You are on page 1of 17

Copyright 1996, American Institute of Aeronautics and Astronautics, Inc.

AIAA Meeting Papers on Disc, January 1996


A9617981, F49620-92-J-0287, F49620-92-J-0038, N00014-92-J-1547, NSF OCE-90-17882, AIAA Paper 96-0001

Control of turbulence
J. Lumley
Cornell Univ., Ithaca, NY

AIAA, Aerospace Sciences Meeting and Exhibit, 34th, Reno, NV, Jan. 15-18, 1996
Applications of active control of turbulent fluid flow are discussed, with their implications for the economy and the
environment. We outline a conceptual basis for control, sketching sensors, algorithm, and actuators. We describe the
physical basis for control of the boundary layer: coherent structures and bursts; the connection between burst
frequency and friction velocity; the change of burst frequency and drag reduction possible with polymers or active
control; and other effects on burst frequency. The state of the flow must be sensed from the surface, and the
information is necessarily incomplete and aliased. Sophisticated techniques may be necessary to interpret the
signals. A control strategy is necessary, and we need a model of the flow as a predictor/interpretor; surface actuators
are also necessary. We present surprising results of Direct Numerical Simulation of an actuator. Before controlling the
fluid, we have tried to control a model; we present results of attempts to control several different models. (Author)

Page 1

AIAA-96-0001

Control of Turbulence
John Lumley*
Sibley School
of
Mechanical and Aerospace Engineering
Cornell University
Ithaca, NY 14853
January 1996

Abstract
A few applications of active control of turbulent fluid flow are
discussed, with their implications for the economy arid the environment. We outline a conceptual basis for control, sketching sensors,
algorithm and actuators. Control requires understanding of turbulent
flows beyond our present capabilities. We describe the physical basis
for control of the boundary layer: coherent structures and bursts; the
connection between burst frequency and friction velocity; the change
of burst frequency and drag reduction possible with polymers or active control; other effects on burst frequency [e.g.- curvature, acceleration / deceleration (pressure gradients), divergence - convergence
(extra rates of strain)]. The state of the flow must be sensed from
*The 1996 Dryden Lecture in Research of the American Institute of Aeronautics and
Astronautics; AIAA Paper No. 96-0001. Supported in part by Contract No. F4962092-J-0287 jointly funded by the U. S. Air Force Office of Scientific Research (Control
and Aerospace Programs), and the U. S. Office of Naval Research, in part by Grant No.
F49620-92-J-0038, funded by the U. S. Air Force Office of Scientific Research (Aerospace
Program), and in part by the Physical Oceanography Programs of the U. S. National
Science Foundation (Contract No. OCE-901 7882) and the U. S. Oflice of Naval Research.
(Grant No. N000H-92-.1-1547).
1

the surface, and the information is necessarily incomplete and aliased.


Sophisticated techniques may be necessary to interpret the signals. A
control strategy is necessary, an algorithm, and we need a model of
the flow as a predictor / interpreter. Surface actuators are necessary what do they do to the fluid? We present surprising results of Direct
Numerical Simulation of a type of actuator. Before controlling the
fluid, we have tried to control a model; we present results of attempts
to control several different models.

Introduction

For the last few years, control of turbulent flows has been a very popular
subject. NASA, for example (in the person of Dennis Bushnell) stated that
we would have to improve our lift to drag ratio by a factor of two during
the next two decades in order to remain competitive with Europe and the
Pacific rim, and that this would have to be done by various clever drag reduction and high-lift devices, among them active control. In the hypersonic
transport, for example, it was pointed out that most of the drag occurred
in the engine, the only place where the density was appreciable, and that
this was an ideal opportunity for drag reduction by active control, since the
area was relatively small; in addition, there was difficulty getting the fuel
and oxidant mixed and burned in the time available, and this was a good
application for active control also, to increase mixing rates. More recently,
the engines of the high speed civil transport have proved too noisy for commercial airport regulations, arid active control has been proposed as a means
to increase the mixing, causing the jets to spread more rapidly, and the jet
velocity consequently to drop, reducing the noise output.
Recently, [10], [17], [9] some of these attempts at active control have
been written up in the print media. In this case, specifically, the stories
featured tiny micro-mechanical actuators developed by Chih-Minh Ho and
his co-workers. None of the articles mentioned, however, the necessary understanding of the flow, and the need for a control algorithm, to make this
work. The articles suggested that such control devices would be tried on real
aircraft within two years [43]. In this test application, the use of a neural
net was envisioned; by the use of a neural net, you hope that you don't need
to understand the flow, or have a clearly developed control algorithm - like

biofeedback, it is hoped that the net will learn to do the right thing without
understanding it, to bring down the drag, or whatever else is required. However, experience (Jan Aschenbach - private communication) has indicated
that allowing the net to learn in this way is slow and inefficient, producing
unimpressive results; better results are obtained only when there is a clear
model and algorithm available to train the net.
To justify continued funding, a practical demonstration is absolutely eseritial. However, the time scale of the funding agencies and that of the research
projects is not usually similar, and practical demonstrations are usually demanded long before the project is ready. This risks a negative outcome which
might have been avoided, but which could kill the project.
No sooner had the possibility of active flow control appeared in the popular press than the funding agencies announced that they were cutting back
the funding for active control. There are a number of possible explanations
here: the agencies may have felt that the technology was so close to practical
development that no further research was necessary; they may have felt that,
on the basis of preliminary tests, the expected gains would not justify further
investment, or the aircraft industry may have said that they were no longer
interested in drag reduction, because the price of oil was relatively low, and
that they were more interested in reduced cycle time. I believe all of these
reasons are ill founded.
The technology is certainly far from mature; rather, it is in its infancy.
Preliminary tests have indicated gains in the neighborhood of 5%-10%, which
probably would not justify continued investment. However, theory indicates
possible gains of 50%, and the very preliminary nature of the tests and immaturity of the field suggests that continued investment at this time might
pay off. So far as the future is concerned, [1] indicates that managers have
not for a considerable time undertaken any R&D effort that does not promise
payback within five years. Research like active control of turbulence, however, may not payback in less than 20 years or more [1]. In 20 years, the price
of crude oil is almost certain to be higher - if we want to have technology
ready then, we must start now.
In what follows, we will present the very particular point of view of our
own group on flow control; this paper is riot intended to be in any sense a
comprehensive survey. A number of these have been published in the last
few years [8], [11], [24], [25], [26], [41], and the reader is referred to these for
a survey of the field from various points of view.

Control as a Concept

We envision sensors somewhere in the flow, perhaps on the skin, perhaps


somewhere downstream, or off axis, depending on the particular flow. The
function of these sensors is to tell us the state of the flow. These sensors might
be V-hot film units, for example, in the case of the boundary layer. Such units
would respond to the streamwise and cross-stream fluctuating wall shear
stress. The sensors might be arranged on one or more lines across the stream,
at different streamwise locations, and of course would have a certain spacing
in the cross-stream direction. We are interested in knowing the velocity field
above the surface in a region, and the location and distribution of the sensors
restricts the information that we can obtain. Since the sensors are on the
surface, they can perceive only a part of the information from the flow above.
In addition, their location on lines, and their discrete distribution opens
the possibility of aliasing of various sorts, in which one kind of information
appears as another kind.
We need help in interpreting the input of the sensors. Here is where a
low-dimensional model would be useful. It could, for example, be used as a
Kalman filter to make the most of the partial information from the sensors.
We must also address the problem of how, when arid where to interfere
with the flow, to have the greatest effect. This is known as the control
algorithm, something that was omitted from [10]. This is probably the most
important aspect of control, and a matter that is far from being resolved in
any flow. In the boundary layer, for example, if we wish to interfere with
what is called the bursting of the coherent structures, what do we want to
do - prevent them from developing, surely, but how? Blow them away from
the wall? Blow them sideways? Add vorticity of opposite sign to them to
weaken them? Just when arid where do we want to do this? How will we
decide in real time?
Finally, we have the question of actuators. Once we have decided what
we want to do, what flow disturbance we wish to introduce, we must have
a device that will produce this disturbance. This might be a flap, perhaps,
or a bump, on the surface under a boundary layer, or a flap on the lip of
a jet orifice, for example. We need to know the flow disturbance that will
be produced by the actuator. These actuators, in the case of the boundary
layer, function in the very viscous, but riot quite viscously dominated region
near the wall, and such flows are difficult to predict. We will see later that

the flow fields are often not at all what we expect.

Physical Basis for Control of the Boundary


Layer

Many turbulent flows display a combination of organized, or coherent, structures and apparently disorganized, or incoherent, structures. Coherent structures make up over 80% of the energy in the wall region of the turbulent
boundary layer (say for y+ < 40, where y+ yurjv , and UT is the friction
velocity and v the viscosity; the friction velocity is defined as fidU/dy = pu^.,
where the gradient is evaluated at the wall). We want to develop a lowdimensional model of the flow in the vicinity of the wall, resolving just these
coherent structures. These structures are observed to be vortices lying parallel to, and close to, the wall, oriented in the streamwise direction. These
vortices are usually observed as unequal pairs, but sometimes as singletons.
On the up draft side of the vortex, slow-moving fluid is lifted away from the
wall, producing a low-speed streak, while a high-speed streak is produced on
the other side.
The coherent structures are observed to "burst"; that is, the updraft becomes more intense, adjacent vortices move toward each other, the inflection
in the streamwise velocity profile in the updraft intensifies and moves farther
from the wall, a secondary instability grows on this inflection, producing a
burst of Reynolds stress, the vortices are weakened as a result, they move
apart, and a "sweep" of high speed fluid comes down from the outer part
of the layer. This process is responsible for the production of most of the
turbulent kinetic energy in the layer, and it is this process that we want to
interfere with.
In the past decade our group at Cornell has developed a new approach
to turbulent flows([2], [47], [3], [33], [7]). We have applied our approach
primarily to the flat-plate boundary layer. The approach has become quite
popular, being picked up by many research groups (c.f. [5], [16], [29], [30],
[27], [31], [28], [32], [36], [37], [38], [42]), [46], [23], [22]). This is the approach
that we will apply to tell us how, when and where to interfere in the boundary
layer.
In the flat-plate boundary layer, we begin from the idea that the near-

wall region is energetically dominated by large-scale coherent structures. It


should thus be possible to construct a low-dimensional model of the wall

region resolving only these coherent structures, and parameterizing the less
energetic, less well-organized, smaller-scale turbulence. Such a model can be
used for many purposes, if only large-scale information is needed. The model
involves the following steps:
First we need a good set of basis functions. We use the Empirical Eigenfunctions of the Proper Orthogonal Decomposition, or the KarhunenLoeve decomposition ([3]), because these are optimal from the viewpoint of convergence. In the wall region, one eigenfunction is sufficient
to capture 80% of the fluctuating energy. These eigerifunctions arc
obtained either from experiment or from DNS.
By Galerkin projection, we obtain a set of ordinary differential equations for the coefficients in the decomposition ([34]). The decomposition is truncated, and the modes that are not represented explicitly
are parameterized by a gradient transport model. Some representation of these unresolved modes is necessary, to provide a sink for the
turbulent energy; however, we believe that the precise nature of the
dissipation mechanism does not affect the structure of the large scales.
A unique feature of our models is the utilization of separation of scales
between the coherent structures and the mean. This results in a cubic
term which globally stabilizes the system and allows a very compact
representation.
Models of the near-wall region of the turbulent boundary layer have
been constructed containing from four to 64 degrees of freedom ([45]);
ten degrees of freedom provides a model of the wall region that is satisfactory in many respects, although even the four-dimensional model

displays some of the characteristrics of the wall region.


The low-dimensional models are explored with various simulation programs constructed for this purpose ([47]), and analyzed using the techniques of dynamical systems theory.
Note that the mean time between bursts (in the laboratory or the
model) is closely related to the drag; in fact, Tv%./v is approximately

constant (where T is the mean time between bursts, UT is the friction


velocity and v is the kinematic viscosity), so that the drag coefficient
is inversely proportional to the inter-burst time.The ten dimensional
model displays the following properties:
1. The model displays an intermittent phenomenon with an ejection
phase and a sweep phase that strongly resembles the bursting phenomenon observed in the boundary layer. The probability distribution of inter-burst times has the observed shape ([48], [49], [50],
[35]). However, the time scales both for bursts and interburst
durations are unrealistically long, a fact that was not appreciated until recently, due to the inadvertent omission of a factor of
[ZqZ/a]1/2 from the equation in Appendix A of [3]. We believe that
this is due to the model's inclusion of only a single coherent structure, when in fact a succession of quasi-independent structures are
being swept past the sensor. We return to this below.
2. The low-dimensional model makes it possible to understand the
dynamics of the bursting process: the instability leading to the
burst, the burst itself and the reformation of streamwise vortices
are determined by wall-region dynamics, and hence will scale with
wall-region parameters; the time of occurrence of a burst, however, is triggered by the pressure signal from the outer part of
the boundary layer ([34]). Thus, the interburst time distribution
should scale jointly with both inner and outer variables.
3. In the presence of favorable and unfavorable pressure gradients,
the mean time between bursts (and hence, the inverse of the drag
coefficient) increases and decreases in qualitative accordance with
observation ([47]). That is, in a favorable pressure gradient, the
mean time between bursts increases, the drag coefficient drops,
and the boundary layer grows more slowly, in both nature and
the model, and vice versa in the case of an unfavorable pressure
gradient.

4. In a flow with curved streamlines, the body force associated with


the streamline curvature has the effect of intensifying or suppressing the coherent structures, making them grow faster or slower,
and come to bursting more or less frequently, according to whether

they are on the concave or convex surface [40].


5. In turbulent drag reduction by polymer additives, the structure
of the wall region remains approximately the same, but increases
in scale. It has been speculated that this is due to an increased
lossiness in the turbulent part of the flow, due to the extensional
viscosity of the polymer ([39]). When the ten-dimensional model
is used to describe a flow with a stretched wall region, mean drag
is reduced, and increased lossiness in the turbulence is required to
maintain the system, consistent with the speculative explanation
([4])6. The low-dimensional model has been used to explore the possibility of active drag reduction. Active drag reduction corresponds
to burst-suppression, that is, to keeping the system from burst-

ing as long as possible, increasing the mean time between bursts


(and consequently reducing the drag) ([20], [21], [19]). It has
been shown using the model that active drag reduction is formally

equivalent to drag reduction by polymer additives, which also acts


to suppress the bursting ([4]). Consequently, active drag reduction
can be expected to produce reductions in drag (50%-80%) similar
to those produced by polymers.
Models of this general type can be used whenever it is desired to simulate the large scales of a boundary layer in response to various effects.
We have just seen how such a model can be used to investigate one
possible drag-reduction technique. It can also be used ([34]):

1. To simulate the fluctuating pressure pattern on a surface under


a boundary layer, either to predict panel vibration or sonar selfnoise;
2. To simulate index of refraction fluctuations occurring in the surface mixed layer of the planetary boundary layer, giving rise to
multi-path interference in digital communication links.
The applications are limited only by the imagination. The point here is, that
through the model, we have established a relation between the time between
bursts (and hence the drag or momentum mixing) and various other physical

effects - polymer additives, pressure gradients, streamline curvature and so


forth.

The control strategy

Careful examination of the low-dimensional model using the tools of dynamical systems theory indicates that there is a circle of fixed points in phase
space, each fixed point corresponding to quiescent streamwise rolls, the coherent structures; each fixed point has an unstable direction, and if the system
trajectory is not quite on the fixed point, it will gradually spiral away from
it; when it is far enough away, it will jump to a diametrically opposed fixed
point, corresponding to quiescent streamwise rolls displaced sideways from
the first ones. The velocity field corresponding to this process matches well
(in a qualitative way) the bursting process; during the jump there is an intensified updraft, followed by a weakening and separation of the rolls, and a
downdraft. In the ten-dimensional model the updraft is not followed by a secondary instability, due to the truncation of the model. The 64-dimensional
model is more realistic, however. The pressure signal from the outer part
of the boundary layer is enough of a disturbance to keep the system a little
away from the fixed point, so that jumps to a diametrically opposed fixed
point occur at irregular intervals.
The time between these jumps is the only unrealistic thing about the
model - it is much longer than the observations. In the real boundary layer,
however, we are not always looking at the same coherent structure - different
coherent structures are swept past our observation point, and if we include
this factor, it restores the correct interburst time [44], [40].
Knowing how the system behaves in phase space, how can we interfere
with the flow in such a way as to temporarily keep the system from jumping,
to increase the mean time between jumps, since this will ultimately reduce
the drag? We have devised a scheme to delay the jump [20], [21], [19], [18].
The scheme involves an unexpected type of interaction: a transverse velocity
field, such as might be produced by a vortex singleton or pair adjacent to the
coherent structure. The effect of this transverse velocity field is to cause the
system point in phase space to rotate around the fixed point. We use this
rotation to direct the system point back to the fixed point when it tries to
escape; if the system point tries to go around the fixed point to the right, a

rotation is induced that brings it to the left, and so forth. (This geometric interpretation corresponds to the two-dimensional case; the multi-dimensional

case is necessarily more complex). The net result of this technique is an


average increase of the bursting period (corresponding to a decrease of the
drag) of the order of 50% in the best case.
We have controlled other sorts of models, in order to get a feel for the
process. For example, we have tried successfully two sorts of control of the
Kuramoto-Sivashinsky equation (a one-dimensional equation that has some
features in common with the Navier Stokes equation) [6].
Of course, this is control of models, not control of the real flow. We
have a successful Direct Numerical Simulation of this flow [13], [15], [14],
[12], with a moving boundary, and with vortices introduced by body forces.
We are implementing various control schemes in this flow, raising a burnp
under the high-speed streak, for example, or placing a vortex pair producing
an updraft just over the high speed streak. The vortex pair simulates the
velocity field produced by a sharp-edged flap. These attempts are not in any
sense optimized, nor are they particularly coordinated with our understanding of what is happening in phase space; nevertheless, we have achieved drag
reductions of the order of 20% in the vicinity of the applied control. Even
more interesting is the fact that the drag on the opposite wall falls by 10%

after a short lag. This presumably reflects the reduced turbulence production
due to the reduced bursting. Raising a bump under the high-speed streak
also reduces the drag locally, by both lifting the high speed fluid away from

the wall, and deflecting it into the low-speed streaks, increasing their speed.
Unfortunately, the effect is somewhat reversed when the burrip is lowered.

Actuators

We have considered as actuators both flaps of the type designed by Chih-Ming


Ho and co-workers [9], [10], [17], and Gaussian bumps. We imagined that a
Gaussian bump would produce a necklace vortex similar to that produced
around a bridge pier or telephone pole. Our simulation of this flow, however,
showed something completely different. We found that, as the flow splits
to pass around the bump, it generates a strong vortex pair with an updraft
between the vortices. Behind the bump, as the flow comes around to rejoin
itself, it generates another strong vortex pair of opposite sign, underneath
10

the first pair. This stack of two pairs is topped by the necklace vortex,
which is the weakest of the three, and is of the same sign as the lowest pair.
Hence, trailing downstream we have a stack of three pairs of alternating sign.
This complicated flow is not particularly useful from a control point of view.
As noted above, we have nevertheless managed to exert some control by
pushing the high-speed streak away from the wall, but for more complex and
controllable interactions, a simpler actuator velocity field would be desirable.
The field of a simple vortex pair, for example, will produce an up- or downdraft, or a cross-wind for adjacent structures. We suppose that a sharpedged flap will produce such a velocity field. It is not possible to resolve the
geometry of such a flap of small size in a DNS of a channel flow, so we have
been using body forces to induce a vortex pair. A DNS could be done of the
flow around such a flap in isolation, at large scale, but has not been done.

Acknowledgments

This manuscript describes work carried out over a ten year period and necessarily involving many past and present students and colleagues. I thank
them all for their contributions, but particularly Phil Holmes, Gal Berkooz
and Sid Leibovich.

References
[1] Anon. R&D scoreboard:1978. Business Week, page 52, July 2 1979.

[2] N. Aubry. A Dynamical System/Coherent Structure Approach to the


Fully Developed Turbulent Wall Layer. PhD thesis, Cornell University.,
1987.
[3] N. Aubry, P. Holmes, J. L. Lumley, and E. Stone. The dynamics of
coherent structures in the wall region of a turbulent boundary layer.
Journal of Fluid Mechanics, 192:115-173, 1988.
[4] N. Aubry, J. L. Lumley, and P. Holmes. The effect of modeled drag

reduction in the wall region. Theoret. Coraput. Fluid Dynamics, 1:229248, 1990.
11

[5] K. S. Ball, L. Sirovich, and L. R. Keefe. Dynamical eigenfunction decomposition of turbulent channel flow. Int. Jour, for Num. Meth. in
Fluids, 12:585-604, 1991.
[6] G. Berkooz. Controlling models of the turbulent wall layer by boundary

deformation. Technical Report FDA-92-16, Cornell University., 1993.


[7] G. Berkooz, P. Holmes, and J. L. Lumley. The proper orthogonal decomposition in the analysis of turbulent flows. Annual Review of Fluid
Mechanics, 25:539-575, 1993.

[8] R. F. Blackwelder. Some ideas on the control of near-wall eddies. AIAA


J., 1989. Paper No. 89-1009.

[9] Stuart F. Brown. Smart wings. Popular Science, page 59, March 1995.
[10] Malcolm W. Browne. Micro-machines help solve intractable problem of
turbulence. The New York Times, page B13, January 3 1995.

[11] D. M. Bushnell and C. B. McGinley. Turbulence control in wall flows.


Annual Review of Fluid Mechanics, 21:1-20, 1989.
[12] H. A. Carlson. Direct numerical simulation of laminar and turbulent
flow in a channel with complex, time-dependent wall geometries. PhD
thesis, Cornell University, 1995.
[13] H. A. Carlson, G. Berkooz, and J. L. Lumley. Direct numerical simulation of flow in a channel with complex, time-dependent wall geometries:
a pseudospectral method. Journal of Computational Physics, 1996. In
press.
[14] H. A. Carlson and J. L. Lumley. Active control in the turbulent bound-

ary layer of a minimal flow unit. Journal of Fluid Mechanics, 1996.


submitted.
[15] H. A. Carlson and J. L. Lumley. Flow over an emerging obstacle. AIAA
J, 1996. Submitted.
[16] D. H. Chambers, R. J. Adrian, P. Moin, D.S. Stewart, and H. J. Sung.
Karhunen-Loeve expansion of burgers model of turbulence. Phys. Fluids,
31:2573-257?, 1988.
12

[17] Jeff Cole. Miniscule flaps face big job on jets' wings. The Wall Street
Journal, page Bl, November 25 1994.
[18] B. D. Coller. Suppression of heterodinic bursts in boundary layer models.
PhD thesis, Cornell University, 1995.

[19] B. D. Coller, P. Holmes, and J. L. Lumley. Control of bursting in boundary layer models. Appl. Mech Rev., 47 (6), part 2:S139-S143, 1994.
Mechanics USA 1994, ed. A. S. Kobayashi.
[20] B. D. Coller, P. Holmes, and J. L. Lumley. Controlling noisy heteroclinic
cycles. Physica D, 72:135-160, 1994.
[21] B. D. Coller, P. Holmes, and J. L. Lumley. Interaction of adjacent bursts

in the wall region. Phys. Fluids, 6 (2):954-961, 1994.


[22] A. E. Deane, I. G. Keverkidis, G. E. Karniadakis, and S. A. Orszag. Lowdimensional models for complex flows: Application to grooved channels
and circular cyliders. Physics of Fluids A, 3(10):2337-2354, 1991.
[23] A. E. Deane and L. Sirovich. A computational study of Raleigh-Benard

convection part I. Rayleigh number scaling. Journal of Fluid Mechanics,


222:231-250, 1991.
[24] H. E. Fiedler and H. H. Fernholz. On management and control of turbulent shear flows. Progress in Aerospace Sciences, 27:305, 1990.

[25] M. Gad-el Hak. Flow control. Applied Mechanics Reviews, 42(10), 1989.
[26] M. Gad-el Hak. Interactive control of turbulent boundary layers: a
futuristic overview. AIAA J., 32(9):1753, 1994.

[27] M. Glauser, S. J. Leib, and W. K. George. Coherent structures in the


axisymmetric turbulent jet mixing layer. In Turbulent shear flows 5.
Springer-Verlag, 1987.

[28] M. Glauser, X. Zheng, and W. K. George. The strearnwise evolution of


coherent structures in the axisymmetric jet mixing layer. In T. Gatski et.

al., editor, The Lumley symposium: Recent developments in turbulence,


Newport News, VA Nov. 1990. Springer, 1990.
13

[29] M. N. Glauser and W. K. George. Orthogonal decomposition of the


axisymmetric jet mixing layer including azimuthal dependence. In
G. Comte-Bellot and J. Mathieu, editors, Advances in turbulence.
Springer-Verlag, 1987.
[30] M. N. Glauser and W. K. George. An orthogonal decomposition of
the axisymmetric jet mixing layer utilizing cross-wire velocity measurements. In 6th symposium turbulent shear flows, 1987.
[31] M. N. Glauser, X. Zheng, and C. R. Doering. The dynamics of organized
structures in the axisymmetric jet mixing layer. In M. Lesieur and
0. Metais, editors, Turbulence and coherent structures. Kluwer, 1989.

[32] A. Glezer, A. J. Kadioglu, and A. J. Pearlstein. Development of an


extended proper orthogonal decomposition and its application to a time
periodically forced plane mixing layer. Physics of Fluids A, 1:1363-73,
1989.
[33] P. J. Holmes. Can dynamical systems approach turbulence? In J.L.
Lumley, editor, Whither turbulence? Turbulence at the Crossroads,
pages 195-249, New York, 1990. Springer.

[34] P. J. Holmes, G. Berkooz, and J. L. Lumley. Turbulence, Coherent


Structures, Dynamical Systems and Symmetry. Cambridge University
Press, 1996.
[35] P. J. Holmes and E. Stone. Heteroclinic cycles, exponential tails and
intermittency in turbulence production. In T. B. Gatski, S. Sarkar, and
C. G. Speziale, editors, Studies in Turbulence, pages 179-189. SpringerVerlag, 1992.
[36] M. Kirby, J. Boris, and L. Sirovich. An eigenfunction analysis of axisymmetric jet flow. Journal of computational physics, 90 no. 1:98-122,
1990.

[37] M. Kirby, J. Boris, and L. Sirovich. A proper orthogonal decomposition of a simulated supersonic shear layer. International journal for
numerical methods in fluids, 10:411-428, 1990.

14

[38] Z-C. Liu, Adrian R. J., and T. J. Hanratty. Reynolds-number similarity


of orthogonal decomposition of the outer layer of turbulent wall flow.
Technical Report TAM 748, UILU-ENG-94-6004, University of Illinois,
Department of Theoretical and Applied Mechanics, 1994.
[39] J. L. Lumley and I. Kubo. Turbulent drag reduction by polymer additives: a survey. In B. Gampert, editor, The Influence of Polymer Additives on Velocity and Temperature Fields, pages 3-24. Springer-Verlag,
1985.
[40] J. L. Lumley and B. Podvin. Dynamical systems theory and extra rates
of strain in turbulent flows. Journal of Experimental and Thermal Fluid
Science, 1996. Peter Bradshaw Symposium; in press.
[41] P Moin and T. Bewley.

Feedback control of turbulence.

In A. S.

Kobayashi, editor, Mechanics USA 1994, volume 47(6)2 of Applied Mechanics Reviews, pages S3-S13, 1994. ASME Reprint No. AMR146.
[42] P. Moin and R. D. Moser. Characteristic-eddy decomposition of turbulence in a channel. Journal of Fluid Mechanics, 200:471-509, 1989.
[43] H. Park and L. Sirovich. Turbulent thermal convection in a finite domain, part ii. numerical results. Physics of Fluids A, 2 (9): 1659-1668,
1990.

[44] B. Podvin, J. Gibson, G. Berkooz, and J. L. Lumley. Lagrangian and


eulerian view of the bursting period. The Physics of Fluids A, 1996. In
preparation.
[45] S. Sanghi and N. Aubry. Mode interaction models for near-wall turbulence. J. Fluid Mech., 247:455-488, 1993.
[46] L. Sirovich and A. E. Deane. A computational study of Raleigh-Benard
convection part II. Dimension considerations. Journal of Fluid Mechanics, 222:251-265,1991.

[47] E. Stone. A Study of Low Dimensional Models for the Wall Region of
a Turbulent Layer. PhD thesis, Cornell University., 1989.

15

[48] E. Stone and P. J. Holmes. Noise induced intermittency in a model of


a turbulent boundary layer. Physica D, 37:20-32, 1989.

[49] E. Stone and P. J. Holmes. Random perturbations of heteroclinic cycles.


SI AM J. on Appl. Math., 50 no. 3:726-743, 1990.
[50] E. Stone and P. J. Holmes. Unstable fixed points, heteroclinic cycles
and exponential tails in turbulence production. Phys. Lett. /4, 155:2942, 1991.

16

You might also like