator, or
 494
Table 3
Summary of selected closed-loop cavity suppression studies

Study Conditions Method Comments

Gharib [106]  UN ¼ 22 cm/s,  Strip heater used to excite TS waves via Joulean heating  Feedback control using manual gain and phase
ReD ¼ 24,000, adjustment reduced rms fluctuations by factor of 2
laminar BL H ¼ 2.5,
Rey ¼ 95
 L/y ¼ 66, 77, 82

Cattafesta et al. [35]  L/D ¼ 0.5, L/W ¼ 0.5,  Used piezoelectric flaps mounted at the leading edge of the cavity to suppress low-speed cavity oscillations  CL control suppressed oscillations with order-of-
UN ¼ 40 m/s,  Used LQG and pole-placement feedback control designs magnitude lesser input actuator power
L/y ¼ 81  Shear-layer meas. w/ & w/o control show no effect on
 L/D ¼ 2.0, L/W ¼ 2.0, mean flow
UN ¼ 45 m/s,
L/y ¼ 328

L.N. Cattafesta III et al. / Progress in Aerospace Sciences 44 (2008) 479–502

Mongeau et al.  UN ¼ 15–29 m/s  Used active spoiler driven via a moving coil loudspeaker (up to 1 mm or less that 11)  Significant attenuation achieved with small actuation
[78,79] flow over a  Used robust loop shaping algorithm effort
Helmholtz resonator  Robust performance for transient operating conditions

Kestens and Nicoud  Laminar BL with  2-D NS simulations  Demonstrated suppression only in vicinity of
[109] Red ¼ 1000 and  Used off-line system identification for system transfer function model microphone

ARTICLE IN PRESS
L/D ¼ 2 for  Used filtered-X algorithm  No effect on flow (active noise control, not active flow
MN ¼ 0.2 control)

Cattafesta et al. [29]  L/D ¼ L/W ¼ 5  Used piezoelectric flaps flush at the leading edge of the cavity  Demonstrated successful single-tone suppression at
 MN ¼ 0.4, 0.6, 0.85  Used adaptive disturbance rejection algorithm & system identification MN ¼ 0.74, L/D ¼ 4, L/W ¼ 3 small-scale test but
 Calculated negligible suppression in larger-scale wind tunnel
d ¼ 10.5 mm @ tests
MN ¼ 0.6,
Re/m ¼ 11 106

Shaw and Northcraft  L/D ¼ 6.46,  Tested pulsed fluidic injection at the cavity leading edge using closed-loop control via rotary valve  Controller optimizes injection mass flow rate and
[27] L/W ¼ 3.67  Quasi-steady tuning of open-loop fixed-frequency forcing based on bandpass-filtered rms pressure signal frequency
 MN ¼ 0.6, 0.85, 0.95,  Suppression of tones demonstrated but new tones
1.05 appear at excitation frequency

Williams et al. [28]  L/D ¼ 2,4, or 5,  Used zero-net mass-flux unsteady bleed actuator to suppress or enhance individual resonant cavity modes  Experiments showed 01 forcing was better than 451 or
L/W ¼ 1.33  Used analog output feedback 901 (vertical)
 MN ¼ 0.2–0.55,  Multiple-mode suppression at MN ¼ 0.48
d ¼ 3.18–2.41 cm,  Studied nonlinear mode interactions
y ¼ 0.30–0.26 cm

Williams et al. [67]  Similar conditions as  Used zero-net mass-flux unsteady bleed actuator used to suppress or enhance individual resonant modes  Found single-mode resonance can occur if cavity
above  Used analog output feedback control circuit mode coincides with Rossiter mode
 BL data: at  at Mach 0.35, input power from 6.3 W (suppressed
MN ¼ 0.2–0.55, oscillations) to max. of 53 W
d ¼ 3.18–2.41 cm,  Could suppress or enhance tones
y ¼ 0.30–0.26 cm

Williams and  Similar to above  Used a commercial adaptive digital controller from Arbor Scientific as a follow-on to earlier work  Comparable results for single-mode analog controller
Morrow [95]  L/D ¼ 5, L/W ¼ 1.33  Used filtered-X LMS algorithm suppression but unable to suppress multiple modes
 MN ¼ 0.25–0.55 simultaneously
 Suggested need for adaptive plant model

Williams et al. [80]  Similar to above  Companion papers to study model-based control of cavity oscillations  Cavity can exhibit limit-cycle behavior or act like a
& Rowley et al.  L/D ¼ 5, L/W ¼ 1.33  Experiments used feedback control to suppress the mode and then perform frequency-response exp. stable noise amplifier
[81,82]  MN ¼ 0.34,  Feedback controller is 2nd-order Butterworth filter with different bandwidths, gain, and time delay  Discussed peak splitting via Nyquist analysis,
fundamental performance limitations of feedback
control using area rule
Cabell et al. [83]  MN ¼ 0.275, 0.4, 0.6  Application of discrete-time, Linear Quadratic control design methods to the cavity tone problem  System model required 150–200 states
 L/D ¼ 5, L/W ¼ 3 s at  Used ERA to obtain state-space model via system identification  Control order reduced to 60 states using balance
MN ¼ 0.275,  Piezosynthetic jet actuator at leading edge directed parallel to free stream realization truncation
d5–6 mm  Frequency-shaping technique used to restrict
controller frequency range
 Peaking and peak splitting observed—features
explained via linear models

Kegerise et al. [84]  MN ¼ 0.275, 0.4, 0.6  Used piezoceramic bimorph flap actuator to suppress multiple cavity tones using output feedback control  Demonstrated multiple modes
 L/D ¼ 5, L/W ¼ 3 s at  Assessed IIR- and FIR-based system identification plant models  Found required actuator tip motion for suppression
MN ¼ 0.275, varied by 6–17 wall units
d5–6 mm  FIR model not suitable for plant model, IIR model
required

Debiasi et al. [87]  L ¼ 50.8 mm,  Zero-net-mass-flux forcing of the shear layer at the leading edge through a slot (50.8 mm  1 mm) of 301 w.r.t.  The peak pressure fluctuations were suppressed up to
W ¼ 50.8 mm flow 23 dB
 L/D ¼ 4  A compression driver with bandwidth 1–20 kHz  Effectiveness of the controller was reduced at high

L.N. Cattafesta III et al. / Progress in Aerospace Sciences 44 (2008) 479–502

 d ¼ 2.5 mm  Logic-based controller with forcing frequency between 2 and 5 kHz and fixed amplitude Mach number (above 0.4) flow presumably due to the
 Red ¼ 2  104, lack of actuation authority
MN ¼ 0.25–0.5

Micheau et al. [107]  L ¼ 800 mm,  A vibrating surface (3 cm by 3 cm) with a shaker located at the trailing edge  Single peak reduction can achieve up to 25 dB
W ¼ 150 mm  Envelope controller

ARTICLE IN PRESS
 L/D ¼ 5.3
 ReL ¼ 3  105,
MN ¼ 0.11

Kegerise et al.  L ¼ 152.4 mm  Piezoelectric bimorph cantilever beam at leading edge  Demonstrated multiple Rossiter mode suppression at
[85,86]  W ¼ 50.89 mm  Bandwidth 1 kHz fixed Mach numbers ranging 0.275–0.38
 L/D ¼ 5  fn1200 Hz  Concluded that the disturbances entering the cavity
 d ¼ 6 mm  DC gain 0.25 mm/V flow were collocated with the control input at the
 Off-line system ID cavity leading edge
 Adaptive generalized predictive control

Yan et al. [88] Similar conditions as in  Linear controllers such as HN, Smith predictor, PID, and PID-based parallel proportional with time delay  Controller design was based on the linear model
Debiasi et al. [87] controllers were implemented introduced in Williams et al. (2002) and Rowley et al.
(2002)
 Rapid switching occurred between the Rossiter modes
by the forcing of linear controllers
 Adding a zero to the linear controller can cancel the
newly excited Rossiter mode

Samimy et al. [89] Similar conditions as in  Linear Quadratic optimal controller  Controller design was based on reduced-order models
Debiasi et al. [87] derived using POD method along with the Galerkin
projection method
 1000 Simultaneous PIV-pressure measurements were
used for POD calculations

Efe et al. [132] Similar conditions as in  Direct and indirect synthesis of the neural architectures to identify and control the system  The performance of the neurocontroller was not
Debiasi et al. [87] satisfactory
 1.6Vrms produced a 20 dB reduction of the resonant
peak at the expense of the excitation of a large
sideband peak

Rowley et al. [129]  L ¼ 152.4 mm,  Feedback control was introduced into 2-D direct
W ¼ 76.2 mm  Zero-net-mass-flow forcing at the leading edge numerical simulations
 L/D ¼ 2  One 200 W compression driver with 900 W amplifier
 MN ¼ 0.6  Exponential horn design
 L/y ¼ 52.8  Heuristic feedback law

495
 ARTICLE IN PRESS

496 L.N. Cattafesta III et al. / Progress in Aerospace Sciences 44 (2008) 479–502

as a drag penalty. By contrast, closed-loop control can linearly 4.2.1. System identification
stabilize a system (i.e., with ‘‘infinitesimal’’ actuation), and Several studies have determined a model for the cavity flow
furthermore it can reduce the amplification of external distur- empirically using input/output data. The general approach is
bances, such as boundary layer turbulence or external acoustic to force the cavity system, comprising the actuator(s), cavity,
waves. In addition, adaptive techniques may be used to tune the sensor(s), and any signal conditioners (e.g., amplifiers, filters),
controller’s behavior in real time. over a broad range of frequencies, measure the response from the
sensor, and then either determine an empirical transfer function
from spectra, or use an adaptive filter approach [108] to tune
4.1. Quasi-static vs. dynamic controllers coefficients in a filter such that the mean squared error signal
between the filter and the data is minimized (see Fig. 9). In the
The first closed-loop approaches were modifications of open- first case, the model is a frequency-response function, which may
loop strategies. We categorize these strategies as quasi-static, be expressed via Bode or Nyquist plots. In the second case, a time-
because in these approaches, the time scales by which feedback domain discrete filter is used. In either situation, a linear model
acts are much slower than the time scales of the flow. In perhaps takes the form of a rational transfer function, whose coefficients
the first known closed-loop cavity control experiments, Gharib are ultimately determined either off-line via a least-squares
[106] used periodically forced strip heaters to excite Tollmien– approach or on-line in a recursive fashion, for instance using a
Schlichting waves in the boundary layer upstream of a cavity. By gradient descent algorithm such as least mean squares (LMS) or
feeding back a velocity measurement from a downstream location recursive least squares (RLS) [108]. Once a model is obtained,
in the cavity, and phase locking the sinusoidal forcing to this many standard tools for control system design may be used.
measurement, they obtained a 40% reduction in velocity fluctua- Several approaches are described below.
tions. As an example of the frequency-response method, Mongeau
Shaw and Northcraft [27] also used feedback to modulate a et al. [78] and Kook et al. [79] estimated open-loop transfer
sinusoidal forcing function. They measured the unsteady pressure functions for low-speed flow past a Helmholtz resonator at several
level in a bandpass-filtered signal and then used an iterative different flow velocities. A similar approach was used by Rowley
search algorithm to adjust the frequency and amplitude of a and Williams [68], who obtained an empirical transfer function
pulsed-jet actuator and achieved a significant suppression of the for a compressible cavity flow. An important difference in this
tones, along with some suppression of broadband noise. Micheau work was that a manually tuned controller was used to first
et al. [107] used a feedback envelope controller to drive a vibrating stabilize the oscillations prior to performing the frequency-
surface located at the trailing edge of the cavity. In their response experiment. The frequency-response experiment was
experiment, the single peak resonance was reduced by up to performed on the stabilized system, and then the effect of the
25 dB. Debiasi and Samimy [87] also used an adaptive learning known controller was removed. The reason for this approach is
approach, in which open-loop forcing is applied at a particular that a frequency-response experiment makes sense only for stable
frequency, and this frequency is automatically adjusted by a linear systems. If the system is unstable, with the amplitude of
learning algorithm to obtain the best suppression. oscillation limited by nonlinearities, then it is not clear how to
The class of closed-loop control that we call dynamic interpret the results of the frequency-response experiment. The
controllers uses feedback at the time scales of the unsteady distinction between a lightly damped, but stable cavity system, vs.
motion of the fluid. This type of control is most amenable to a self-sustained cavity oscillation is discussed further in Ref. [13].
techniques from classical and modern control theory and has Several system identification techniques have been used to
some distinct advantages over modulated open-loop techniques, determine models in the form of rational transfer functions or
as discussed at the beginning of this section. Since the majority of state-space representations. Most of these involved frequency-
recent closed-loop control studies have used dynamic controllers, response experiments on the uncontrolled cavity, and as men-
the remainder of this section focuses on these. tioned above, if the linearized system is indeed unstable, and the
Nearly all of the dynamic closed-loop cavity control studies to
date have used linear control techniques. While this might seem
restrictive, it is actually reasonable to expect that linear
controllers would perform well. The reasons for this are twofold. Desired Output
First, recent experiments have indicated that in some regimes, Unknown d
purely linear mechanisms can describe even finite-amplitude Input System
oscillations of the cavity [80–82]. Second, even in regimes in u
which nonlinearities play a role in the naturally oscillating cavity,
one hopes that in the controlled cavity, the oscillations will be
small, and thus linear models will remain valid.
Filter
Output _ +
4.2. Models Adaptive y
Filter Σ
Most linear control approaches rely on an accurate mathema-
tical model of the system to be controlled. In this setting, the
cavity flow is viewed as an input–output system, where the input
is, for instance, the voltage signal supplied to the actuator, and the
Error
output is a sensor measurement. There may be multiple sensors, e
or even multiple actuators, in which case the input and/or output
are vectors. Many different modeling techniques have been used Fig. 9. Schematic of system identification approach using adaptive filters to obtain
a low-order dynamical system model. A broadband input signal is supplied to the
in recent years, either based on flow physics or empirically
unknown cavity plant and the filter with unknown coefficients. The error signal
identified directly from an experiment, and we describe some of from the filter is fed back to adjust the filter coefficients to minimize the mean
these techniques below. square error signal.
 ARTICLE IN PRESS

L.N. Cattafesta III et al. / Progress in Aerospace Sciences 44 (2008) 479–502 497

oscillations are self-sustained, then the meaning of such experi- scalar-valued POD modes. However, to use vector-valued modes,
ments is not clear. one must choose an inner product that appropriately weights the
Kestens and Nicoud [109] used a filtered-X LMS algorithm to thermodynamic variables (e.g., density) and kinematic variables
determine a model for a two-dimensional Navier–Stokes simula- (e.g., velocity). See Rowley et al. [116] for a thorough description of
tion of a forced cavity flow. The model was determined with no the method for compressible flows.
flow, so effects such as shear-layer convection and amplification The result of the Galerkin projection is a set of nonlinear
would not be captured by such a system identification. ordinary differential equations. A key issue is that the standard
Cattafesta et al. [35] used an off-line least-squares method to Galerkin model does not explicitly include the control input. This
identify the parameters in a discrete-time difference equation, limitation can be addressed in various ways, and Samimy et al.
and then used an eigensystem realization algorithm to convert [89] provide an example application of a spatial subdomain
this to state-space form [110]. Cabell et al. [83] used a similar separation method in a subsonic cavity flow application. The
approach and obtained state-space models of very high order experimental implementation of the resulting controller requires
(150–200 states). In these experiments, low coherence was real-time updates of the state variables (coefficients of the POD
observed between input and measured signals in the system modes) based on sensor measurements, and various methods for
identification experiment, so long time records (410 s) were such state estimation are discussed in Section 4.3.1.
required. Controllers designed from these models yielded reason-
ably good suppression, and the models predicted general trends
4.2.3. Rossiter-type models
observed in the experiments.
In contrast to POD models, in which the full Navier–Stokes
Cattafesta et al. [29] and Kegerise et al. [84] subsequently used
equations are simplified, another approach is to model different
an on-line adaptive system identification algorithm together with
components of the Rossiter mechanism, and connect these
an adaptive feedback controller. In this case, an independent
together, building more complex models from simple models
random signal is added to the computed control signal so that the
of the individual components. Cain et al. [117] used a nonlinear
system model can be updated while the adaptive controller is
model for shear-layer amplification, coupled with models for
active. Such closed-loop system identification avoids the problem
acoustic scattering and receptivity, to predict amplitudes of cavity
mentioned above of identifying an unstable system. Adaptive
oscillations. Their procedure was iterative, and provided an
identification algorithms, including extensions to nonlinear
estimate for steady-state amplitude, but was not time-accurate.
system models, are discussed further in Pillarisetti and Cattafesta
Nonetheless, it represented a significant improvement over the
[111]. Nonlinear models provide greater model accuracy at the
basic Rossiter model [2] and the more advanced Tam and Block
expense of increased computational complexity. The resulting
[118] model.
nonlinear model can either be linearized and used with standard
Rowley and Williams [68] and Rowley et al. [82] used purely
linear control schemes, or they can be used with nonlinear control
linear models for the various components as shown in Fig. 10,
schemes.
including shear-layer amplification, and obtained time-accurate
models in the form of transfer functions. The model suggested an
alternative mechanism for cavity oscillations: while the conven-
4.2.2. POD/Galerkin models
tional view is that the oscillations are self-sustained; an
The system identification procedures described above treat the
alternative view is that the cavity could act as a lightly damped
cavity as a black box, and models are obtained by observing the
oscillator that amplifies noise at resonant frequencies. In this case,
response to forcing. An alternative approach is to obtain models
the oscillations are not self-sustaining. The linear models obtained
based solely on the physics of the flow system. The advantage
in their paper explained some peak-splitting effects observed in
of this approach is that it can provide insight into the mechanisms
experiments, and implied some fundamental performance limits
of cavity oscillations that might not be apparent otherwise. In
[13], as discussed in Section 4.4.
addition, one may obtain scaling laws that determine how the
The models described above provide physical insight, but
models vary as a parameter such as Mach number or cavity length
without empirical tuning, they are not sufficiently accurate to
is varied.
design a control system. Recently, Kerschen and Tumin [119],
The governing equations for most fluids are the Navier–Stokes
Alvarez et al. [120], and Alvarez and Kerschen [121] described a
equations, so in a sense a very accurate model of cavity flows is
theoretical model of cavity resonance that shows promise for
already known. However, standard tools for control analysis and
physics-based cavity control design. It combines a propagation
design do not apply to nonlinear partial differential equations. So
model based on a finite-thickness shear layer with scattering
in order to use these techniques, it is desirable to approximate
models for the leading and trailing-edge regions of the cavity
the Navier–Stokes equations by a finite-dimensional system. An
using an exact Wiener–Hopf technique. The model predicts the
increasingly popular method for obtaining such low-dimensional
cavity resonance frequencies without any empirical constants and
models is Proper Orthogonal Decomposition (POD) and Galerkin
also provides the temporal growth rate of each mode. It has also
projection. In this method, data from simulations or experiments
been used to demonstrate the effect of tunnel modes [13].
are used to determine a finite-dimensional subspace that contains
the ‘‘most important’’ features of the flow (based on energy). The
Navier–Stokes equations are then projected onto this subspace to 4.3. Control algorithms
obtain a low-dimensional model. For a comprehensive review of
this method, see Holmes et al. [112]. This section discusses dynamic-control algorithms, in which
POD/Galerkin models for 2-D flow past a cavity were demon- the feedback occurs at the time scales of the unsteady flow. The
strated by Rowley et al. [113,114], and POD modes were also simplest control strategies do not require a model, and the
determined by Ukeiley et al. [115]. Similar models were obtained parameters of the controller are tuned manually. Williams et al.
from experimental data and used for control design by Samimy [28] used such a strategy, in which a pressure signal was bandpass
et al. [89] and Caraballo et al. [90]. The standard method for filtered about the frequency of a cavity tone, and a phase shifter
incompressible flows needs to be modified for compressible flows, was manually tuned until the oscillations were suppressed.
in which case the thermodynamic variables become important. It Several bandpass filters were used in parallel to achieve suppres-
has been found that vector-valued POD modes work better than sion of multiple modes, and a similar approach was used by
 ARTICLE IN PRESS

498 L.N. Cattafesta III et al. / Progress in Aerospace Sciences 44 (2008) 479–502

Fig. 10. Physics-based model of cavity oscillations that uses a linear model (i.e., transfer function) for each component of the system (from Rowley et al. [82]).

Fig. 11. Control architecture to bandpass filter each cavity tone of interest and feed back a scaled and delayed version of each for maximum suppression (from Kegerise et al.
[84]).

Kegerise et al. [84] to suppress multiple modes, as illustrated in well as pole placement (also discussed below), to suppress the
Fig. 11. oscillations with an order-of-magnitude lesser input power than
While manual tuning of controller parameters can work open-loop techniques. Cabell et al. [83] also used an LQG
reasonably well, often new tones are produced by the controller. regulator, with a frequency-dependent weighting on the control
Furthermore, obtaining good suppression with manual tuning is effort. The models used for these control designs were identified
usually difficult, especially when multiple modes are present and directly from experimental data. POD/Galerkin models have also
the actuator bandwidth is limited. More predictable design been used for control design using LQR, in direct numerical
methodologies are available if one has a model of the system. simulations [129] and experiments [89,90]. However, controllers
For instance, Mongeau et al. [78] and Kook et al. [79] used a based on POD/Galerkin models often suffer from limitations in the
classical loop-shaping technique to design a control law for the model fidelity. In the simulations, for example, careful tuning of
flow past a Helmholtz resonator. Given a plant model P(s), then the LQR weights was necessary to achieve suppression of tones,
with a controller given by C(s), where s ¼ io, one may design the and the performance of the model on full simulations did not
loop gain L(s) ¼ P(s)C(s) such that the closed-loop system has match the behavior of the reduced-order model. In the experi-
some desired properties, such as good disturbance rejection over a ments, the gains needed to be reduced by an adjustable multi-
certain frequency range (characterized by the sensitivity function plicative factor to avoid actuator saturation. These results
described in Section 4.4). One then determines C(s) by inverting highlight the need for developing methods to provide more
the plant: C(s) ¼ L(s)/P(s). If the plant model has zeros or poles in accurate reduced-order models and also to adapt or update the
the right-half plane (RHP), then in order to avoid undesirable RHP model in an automated fashion as the system moves along a
pole-zero cancellation, one must place certain restrictions on L(s), trajectory from its baseline to controlled state.
which can be cumbersome (see Doyle et al. [122] for more For both pole placement and LQG, one begins with a model or
details). realization of the system in state-space form:
Modern control tools provide systematic ways of designing
controllers, given an adequate model [123]. Two types of optimal xkþ1 ¼ Axk þ Buk ,
control design, discussed below, are the Linear Quadratic yk ¼ Cxk þ Duk , (2)
Regulator (LQR), used when the full state vector is available, and
Linear Quadratic Gaussian (LQG), when only noisy sensor where uk is the input (actuator voltage) at time tk ¼ k Dt, and yk is
measurements are available. Cattafesta et al. [35] used LQG, as the corresponding output (sensor measurement). Here xk is the
 ARTICLE IN PRESS

L.N. Cattafesta III et al. / Progress in Aerospace Sciences 44 (2008) 479–502 499

state vector, and A–D are matrices of appropriate dimension. If In LSE, the state estimate is represented by a static linear
one chooses a feedback law function of the sensors

uk ¼ Kxk , (3) x^ k ¼ Myk , (7)

where K is a matrix of gains, then the closed-loop system becomes where the matrix M is determined by correlating known states x
with their corresponding sensor measurements y. This technique
xkþ1 ¼ ðA þ BKÞxk , does not require a model but requires a larger number of sensors,
yk ¼ ðC þ DKÞxk . (4) and is more sensitive to noise than dynamic observers such as (6)
[129]. Extensions to quadratic stochastic estimation provide a
A basic fact from control theory is that if the realization is significant improvement over (7) but still require a large number
controllable, the matrix K may be chosen to place the eigenvalues of sensors [128,130]. For these reasons, we strongly endorse the
of A+BK at any desired locations. Thus, even if the original system use of dynamic observers such as (6) vs. static estimators such as
is unstable (i.e., the eigenvalues of A lie outside the unit circle), K (7), even if only a crude model of the dynamics is available. For
can be chosen such that the closed-loop system is stable, and in cavity flows, an observer of the form (6) was used by Rowley and
fact such that the response decays arbitrarily quickly. This design Juttijudata [131] to reproduce the first few POD modes in a direct
approach is called pole placement; one specifies the desired numerical simulation, using a single noisy pressure sensor.
locations of the closed-loop poles, and solves for the gains K that Adaptive disturbance rejection control techniques have also
achieve them. The tradeoff in control design is to choose K to been used, in which parameters of the controller are updated
achieve a balance between fast response (good) and large values in real-time in order to drive the sensor measurements to zero.
of the gains (bad). Large gains require large actuator power and Such controllers have been used by Kestens and Nicoud [109],
amplify sensor noise. Cattafesta et al. [29], and Kegerize et al. [86] with good success.
An optimal way of achieving this balance between fast One promising area for future research lies in the combination of
response and large gains is to use a Linear Quadratic Regulator adaptive physics-based system identification combined with
(LQR) [124], which chooses K to minimize a cost function adaptive control. The idea here is to replace the ‘‘black-box’’
X
1 model with a physics-based model but update the model (and the
J¼ ðxTk Qxk þ uTk Ruk Þ, (5) controller) with experimental data.
k¼0

where Q and R are symmetric, positive-definite matrices. The first 4.4. Fundamental limits on achievable performance
term is related to the performance, while the second term
provides a measure of the cost of the control input. These Though closed-loop controllers have provided reasonably good
matrices are the parameters in the control design, and by choosing suppression, they often produce some adverse effects, such as an
different matrices we choose how to balance good performance increase in noise at other frequencies. For instance, if the gain of a
against the cost of control. One determines the value of the gain controller is increased too much, then a peak-splitting phenom-
matrix K that minimizes this cost function by solving a quadratic enon may be observed as shown in Fig. 12 (Rowley et al. [81] and
matrix equation (an algebraic Riccati equation), and there are Cattafesta et al. [29]), in which the main resonant peak splits into
commercial software tools available to do this. two peaks on either side of the original peak. Because of
fundamental limitations of feedback control, such adverse effects
are unavoidable to some extent, though they may be minimized
4.3.1. State estimation: observers and static estimators by the controller design as described below.
In order to use the state feedback techniques described above, In Fig. 4 with unity gain feedback H(s) ¼ 1, the transfer
one needs to know the state xk, which usually is not directly function from an external disturbance to the output is given by
available. For instance, in a POD model, the state xk consists of the
amplitude of each POD mode, which usually cannot be measured PðsÞ
¼ SðsÞPðsÞ, (8)
directly. In order to obtain the state, one typically designs an 1 þ PðsÞCðsÞ
observer to estimate the state, for instance by solving the equation
160
x^ kþ1 ¼ Ax^ k þ Buk þ LðC x^ k þ Duk  yk Þ, (6)
Control Off
where x^ k is the state estimate at time tk and L is a matrix of Control On
155
observer gains. One may choose these observer gains such that
Sound Pressure Level (dB)

the estimate is guaranteed to converge to the actual state (as long

as the realization is observable). Again, there is a tradeoff between 150
fast convergence and large amplification of sensor noise, and an
optimal way of balancing these tradeoffs uses a procedure that
precisely parallels LQR. Such an optimal observer is called a 145
Kalman filter. When a Kalman filter is combined with state
feedback using LQR, the resulting controller is called an LQG
140
regulator. For more information about LQR and LQG regulators,
see standard controls texts [124–126].
An alternative approach is to use a static estimator, such as 135
Linear Stochastic Estimation (LSE), originally proposed by Adrian
[127]. In this approach, the state estimate depends only on the
sensor measurements at the present time, unlike the above 130
0 500 1000 1500
dynamic approach, which has a memory of the time history of the
Frequency (Hz)
sensor measurements. An extension of LSE to include quadratic
terms for improved accuracy has been successfully employed by Fig. 12. The reduction of the first cavity at Mach 0.74 illustrates the peak-splitting
Murray and Ukeiley [128] and Samimy et al. [89] in cavity flows. phenomenon (from Cattafesta et al. [29]).
 ARTICLE IN PRESS

500 L.N. Cattafesta III et al. / Progress in Aerospace Sciences 44 (2008) 479–502

where S ¼ 1/(1+PC) is the sensitivity function. Since the transfer practical interest, the search for a solution to this problem
function from disturbances to output for the open-loop (i.e., no- combined with budgetary, time, and scientific constraints often
control) system is simply P(s), the sensitivity function is the ratio limits flow-physics experiments. Where possible, flow-field
of the closed-loop to open-loop transfer functions and, hence, measurements (both mean and fluctuating components) should
quantifies how feedback alters the effect of disturbances. One of be conducted for the baseline and controlled cases. In addition,
the goals of feedback design for cavity flows is to reduce the the interaction of the actuator with the flow should be
amplification disturbances, so we would like to make |S(io)|o1 characterized. Beyond improving our understanding of cavity
for all frequencies o 2 <. Unfortunately, this is not possible in a oscillations, these data will provide the basis for future compar-
linear control system, due to Bode’s integral formula, or area rule, ison and scaling results. It will also permit the evaluation of
which states that if the relative degree of the loop gain is at least 2 critical issues, for example, isolating high-frequency excitation vs.
(i.e., the degree of the denominator of P(s)C(s) is at least 2 greater mean-flow-modification effects.
than the degree of the numerator, which is virtually always true in Furthermore, several active open-loop and passive schemes
practice), then have demonstrated the ability to reduce broadband levels, while
Z 1 X closed-loop control methods explored to data have not modified
log jSðioÞj do ¼ pðlog eÞ Reðpi Þ, (9) the mean flow significantly. Evidence suggests that broadband
0 i suppression implies a modification of the mean flow.
where pi denotes the RHP poles of P(s)C(s). For instance, if the While the closed-loop results have been promising, only
plant is stable (i.e., the right-hand side of the equation is zero), passive and active open-loop methods have been successful at
then in a log–linear plot of |S(io)|, the area of attenuation supersonic speeds. It is conceivable that a hybrid scheme
(|S(io)|o1) must be balanced by an equal area of amplification consisting of some combination may be effective. For closed-loop
(|S(io)|41), in which the closed-loop system amplifies distur- control to have an even greater impact, several things should
bances even more than the open-loop one. If the plant is unstable, occur. First, adaptive physics-based dynamic models are required
the area of amplification must be larger than the area of to enable design of suitable controllers. Second, better ‘‘Type A’’
attenuation. Of course, all of this attenuation and amplification (discussed in Section 3) actuators are required that have higher
must occur within the bandwidth of the controller, since output, larger bandwidth, and faster time response, such that
|S(io)| ¼ 1 when the control goes to zero. Hence, narrow band- multi-mode tonal (and perhaps broadband) closed-loop control at
width controllers or actuators will have severe limitations. If good supersonic speeds is possible. Third, it is our opinion that
performance is desired at one frequency, then a penalty of bad researchers in the field of flow control must accept, if not
performance at other frequencies within the passband must be embrace, the role of multiple disciplines (e.g., fluid dynamics,
incurred. Wider bandwidth actuators will not have such severe control theory, and transducers). Only in this manner can active
limitations, since the area of amplification may be spread out over flow control achieve its full potential.
a large frequency range. This emphasizes yet again the need to
develop high-bandwidth actuators.

Acknowledgements
5. Summary and outlook
The research program at Florida A&M University and Florida
This paper has provided a review of active control of flow- State University (FAMU-FSU) is supported by DARPA (Technical
induced cavity oscillations, where emphasis has been placed on monitor: Dr. Steven Walker) and Boeing (Technical monitor:
recent experimental investigations of open- and closed-loop Dr. William Bower). The research at UF, IIT and Princeton has been
suppression techniques. It is worth noting that, with a few supported by AFOSR (Technical monitors John Schmisseur and
exceptions, most of the experimental results available in the Sharon Heise). The authors are grateful to Mr. Ning Zhuang for his
literature and discussed herein have been obtained from studies help in reviewing some of the literature for this paper and would
conducted in relatively small facilities. There have been some also like to acknowledge their collaborators on cavity flow work,
questions regarding the value of such studies at ‘‘impractically Profs. A. Annaswamy (MIT) and C. Shih (FAMU-FSU), Larry Ukeiley
small’’ scales as to whether they capture the key flow physics (UF), and Tim Colonius (Caltech).
encountered in the ‘‘real/full-scale’’ world. However, comparison
of results obtained in small-scale experiments with those at larger References
scales has clearly shown that at least as far as cavity flows are
concerned, these experiments do capture the relevant flow [1] Zhuang N, Alvi F, Shih C. Another look at supersonic cavity flows and their
physics and can be accurately scaled to larger facilities and flight control. AIAA 2005-2803, May 2005.
[2] Rossiter JE. Wind-tunnel experiments on the flow over rectangular cavities
tests provided care is taken to avoid tunnel modes [13]. More at subsonic and transonic speeds. Aeronautical Research Council Reports and
importantly, small-scale studies can be successfully used to Memoranda no 3438, October 1964.
design and scale actuators for large-scale tests—see, for example, [3] Powell A. On edge tones and associated phenomena. Acustica 1953;3:
233–43.
Zhuang et al. [50] and Ukeiley et al. [49]. Since the costs [4] Krishnamurty K. Acoustic radiation from two dimensional rectangular
associated with smaller scale studies are significantly lower, and cutouts in aerodynamic surfaces. NACA Technical Note 3487, August 1955.
the results obtained are more detailed, it is imperative that [5] Roshko A. Some measurements of flow in a rectangular cutout. NACA
Technical Note 3488, August 1955.
experiments at these small, research laboratory scales strive to [6] Rockwell D, Naudascher E. Review: self-sustaining oscillations of flow past
achieve a better understanding of the fundamental flow physics Cavities. Trans ASME J Fluids Eng 1978;100:152–65.
governing cavity flows and their response to various actuation [7] Rockwell D, Naudascher E. Self-sustained oscillations of impinging shear
layers. Annu Rev Fluid Mech 1979;11:67–94.
methods. Once better understood, the more promising concepts
[8] Rockwell D. Oscillations of impinging shear layers. AIAA J 1983;21(5):
can be intelligently scaled for further validation at increasingly 645–64.
larger scale. [9] Blake WK, Powell A. The development of contemporary views of flow tone
Based on the results discussed in this article, we conclude with generation. In: Recent advances in aeroacoustics. New York: Springer; 1986.
p. 247–345.
the following observations/recommendations. First, while the [10] Komerath NM, Ahuja KK, Chambers FW. Prediction and measurement of
suppression of cavity oscillations is an important problem of flows over cavities—a survey. AIAA 87-0166, January 1987.
 ARTICLE IN PRESS

L.N. Cattafesta III et al. / Progress in Aerospace Sciences 44 (2008) 479–502 501

[11] Chokani N. Flow induced oscillations in cavities—a critical survey. DGLR/ [50] Zhuang N, Alvi FS, Alkislar MB, Shih C. Supersonic cavity flows and their
AIAA 92-01-159, May 1992. control. AIAA J 2006;44(9):2118–28 See also AIAA 2003-3101.
[12] Colonius T. An overview of simulation, modeling, and active control of flow/ [51] Bower WW, Kibens V, Cary AW, Alvi F, Raman G, Annaswamy A, et al. High-
acoustic resonance in open cavities. AIAA 2001-0076, January 2001. frequency excitation active flow control for high-speed weapon release
[13] Rowley CW, Williams DR. Dynamics and control of high-Reynolds number (HIFEX). AIAA 2004-2513, June 2004.
flow over cavities. Annu Rev Fluid Mech 2006;38:251–76. [52] Shalaev VI, Fedorov AV, Malmuth ND. Dynamics of slender bodies separating
[14] Williams DR, Rowley CW. Recent progress in closed-loop control of cavity from rectangular cavities. AIAA J 2002;40(3):517–25.
tones. AIAA 2006-0712, January 2006. [53] Sahoo D, Annaswamy A, Alvi FS. Microjets-based active control of store
[15] Bruggeman JC, Hirschberg A, van Dongen MEH, Wijnands APJ. Self-sustained trajectory in a supersonic cavity using a low-order model. AIAA J 2007;
aero-acoustic pulsations in gas transport systems: experimental study of the 45(3):516–31 See also AIAA 2005-3097.
influence of closed side branches. J Sound Vibr 1991;150:371–93. [54] Pinney MA, Leugers JE. Experimental ivestigation of the ipact of internal/
[16] Kook H, Mongeau L, Brown DV, Zorea S. Analysis of the interior pressure external weapons carriage on a generic aircraft configuration. L-TR-96-3110,
oscillations induced by flow over vehicle openings. Noise Control Eng J 1997; Final Report, Wright Laboratory, 1996.
45:223–34. [55] Schmit RF, Schwartz DR, Kibens V, Raman G, Ross JA. High and low frequency
[17] Srinivasan GR. Acoustics and unsteady flow of telescope cavity in an actuation comparison for a weapons bay cavity. AIAA 2005-0795, January
airplane. AIAA J Aircr 2000;37(2):274–81. 2005.
[18] Jumper EJ, Fitzgerald EJ. Recent advances in aero-optics. Prog Aerosp Sci [56] Ukeiley LS, Ponton MK, Seiner JM, Jansen B. Suppression of pressure loads in
2001;37:299–339. resonating cavities through blowing. AIAA 2003-0181, January 2003.
[19] Stanek M, Raman G, Kibens V, Ross J, Odedra J, Peto J. Control of cavity [57] Mongeau L, Franchek MA, Kook H. Control of interior pressure fluctuations
resonance through very high frequency forcing. AIAA 2000-1905, June 2000. due to flow over vehicle openings. In: Proceedings of the 1999 noise and
[20] Heller HH, Bliss DB. The physical mechanisms of flow-induced pressure vibration conference, May 1999, vol. 2. p. 1257–66.
fluctuations in cavities and concepts for their suppression. AIAA 75-491, [58] Shaw L, McGrath S, Weapons bay acoustics—passive or active control. AIAA
March 1975. 96-1617, April 1996.
[21] Shaw LL. Suppression of aerodynamically induced cavity oscillations. AFFDL- [59] Cattafesta III LN, Garg S, Shukla D. The development of piezoelectric
TR-79-3119, November 1979. actuators for active flow control. AIAA J 2001;39(8):1562–8.
[22] Chokani N, Kim I. Suppression of pressure oscillations in an open cavity by [60] Cattafesta L, Mathew J, Kurdila A. Modeling and design of piezoelectric
passive pneumatic control. AIAA 91-1729, June 1991. actuators for fluid flow control. SAE 2000 Trans—J Aerosp (Sect 1) 2001;109:
[23] Raman G, Cain AB. Innovative actuators for active flow and noise control. 1088–95.
Proc Instn Mech Eng G J Aerosp Eng 2002;216:303–323. [61] Mathew J, Song Q, Sankar B, Sheplak M, Cattafesta L. Optimized design of
[24] McGrath SF, Shaw Jr LL. Active control of shallow cavity acoustic resonance. piezoelectric flap actuators for active flow control. AIAA J 2006;44(12):
AIAA 96-1949, June 1996. 2919–28.
[25] Ukeiley LS, Ponton MK, Seiner JM, Jansen B. Suppression of pressure loads in [62] Yokokawa Y, Fukunishi Y, Kikuchi S. Suppression of aero-acoustic noise by
cavity flows. AIAA J 2004;42(1):70–9 See also AIAA 2002-0661. separation control using piezo-actuators. AIAA 2000-1931, June 2000.
[26] DiStefano III JJ, Stubberud AR, Williams IJ. Feedback and control systems. In: [63] Grove J, Birkbeck RM, Kreher JM. Acoustic and separation characteristics
Schaum’s outlines. 2nd ed. New York: McGraw-Hill; 1990. with bay leading edge blowing. AIAA 2000-1904, June 2000.
[27] Shaw L, Northcraft S. Closed loop active control for cavity resonance. AIAA [64] Grove J, Leugers J, Akroyd G. USAF/RAAF F-111 flight test with active
99-1902, May 1999. See also Rothfuss DA, Northcraft SA. Active control of separation control. AIAA 2003-0009, January 2003.
weapons bay acoustics. AFRL-VA-WP-TR-1998-3039, June 1998. [65] Smith BR, Jordan JK, Bender EE, Rizk SN, Shaw LL. Computational simulation
[28] Williams D, Fabris D, Iwanski K, Morrow J. Closed loop control in cavities of active control of cavity acoustics. AIAA 2000-1927, June 2000.
with unsteady bleed forcing. AIAA 2000-0470, January 2000. [66] Williams DR, Cornelius D, Rowley CW. Supersonic cavity response to open-
[29] Cattafesta III LN, Shukla D, Garg S, Ross JA. Development of an adaptive loop forcing. In: Proceedings of the first Berlin conference on active flow
weapons-bay suppression system. AIAA 99-1901, May 1999. control, September 2006.
[30] Sarohia V, Massier PF. Control of cavity noise. J Aircr 1977;14(9):833–7. [67] Williams DR, Fabris D, Morrow J. Experiments on controlling multiple
[31] Vakili AD, Gauthier C. Control of cavity flow by upstream mass-injection. acoustic modes in cavities. AIAA 2000-1903, June 2000.
J Aircr 1994;31(1):169–74. [68] Rowley CW, Williams DR. Control of forced and self-sustained oscillations in
[32] Sarno RL, Franke ME. Suppression of flow-induced pressure oscillations in the flow past a cavity. AIAA 2003-0008, January 2003.
cavities. J Aircr 1994;31(1):90–6. [69] Raman G, Kibens V, Cain A, Lepicovsky J. Advanced actuator concepts for
[33] Mendoza JM, Ahuja KK. Cavity noise control through upstream mass active aeroacoustic control. AIAA 2000-1930, June 2000.
injection from a Coanda surface. AIAA 96-1767, May 1996. [70] Stanek MJ, Raman G, Kibens V, Ross JA, Odedra J, Peto JW. Suppression of
[34] Hsu JS, Ahuja KK. Cavity noise control using Helmholtz resonators. AIAA 96- cavity resonance using high frequency forcing—the characteristic signature
1675, May 1996. of effective devices. AIAA 2001-2128, May 2001.
[35] Cattafesta III LN, Garg S, Choudhari M, Li F. Active control of flow-induced [71] Chan S, Zhang X, Gabriel S. Attenuation of low-speed flow-induced cavity
cavity resonance. AIAA 97-1804, June 1997. tones using plasma actuators. AIAA J 2007;45(7):1525–38.
[36] Shaw L. Active control for cavity acoustics. AIAA 98-2347, June 1998. [72] Schlichting H. Boundary layer theory. 7th ed. New York: McGraw-Hill; 1979.
[37] Fabris D, Williams DR. Experimental measurements of cavity and shear layer p. 428–32.
response to unsteady bleed forcing. AIAA 99-0606, January 1999. [73] Little J, Debiasi M, Caraballo E, Samimy M. Effects of open-loop and closed-
[38] Lamp AM, Chokani N. Computation of cavity flows with suppression using loop control on subsonic cavity flows. Phys Fluids 2007;19.
jet blowing. J Aircr 1997;34(4):545–51. [74] Stanek MJ, Sinha R, Seiner J, Pierce B, Jones M. High frequency flow
[39] Raman G, Raghu S, Bencic TJ. Cavity resonance suppression using miniature control—suppression of aero-optics in tactical directed energy beam
fluidic oscillators. AIAA 99-1900, May 1999. propagation and the birth of a new model (part I), AIAA 2002-2272, May
[40] Raman G, Raghu S. Cavity resonance suppression using miniature fluidic 2002.
oscillators. AIAA J 2004;42(12):2608–12. [75] Rizzetta DP, Visbal MR. Large-eddy simulation of supersonic cavity flowfield
[41] Stanek MJ, Raman G, Ross JA, Odedra J, Peto J, Alvi F, et al. High frequency including flow control. AIAA 2002-2853, June 2002.
acoustic suppression—the role of mass flow, the notion of superposition, and [76] Kim K, Debiasi M, Schultz R, Serrani A, Samimy M. Dynamic compensator for
the role of inviscid instability—a new model (part II). AIAA 2002-2404, June a synthetic-jet-like compression driver actuator in closed-loop cavity flow
2002. control. AIAA 2007-880, January 2007.
[42] Stanek MJ, Raman G, Ross JA, Odedra J, Peto J, Alvi FS, et al. High frequency [77] Ziada S. Flow excited resonance of a confined shallow cavity in low
acoustic suppression—the mystery of the rod-in-crossflow revealed. AIAA mach number flow and its control. In: Proceedings of IMECE2002 ASME
2003-0007, January 2003. international mechanical engineering congress and exposition, November
[43] Wiltse JM, Glezer A. Direct excitation of small-scale motions in free shear 2002.
flows. Phys Fluids 1998;10(8):2026–36. [78] Mongeau L, Kook H, Franchek MA. Active control of flow-induced cavity
[44] Stanek MJ. A numerical study of the effect of frequency of pulsed flow resonance. AIAA 98-2349, June 1998.
control applied to a rectangular cavity in supersonic crossflow. PhD thesis, [79] Kook H, Mongeau L, Franchek MA. Active control of pressure fluctuations
Department of Aerospace Engineering and Engineering Mechanics, Uni- due to flow over Helmholtz resonators. J Sound Vibr 2002;255(1):61–76.
versity of Cincinnati, 2005. [80] Williams DR, Rowley C, Colonius T, Murray R, MacMartin D, Fabris D,
[45] Arunajatesan S, Shipman JD, Sinha N. Mechanisms in high frequency control et al. Model-based control of cavity oscillations—part 1: experiments. AIAA
of cavity flows. AIAA 2003-0005, January 2003. 2002-0971, January 2002.
[46] Blackstock DT. Fundamentals of physical acoustics. New York: Wiley; 2000. [81] Rowley CW, Williams DR, Colonius T, Murray RM, MacMartin D, Fabris D.
p. 218. Model-based control of cavity oscillations—part II: system identification
[47] Le KC. Vibrations of shells and rods. New York: Springer; 1999. p. 155–61. and analysis. AIAA 2002-0972, January 2002.
[48] Bueno PC, Ünalmis ÖH, Clemens NT, Dolling DS. The effects of [82] Rowley CW, Williams DR, Colonius T, Murray RM, MacMynowski DG.
upstream mass injection on a Mach 2 cavity flow. AIAA 2002-0663, January Linear models for control of cavity flow oscillations. J Fluid Mech 2006;547:
2002. 317–30.
[49] Ukeiley LS, Sheehan M, Coiffett F, Alvi FS, Srinivasan A, Jansen B. Control [83] Cabell RH, Kegerise MA, Cox DE, Gibbs GP. Experimental feedback control
of pressure loads in complex cavity configurations. AIAA 2007-1238, January of flow-induced cavity tones. AIAA J 2006;44(8):1807–16 See also AIAA
2007. 2002-2497.
 ARTICLE IN PRESS

502 L.N. Cattafesta III et al. / Progress in Aerospace Sciences 44 (2008) 479–502

[84] Kegerise MA, Cattafesta LN, Ha C-S. Adaptive identification and control of [107] Micheau P, Chatellier L, Laumonier J, Gervais Y. Active control of a self-
flow induced cavity oscillations. AIAA 2002-3158, June 2002. sustained pressure fluctuation due to flow over a cavity. AIAA 2004-2851,
[85] Kegerise M, Cabell R, Cattafesta L. Real-time feedback control of flow- May 2004.
induced cavity tones. Part 1: fixed-gain control. J Sound Vibr 2007;307: [108] Haykin S. Adaptive filter theory. 4th ed. Upper Saddle River, NJ: Prentice-
906–23. Hall, Inc.; 2002.
[86] Kegerise M, Cabell R, Cattafesta L. Real-time feedback control of flow- [109] Kestens T, Nicoud F. Active control of unsteady flow over a rectangular
induced cavity tones. Part 2: adaptive control. J Sound Vibr 2007;307: cavity. AIAA 98-2348, June 1998.
924–40. [110] Juang J-N. Applied system identification. Upper Saddle River, NJ: Prentice-
[87] Debiasi M, Samimy M. Logic-based active control of subsonic cavity flow Hall, Inc.; 1994.
resonance. AIAA J 2004;42(9):1901–9 See also AIAA 2003-4003. [111] Pillarisetti A, Cattafesta III, LN. Adaptive Identification of Fluid Dynamic
[88] Yan P, Debiasi M, Yuan X, Little J, Özbay H, Samimy M. Experimental study of Systems, AIAA 2001-2978, June 2001.
linear closed-loop control of subsonic cavity flow. AIAA J 2006;44(5): [112] Holmes P, Lumley JL, Berkooz G. Turbulence, coherent structures, dynamic
929–38. systems and symmetry. Cambridge: Cambridge University Press; 1996.
[89] Samimy M, Debiasi M, Caraballo E, Serrani A, Yuan X, Little J, et al. Feedback [113] Rowley CW, Colonius T, Murray R. POD based models of self-sustained
control of subsonic cavity flows using reduced-order models. J Fluid Mech oscillations in the flow past an open cavity. AIAA 2000-1969, June 2000.
2007;579:315–46. [114] Rowley CW, Colonius T, Murray R. Dynamic models for control of cavity
[90] Caraballo E, Little J, Debiasi M, Samimy M. Development and implementa- oscillations. AIAA 2001-2126, May 2001.
tion of an experimental-based reduced-order model for Feedback control of [115] Ukeiley LS, Kannepalli C, Arunajatesan S, Sinha N. Low-dimensional
subsonic cavity flows. J Fluids Eng 2007;129(7):813–24. description of variable density flows. AIAA 2001-0515, January 2001.
[91] Gallas Q, Holman R, Nishida T, Carroll B, Sheplak M, Cattafesta L. Lumped [116] Rowley CW, Colonius T, Murray RM. Model reduction for compressible flows
element modeling of piezoelectric-driven synthetic jet actuators. AIAA J using POD and Galerkin projection. Physica D 2004;189(1–2):115–29.
2003;41(2):240–7. [117] Cain AB, Bower WW, McCotter F, Romer WW. Modeling and prediction of
[92] Shaw L. High speed application of active flow control for cavity acoustics. weapons bay acoustic amplitude and frequency. VEDA Inc.; February 1996.
AIAA 2000-1926, June 2000. [118] Tam CKW, Block PJW. On the tones and pressure oscillations induced by flow
[93] Shaw L. Active flow control of weapons bay acoustics and vibration over rectangular cavities. J Fluid Mech 1978;89(2):373–99.
environments—current status. In: Proceedings of the 72nd shock and [119] Kerschen EJ, Tumin A. A theoretical model of cavity acoustic resonances
vibration symposium, November 2001. based on edge scattering processes. AIAA 2003-0175, January 2003.
[94] Lou H, Alvi FS, Shih C, Choi J, Annaswamy A. Active control of supersonic [120] Alvarez JO, Kerschen EJ, Tumin A. A theoretical model for cavity acoustic
impinging jets: Flowfield properties and closed-loop strategies. AIAA 2002- resonances in subsonic flow. AIAA 2004-2845, May 2004.
2728, June 2002. [121] Alvarez JO, Kerschen EJ. Influence of wind tunnel walls on cavity acoustic
[95] Williams D, Morrow J. Adaptive control of multiple acoustic modes in resonances. AIAA 2005-2804, May 2005.
cavities. AIAA 2001-2769, June 2001. [122] Doyle JC, Francis BA, Tannenbaum AR. Feedback control theory. New York:
[96] Ünalmis ÖH, Clemens NT, Dolling DS. Experimental study of shear layer/ Macmillan; 1992.
acoustics coupling in Mach 5 cavity flow. AIAA J 2001;39(2):242–52. [123] Kailath T. Linear systems. Englewood Cliffs, NJ: Prentice-Hall; 1980.
[97] Garg S, Cattafesta III LN. Quantitative Schlieren measurements of coherent [124] Kwakernaak H, Sivan R. Linear optimal control systems. New York: Wiley-
structures in a cavity shear layer. Exp Fluids 2001;30(2):123–34. Interscience; 1972.
[98] Kegerise MA. An experimental investigation of flow-induced cavity oscilla- [125] Bélanger PR. Control engineering: a modern approach. Oxford: Oxford
tions. PhD thesis, Syracuse University, Syracuse, NY, August 1999. University Press; September 2005.
[99] Smits AJ, Hayakawa K, Muck KC. Constant temperature hot-wire anem- [126] Skogestad S, Postlethwaite I. Multivariable feedback control analysis and
ometer practice in supersonic flows. Part 1: the normal wire. Exp Fluids design. 2nd ed. New York: Wiley-Interscience; November 2005.
1983;1(2):83–92. [127] Adrian RJ. On the role of conditional averages in turbulence theory. In:
[100] Vakili AD, Wolfe R, Nagle T, Lambert E. Active control of cavity aeroacoustics Proceedings of the fourth biennial symposium on turbulence in liquids.
in high speed flows. AIAA 95-0678, January 1995. University of Missouri, Rolla, MO, September 1975.
[101] Kegerise MA, Spina EF, Cattafesta III LN. An experimental investigation of [128] Murray N, Ukeiley L. Estimation of the flowfield from surface pressure
flow-induced cavity oscillations. AIAA Paper 99-3705, June 1999. measurements in an open cavity. AIAA J 2003;41(5):969–72.
[102] Kegerise MA, Spina EF, Garg S, Cattafesta III LN. Mode-switching and [129] Rowley CW, Juttijudata V, Williams DR. Cavity flow control simulations and
nonlinear effects in compressible flow over a cavity. Phys Fluids experiments. AIAA 2005-0292, January 2005.
2004;16(3):678–87 See also AIAA 98-2912. [130] Ukeiley L, Murray N, Song Q, Cattafesta L. Surface pressure based estimation
[103] Forestier N, Jacquin L, Geffroy P. The mixing layer over a deep cavity at high- for control. In: Morrison JF, Birch DM, Lavoie P, editors. IUTAM symposium
subsonic speed. J Fluid Mech 2003;475:101–45. on flow control and MEMS, September 2006, IUTAM Bookseries, vol. 7, 2008.
[104] Murray RC, Elliott GS. Characteristics of the compressible shear layer over a [131] Rowley CW, Juttijudata V. Model-based control and estimation of cavity flow
cavity. AIAA J 2001;39(5):846–56. oscillations. In: Proceedings of the 44th IEEE conference on decision and
[105] Meganathan AJ, Vakili AD. An experimental study of open cavity flows at low control, December 2005.
subsonic speeds. AIAA 2002-0280, January 2002. [132] Efe MÖ, Debiasi M, Yan P, Özbay H, Samimy M. Control of subsonic cavity
[106] Gharib M. Response of the cavity shear layer oscillations to external forcing. flows by neural networks—analytical models and experimental validation.
AIAA J 1987;25(1):43–7. AIAA 2005-0294, January 2005.