You are on page 1of 12

Global stability of a lid-driven cavity with throughflow: Flow visualization studies

C. K. Aidun, N. G. Triantafillopoulos and J. D. Benson

Citation: Physics of Fluids A: Fluid Dynamics 3, 2081 (1991); doi: 10.1063/1.857891


View online: https://doi.org/10.1063/1.857891
View Table of Contents: https://aip.scitation.org/toc/pfa/3/9
Published by the American Institute of Physics

ARTICLES YOU MAY BE INTERESTED IN

Linear stability of lid-driven cavity flow


Physics of Fluids 6, 2690 (1994); https://doi.org/10.1063/1.868158

Reynolds number and end-wall effects on a lid-driven cavity flow


Physics of Fluids A: Fluid Dynamics 1, 208 (1989); https://doi.org/10.1063/1.857491

Three-dimensional centrifugal-flow instabilities in the lid-driven-cavity problem


Physics of Fluids 13, 121 (2001); https://doi.org/10.1063/1.1329908

Direct numerical simulation of the flow in a lid-driven cubical cavity


Physics of Fluids 12, 1363 (2000); https://doi.org/10.1063/1.870387

Simulation of lid-driven cavity with top and bottom moving boundary conditions using implicit finite
difference method and staggered grid
AIP Conference Proceedings 2021, 020002 (2018); https://doi.org/10.1063/1.5062719

Experimental observation of the steady-oscillatory transition in a cubic lid-driven cavity


Physics of Fluids 23, 084106 (2011); https://doi.org/10.1063/1.3625412
Global stability of a lid-driven cavity with throughflow: Flow
visualization studies
C. K. Aiduma) N. G. Triantafillopoulos, and J. D. Benson
Institute of Paper Science and Technology, and School of Mechanical Engineering,
Georgia Institute of Technology. 575 Fourteenth Street, N. W;, Atlanta, Georgia 30318
(Received 16 April 1990; accepted 14 May 199 1)
Flow visualization studies of a lid-driven cavity (LDC) with a small amount of throughflow
reveal multiple steady states at low cavity Reynolds numbers. These results show that
the well-known LDC flow, which consists of a primary eddy and secondary corner eddies, is
only locally stable, becomes globally unstable, and competes with at least three other
steady states before being replaced by a time-periodic flow. The small amount of throughflow
present in this system seems to have no qualitative effect on the fluid flow characteristics.
These observations suggest that multiple stable steady states may also exist in closed LDC’s.
Since stability properties of the closed LDC flows are virtually unexplored, the present
flow visualization results are interpreted by first proposing an expected behavior of an idealized
(free-slip end walls) LDC and then treating the problem at hand as a perturbation of
the ideal case. The results also suggest that there are nonunique and competing sequences of
transitions that lead the flow in a LDC from laminar steady state toward turbulence.

I. INTRODUCTION fer into the blade nip and, consequently, affecting coat
weight and its uniformity. There is a fundamental differ-
Understanding the stability properties of flow in a lid-
ence between the characteristics of the layer entering the
driven cavity (LDC) is important because of fundamental
blade nip of a short-dwell coater compared to the nearly
as well as practical reasons. This system represents a sim-
irrotational layer’ in a blade coater with a roll applicator
plified version of many manufacturing devices. Short-dwell
system. In the case of a short-dwell coater, a viscous layer
coaters’ and flexible blade coaters2 used for production of
forms near the dynamic contact line, upstream of the
high-grade paper and photographic films, and melt-spin-
ning processes in forming continuous metal ribbons of mi- blade. This layer is susceptible to instabiIity and interaction
crocrystalline material3 are a few examples. Flow instabil- with the three-dimensional patterns that form when the
ity in the cavity of these processes could have adverse nearly irrotational eddies inside the pond destabilize. These
effects on the quality of the manufactured product. Let us interactions generate a nonuniform wavy layer entering the
examine one of these systems in detail. blade nip. This in turn can have adverse effects on the coat
The trend toward increasing machine speed while re- weight profile.
ducing coat weight with a short-dwell coater (Fig. 1) has The idea that the flow inside the pond upstream of the
been hindered by the difficulty in maintaining uniform blade influences the coating surface quality and the coat
cross-directional coat weight profile, especially in the pro- weight distribution was first suggested by Higgins6 during
duction of lightweight coated papers. Spatially and tempo- experimental investigation of puddle coaters, which are
rally periodic, sometimes even random, coating thickness characterized by a free surface pool of liquid that forms
variations create an uneven coat weight profile, which, in upstream of the blade. Extensive experiments with short-
its extreme case, appears as a patterned surface character- dwell coaters4 also indicate that the flow in the pond in-
ized by streaks running along the machine direction. Our fluences the coated surface quality. It is therefore impor-
experiments with short-dwell coaters show that streaks 2-8 tant to understand the flow behavior inside the pond for
cm in width, also called patches, comprise areas of com- the Reynolds numbers ranging from 300 to 1000, which
paratively low film thickness that are independent of the represent the operating conditions of these systems.
substrate’s surface. The hydrodynamic characteristics of The flow characteristics in these and many other engi-
this system are not well understood. There are reports that neering devices share common features with the simple
disturbance of the coating head can permanently change LDC system. Therefore, understanding the stability prop-
the coated film characteristics-a condition that cannot be erties of flow in a LDC not only increases our knowledge of
tolerated in the manufacturing of commercial products. In fundamental issues but also provides the foundation for
particular, it is suggested4 that three-dimensional hydrody- analyzing and improving the performance of a broad class
namic instabilities inside the pond may be responsible for, of engineering and manufacturing systems.
and hence correlate with, wet-film thickness variations On the fundamental side, a LDC is the simplest system
across the machine direction. Disturbances in the pond can that can be used to study the stability properties of a con-
be transported, influencing the mass and momentum trans- fined flow with closed streamlines. In fact, this is exactly

‘)Author to whom correspondenceshould be addressed.

2081 Phys. Fluids A 3 (9), September 1991 0899~8213/91 I092081 -11$02.00 @ 1991 American Institute of Physics 2081
start-up, and development of Taylor-Goertler-like vortices
(TGL) at the downstream secondary eddy (DSE) before
the appearance of turbulence at Re> 6000. The existence
of TGL vortices is attributed to the concave viscous layer
that forms between the primary and the DSE and their
appearance is viewed as a mechanism for a transition to
turbulence. Using cavities with different aspect ratios, Ko-
seff and Street show that because of the end-wall effects,
the size of DSE increases as Re increases from 1000 to
10 000 for the span-to-width aspect ratio (SAR) equal to
3:l; while for SAR’s of 1:l and 21 DSE reduces in size for
Re > 2000. Extensive laser Doppler anemometer (LDA)
measurements of flow in cavities with SAR less than or
equal to 1:l by Prasad and Koset? and by Bogatyrev and
Gorin16 in cavities with SAR greater than or equal to l:l,
shed light on the effects of geometry and end walls. For
SAR > l:l, the effects of SAR on turbulent flow at high Re
are less significant than at lower Re. Prasad and Koseff
FIG. 1. Schematic of a short-dwell coater, conclude from their measurements that ( 1) reducing the
span increases the viscous drag from the end walls and
thereby reduces the overall velocity inside the cavity, and
the reason it was chosen by Burggraf7 and Pan and Ac-
rives* in their studies of viscous effects on confined vorti- (2) the major portion of the velocity fluctuations at lower
ces. Re is due to TGL vortices.
There has also been significant progress in the numer-
ical simulation of cavity flow at high Re values. The fact
A. Lid-driven cavity flow that TGL vortices appear at the DSE region is shown nu-
The two-dimensional LDC was first studied in detail merically by Kim and lvloin,” Freitas et a/.,‘* Freitas and
by Burggraf,7 who solved the Navier-Stokes equation for Street,” Perng and Street,” and Iwatsu et aZ.2’With the
flow in a square cavity for Reynolds number Re = H/v exception of Kim and Moin, who use periodic boundary
from 0 to 400 (here, V, Z, and Y are the velocity of the lid, conditions at the end walls, all the other investigators men-
width of the cavity, and the kinematic viscosity of the fluid, tioned above use no-slip condition for all boundaries. Fre-
respectively). Pan and Acrivos’ used LDC’s to examine itas and Street provide a detailed analysis of the three-
Prandtl’ and Batchelor’s’” theorem, which states that a dimensional flow with SAR = 3:l at Re = 3200. In
steady flow with closed streamlines at high Reynolds num- addition to extensive particle track plots showing the three-
ber should form an inviscid core surrounded by a thin dimensionality of the flow patterns, they show temporal
shear layer at the boundary. Through extensive experi- variations of the TCL vortices, which agree with Koseff
ments, they concluded that ( 1) as Re- ~13,the flow in a and Street I4 flow visualizations and our observations dis-
LDC with a finite aspect ratio consists of a single inviscid cussed below. Iwatsu et al. simulate a cubic LDC flow at
core of uniform vorticity with an infinitesimally thin Re from 100 to 4.000. They find steady solutions for Re up
boundary layer along the solid boundary, and (2) in cav- to and including 2000 and unsteady solutions for
ities with infinite depth, since the size of the primary vortex Re = 3000. They indicate that at high Re but steady how,
continues to increase with Re, its core does not become the significant transverse variations of flow suggest occur-
inviscid, even as Re-+ 00, or the fluid viscosity becomes rence of TGL vortices. Ku et alz2 solve the three-dimen-
infinitesimally small-a paradoxical conclusion. sional flow in a cubic cavity using a pseudospectral
The two-dimensional cavity flow has been computed method; but contrary to Iwatsu et al., they do not obtain
accurately for Re up to 10 000 by several investigators, steady-state TGL vortices. Pemg and Street” also show
including Ghia et al.,” Schreiber and Keller,12 and more that TGL vortices are not steady for cavities with
recently, Thompson and Ferziger I3 (maximum SAR> 1:l.
Re = 5000). These studies, along with numerous other nu- The studies mentioned above have shown the existence
merical calculations, focus on flow in a two-dimensional of complex hydrodynamic phenomena in this system. Ma-
plane and do not consider the three-dimensionality of the jor fundamental questions have remained unanswered; ( 1)
system. what is the sequence of transitions or the initial bifurca-
To study the transition to turbulence in confined recir- tions in the laminar regime that lead the system toward
culating flows, Koseff and Street14 studied the three-dimen- turbulent flow; and (2) is there a unique sequence of tran-
sional aspects of flow in a cavity for Re from 1000 to sitions to turbulent flow in this system? The stability char-
10000. In their study, the flow starts impulsively by a acteristics and the dynamics of laminar flow in this system
sudden acceleration of the top surface to the desired speed. have virtually remained unexplored.
Several interesting phenomena are revealed, including a Recent developments and conclusions from a set of
local Taylor instability near the downstream lip during numerical studies of flow in a grooved channel (Fig. 2) are

2082 Phys. Fluids A, Vol. 3, No. 9, September 1991 Aidun, Triantafillopoulos, and Benson 2082
h
(a)
G
K7/ 0 HYDRAULIC
FLUID INLET

= ARM

FIG. 2. Schematic of flow in a grooved channel.

of interest to this discussion. Flow in a channel with a


rectangular groove consists of a distorted channel flow cou-
pled with a shear-driven cavity system.23 The flow in the
cavity has features similar to a LDC flow, that is, a pri-
mary eddy forms at the core with secondary eddies in the
comers. This system has been studied numerically by
Ghaddar et a1.24 and Amon and Patera using a varia- -.
interface between DSE and DSE prim& vortex
tional spectral element method. They assume a system with the primary vortex
infinite span and therefore impose periodic boundary con- 4 S w
ditions in the spanwise direction. One of their conclusions
is that, depending on the stability and criticality of the
primary bifurcation in a wall-bounded shear flow, the tran- FIG. 3. Schematics of the experimental setup; (a) showing the coordinate
sition to turbulence could be broadband and abrupt, as in system (xg,z) and the corresponding velocity components (u,u,w); (b) a
plane Poiseuille flow, or low dimensional and ordered. three-dimensional view of the cavity demonstrating the roll-shaped pat-
terns of the steady-state primary mode.
Similarly, the initial bifurcations in a LDC flow could pro-
vide the key to understanding the physics of transition to
unsteadiness and turbulence in confined flows with closed
streamlines. the lower section that is connected to a pump and acts as a
By focusing on the flow states at low Reynolds number reservoir for replenishing the fluid that escapes the cavity.
(Re < 2000) in a LDC with a small amount of through- Fluid is supplied to the lower section of the cavity via
flow, we show that multiple locally stable states exist and a 1.00 cm pipe tap from one side. The symmetry of the flow
compete with each other. This suggests that different patterns show that the flow through the slot into the cavity
mechanisms for transition to turbulence may exist in this is uniform over the span. All bounding walls of the cavity
system. Furthermore, we propose that the steady states are 12.5 mm thick, fabricated from Plexiglas. An impor-
observed in our experiments may be identified as the de- tant feature considered in the design of the cavity assembly
coupling of the nongeneric state diagram of the idealized is that of allowing a precisely uniform film of fluid to es-
confined LDC system characterized by free-slip end walls. cape downstream when the roll, which drives the flow, is
rotating. This is achieved by mounting the cavity on a
finely adjustable support plate, which can be pivoted on
II. EXPERIMENTAL SETUP
two removable screw pins, one on each of the opposite
The cavity, placed on top of a lower compartment [Fig. sides of the plate. The adjustable plate, in turn, is mounted
3 (a)], makes contact with the roll and has width H = 5.08 on a support frame that is hydraulically pressurized with
cm in the direction of roll motion, depth D = 5.08 cm, and two air cylinders so that the cavity makes contact with a
span S = 15.24 cm transverse to the direction of roll mo- chrome-plated roll 59.7 cm in diameter. A 1.0 cm thick
tion. Thus the characteristic geometric parameters are 1:l Teflon collar around the cavity lip [ABCE in Fig. 3(b)]
and 3: 1 for depth- and span-to-width aspect ratios, respec- reduces friction and allows for smooth contact between the
tively. The lower compartment has the same dimensions cavity and the roll surface.
but a depth of only 2.54 cm and serves the purpose of In addition to the cavity Reynolds number, a second
feeding fluid uniformly into the cavity. parameter that depends on the net mass flow rate m is
A very thin film of fluid adheres to the roll and escapes defined as R’ = m/,uS. Therefore when RI-0 (i.e., m-+0)
the cavity through the downstream lip EC. This layer of the system approaches a closed LDC. In the experiments
fluid is scraped off from the roll by a sharp blade down- reported here, the value of R’ is varied from 0.5 to 1.5. In
stream of the cavity. At the bottom of the cavity adjacent this range, variations in R’ did not change the qualitative
to the GK comer, a narrow slot (3 mm wide) opens into behavior of the flow inside the cavity. Based on this, we

2083 Phys. Fluids A, Vol. 3, No. 9, September 1991 Aidun, Triantafillopoulos, and Benson 2083
suspect that at least the qualitative behavior of the flow is
identical to a closed LDC.
All the experiments reported here contain a positive
net mass flow rate. It is difficult to reduce R' to absolute
zero. The difference between various experimental setups is
in the manner by which the small amount of lost fluid is
replenished. For example, in an alternative setup, the cav-
ity can be submerged completely into a pool of liquid and
the fluid lost from the downstream lip will then be replen-
ished from the upstream side.14’i5 When the focus is on
closed LDC Bow, then the loss in fluid is carefully mini-
mized to avoid any significant influence on the flow inside
the cavity.
The experiments are conducted with a viscous New-
tonian fluid and the suspended-particle technique is used
for flow visualization. The fluid continuum is a glycerine
and water mixture with disk-shaped high reflectance alu-
minum (20-40 pm long and -5 pm thick) or Kalliro-
scope (6 X 30X 0.07 pm) flakes. When in motion, these
particles align along the axis of principal normal strain26 (d)
and rapidly reorient themselves due to changes in the flow
pattern. The time scale for particle realignment is much FIG. 4. Sequenceof transitionsin the primary state; (a) one-cell steady
state, Re- 500; (b) time-periodic state, Re - 9(w; (c) wavy downstream
shorter than the time scale of the flow. This technique is secondary eddy starts to fold Re- 1200; (d) Goertler-like vortices appear
ideal for obtaining single images of the instantaneous state in the form of mushroom-shaped structures, Rez 1900.
of flow and, by utilizing a video, for observing the devel-
opment and evolution of time-periodic and unsteady flows.
The concentrations of the suspended particles in all of the nel, while in our experiment, the depth of view is about l-3
experiments reported here are less than 0.1% by weight. cm into the cavity. This provides a good view of the flow
Illumination of the cavity is achieved with a regular white pattern features and their cellular structure.
light concentrated on the downstream vertical wall.
Since the working fluid is Newtonian, a dilute suspen- III. FLOW VlSUALlZATlON RESULTS
sion of flakes also behaves as a Newtonian fluid with an The fn-st set of experiments focused on the sequence of
effective viscosity, ,u,f = ,u( 1 + 2.5~~)) where p and a! are transitions from steady flow to an unsteady state in the
the fluid viscosity and the volume fraction occupied by the range of Re from 100 to 2000.
solid. This relation, first derived by Einstein (see Batche- We evaluated phenomena appearing in a driven cavity
lor27) is valid when a < 0.02. A concentration of 0.1% by with throughflow using real-time observation and photo-
weight of aluminum flakes in our experiments represents a graphs of flow patterns. In the first set of experiments, the
volume fraction a! = 0.0005, therefore peff = l.OOlp, and cavity Reynolds number is increased in increments by in-
for all practical purposes, the effective viscosity can be as- creasing the speed of the roll. The basic steady-state flow
sumed to be equal to the fluid viscosity. pattern, from the side endview, in a cavity with through-
Most of the past flow visualizations of the LDC have flow is depicted schematically in Fig. 3 (b). Fluid dragged
used particle tracing techniques where a sheet of light is downstream reaches edge EC, where most of the fluid fol-
formed to illuminate a cross-sectional plane of the flow lows wall EFNC and only a relatively small amount exits
(for example, see Rhee et aL2*). The photographs show the cavity as a thin film deposited onto the roll surface.
particle streaks, which for a steady-state flow are essen- Flow separation occurs along the wall EFNC, resulting in
tially the projection of the velocity vectors onto a two- the formation of a secondary recirculation in the vicinity of
dimensional plane. In this investigation, we wanted to vi- the edge FN. This is the so-termed downstream secondary
sualize the flow patterns and features that do not eddy (DSE), which counter-rotates with respect to the
necessarily remain with a group of particles. After exam- primary eddy. In contrast, the upstream secondary vortex
ining several techniques (dye injection, glass fibers, and does not exist in the throughflow cavity due to positioning
x-ray radiography) aluminum flakes with regular white of the feeding stream at GK.
light illumination were found to be most effective. In con- The above simplistic description refers to the basic
trast to the tracer particle techniques where a light sheet is “two-dimensional” steady-state flow. In three dimensions,
used to extract information from a single cross-sectional the primary and downstream secondary vortices comprise
plane, here the depth of field is not necessarily uniform cylindrical rolls extending along the span of the cavity,
throughout the plane of view, but is dependent on several while the line JI separating the two rolls is an interface.
factors, such as particle concentration, illumination tech- The focus in the first set of experiments is on the evolution
nique, and the flow pattern itself. Carlson et a1.26were able of the steady state into time-periodic and unsteady states as
to visualize the flow through the entire depth of their chan- Re increases.

2084 Phys. Fluids A, Vol. 3, No. 9, September 1991 Aidun, Triantafillopoulos, and Benson 2084
f 800
flDryCD DD
f fb
Results
(2-D)
Numerical
I

f” I). 0 Burrgraf.
h
Boz‘
1966
&Dalton,
xna”
[ Re
c
Nallashamy & kmc
1973
1977
600

400

200
0 100 200 300 400 500 600 1
w
01
0.5 0.6 0.7 0.9 1.0
FIG. 5. Size of the downstream secondaryeddy.

FIG. 6. Flow visualization measurementsof the critical Reynolds number


All in this study are from the EFNC
the pictures for transition from steady to time-periodic Bow of the primary state
plane. Figure 4 shows four qualitatively distinct flow re- (0.5 > R'> 1.5,~(,,, = 120mPa set).
gimes as Re gradually increases. As soon as the flow starts,
a steady roll-shaped pattern featuring primary and second-
ary vortices appears in the cavity. Fig. 4(a), the cross sec- the time-periodic waves become more clearly visible.
tion of which is the familiar LDC flow [Fig. 3(a)], shows Many experiments were run with the same procedure
the primary eddy and the DSE as roll-shaped patterns. outlined above to determine the critical Reynolds number
This is the basic steady-state flow from the front view of for transition to time-periodic state using the flow visual-
the cavity with throughflow. The dark horizontal line that ization procedure outlined above. Our estimate of the crit-
appears along the span of the cavity is the separation line ical Reynolds number is between 825 to 925. The reason
between the primary and the secondary downstream rolls we give a range for the critical Re is because it is difficult
depicted schematically in Fig. 3(b) as line JI. Both the to judge when the flow becomes time periodic. At onset,
separation and the reattachment points of the flow on the the amplitude of the growing disturbance mode is small
front and bottom walls, respectively, remain stationary up and difficult to detect. Twenty tests were run at a viscosity
to Re-825. range between 70-80 mPa set and the average critical Rey-
The separation line in Fig. 4(a) is straight and almost nolds number was about 870. Twenty tests were also run at
parallel to the lower boundary everywhere, except near the a viscosity range between 95-120 mPa sec. The average
side walls where it curves upward. It is symmetric with critical Reynolds number within that range was 880. The
respect to the centerplane QRST. To confirm that this line minimum Re for the transition in all the experiments was
represents the demarcation between the two primary and 825. Figure 6 presents the experimental data as the fluid
secondary recirculating rolls, the growth of its character- viscosity was increased from 70 mPa set up to the maxi-
istic length scale (i.e., the distance d of the separation line mum value, pmax, of 120 mPa sec.
from the bottom boundary) is compared with other exper-
imental and numerical results. The value of d/D was visu-
ally estimated by measuring the distance of the separation
line from the edge of the bottom wall at the vertical cen-
terline, SR, which is then normalized by the cavity depth,
D, and plotted in Fig. 5 along with results from other
investigators. The height d increases sharply with Re up to
400; thereafter its rate of growth reduces substantially as
its value reaches a plateau. Under the standard conditions
of experimentation, the trend observed here is in good
agreement with other investigators.
In most cases during the beginning of the experiment,
the fluid is allowed to stabilize for about 10 min at the
Reynolds number of 500. At this stage, the primary and
the downstream secondary eddies are fully visible and
share a smooth interface across the span. The lid speed is
increased in intervals of 3 min by 1 m/min. The primary
and downstream secondary eddies continue to be stable up 0.7! *,-I-
880 900 920 940 960 980 1000 1020
to a Reynolds number of about 825. As the Reynolds num-
ber is increased beyond 825, small-amplitude time-periodic Re
waves appear on the DSE starting out in the middle of the
DSE and extending toward the end walls. As Re increases, FIG. 7. Oscillation period of the primary state (TV = 3 set).

2085 Phys. Fluids A, Vol. 3, No. 9, September 1991 Aidun, Triantafillopoulos, and Benson 2085
FIG. 8. A schematic demonstration of the time-periodic vortex sup&m-
posed on the downstream secondary eddy.

The critical oscillation period at onset, r. is about 3 w I


set and decreases, as shown in Fig. 7, by about 20% at
Re = 1000, where a second transition takes place. It ap-
FIG. 9. Multiple stable (a) two-cell, (b) three-cell, and (c) four-cell
pears that in the time-periodic mode, spiral-shaped vortices secondary steady states competing with the primary (one-cell) steady
superimpose on the primary and the downstream second- state of Fig. 4(a).
ary eddies traveling from the symmetry plane outward to
the side walls [Fig. 4(b)]. A schematic diagram of these
vortices is presented in Fig. 8. The disturbance structure from left to right, thus giving the appearance they are dis-
near onset is more visible on the DSE rather than the appearing into their neighbors and reappearing a short
primary eddy. The motion of these vortices is similar to the while later.
rotation of a spring around its axis. Pairs of vortices are In general, these mushrooms are nonuniform in size
generated at the centerline and each member of a pair and have irregular spacing. Their size shrinks with increas-
accelerates toward the end walls. It is interesting to note ing Re but their number remains unchanged. Six mush-
the similarity between the behavior at the onset of the roomlike structures are counted across the cavity in the
time-periodic state, presented in Fig. 8, and the spanwise whole range of Re tested, i.e., the TGL vortices having an
motion of spiral-shaped vortices at Re = 5000, illustrated average wavelength of half the cavity width (i.e., 2.54 cm),
by Kaseff and Street ’s 14(b) Fig. 7, where the flow is un- as shown in Fig. 4(d). This agrees with the numerical
steady and nonperiodic. simulations of Kim and Moini7 at Re = 1000, which show
As the Reynolds number approaches 1000, the separa- two pairs of TGL vortices per cavity width. The wave-
tion line between the primary and secondary recirculations length decreases with increasing Re. This was shown by
becomes irregular, featuring wavy patterns like the ones in Koseff and Street,‘4(a) who observed eight pairs of TGL
Fig. 4(c). Discrete vertical spikes appear simultaneously vortices at Re = 3000, and 11 pairs at Re = 6000, and later
inside the DSE and in the middle of the crest. Irregularity confirmed numerically by Freitas and Street,” who in their
of the separation line and the frequency of the waves’ ap- numerical simulation at Re = 3200 obtain eight pairs in
pearance increases with the Reynolds number Re. At Re in addition to end wall eddies.
the range of 1000-1300, traveling waves are engaged in a The second set of experiments concentrates on the
rapid movement that appears as oscillation of wave crests steady-state regime at low Re. The roll-shaped flow of Fig.
along the span of the cavity. Spikes inside the DSE ap- 4(a) consistently appears when the lid speed increases
proach the end walls and seemingly bounce back, thus gradually from zero. We found that when the lid suddenly
generating the perception of a complex spanwise move- decelerates from the unsteady state (Re- 2000) to a low
ment throughout the whole cavity. It is clear that at this speed (Re < 500), however, the steady roll-shaped primary
stage the dynamics of the flow are more complex than a state [Fig. 4(a)] may or may not recover. In its place
periodic system with a single frequency. steady cellular patterns with unique features may stabilize.
Upon continuous increase of the roll speed, the spikes We have identified three secondary steady states that com-
grow small crowns at their tops, which give them a mush- pete with each other and with the primary state. The front
roomlike form, like the one depicted in Fig. 4(d). These view (i.e., through the CEFN plane) of these structures is
are most likely the TGL vortices reported in previous stud- shown in Fig. 9. This behavior of the system shows that the
ies.14Originally mushrooms move about a stationary posi- primary steady state is globally stable only at very low Re;
tion keeping their points of contact fixed on the bottom it becomes globally unstable to finite-amplitude distur-
wall. At this point, the structures in the DSE interact with bances before destabilizing locally to infinitesimal time-pe-
the primary recirculation by creating vertical ribs between riodic disturbances. It is well known that in the limit as Re
mushrooms, which extend to the full depth of the cavity. approaches zero, the only possible flow is the primary
As the speed slowly increases, the flank and stem of the (Stokes) state. Therefore the secondary steady states,
mushrooms oscillate in phase. Eventually mushroomlike which are described below, can exist only at a finite value
structures fluctuate, causing the flanks to rapidly alternate of Re.

2086 Phys. Fluids A, Vol. 3, No. 9, September 1991 Aidun, Triantafillopoulos, and Benson 2086
center
points

saddle point

no-slipboundary
saddle points
t (3-CELL) (b)
saddle point

FIG. 10. The critical points in a two-dimensional cavity flow.

Due to the complex nature of the three-dimensional


flow patterns that are observed in the experiments, it is
difficult to conceptually visualize their structures. Very of-
ten, two patterns that appear different due to different vi- (4-CELL) Cc)

sualization techniques used in the experiment, may, in fact,


represent identical flow states. A topological description of
the flow patterns based on critical point theory29 provides
a framework and methodology for avoiding confusion. A
three-dimensional flow pattern can have one or more crit-
ical points, which are defined as points in the flow field
where the fluid velocity goes to zero (stagnation or sepa-
ration points) and the slope of the streamlines becomes
indeterminate. Several types of critical points are identified FIG. 11. Topological characterization of the two-dimensional projection
depending on how the streamlines form. of the three-dimensional patterns based on critical-point theory. The (a)
A flow field can be uniquely characterized once all of two-cell, (b) three-cell, and (c) four-cell structures correspond to the
its critical points are correctly identified. In an experimen- three patterns presented by Figs. 7(a), 7(b), and 7(c), respectively.
tal approach, as in this study, these points are identified by
tracing the streamlines with various visualization tech-
niques. In a computational approach, however, a Taylor line consists of two saddle points; one at the no-slip bottom
expansion of the velocity, in terms of the space coordinates wall (at the center of the bottom plane), and the other one
and subsequent analysis of the leading terms, provides about the center of the cavity, as shown in Fig. 11 (a). Note
complete information about the points. the difference between a complete saddle point inside the
For illustrative purposes, let us first analyze the two- flow and one formed on a no-slip boundary. There are two
dimensional lid-driven cavity flow. Because of its limited stable foci on each side of the symmetry line, and each side
degree of freedom, a two-dimensional flow pattern can only wall contains a no-slip boundary saddle. Also, the bottom
consist of centers and saddles. Each of the primary and wall contains two additional boundary saddle points on
secondary vortices forms a center that essentially consists each side of the symmetry line. Thus the first three-dimen-
of a stagnation point at the center of closed streamlines. sional flow pattern consists of one saddle, five no-slip
The interface between the primary and secondary vortices boundary saddle points, four corner saddle points, and four
is a separation line that connects two saddle points on the stable foci. We shall refer to this flow pattern as a two-cd
no-slip boundary (wall) of the cavity, as shown in Fig. 10. structure. The number of cells in the flow is determined by
Therefore neglecting the tertiary eddies, a two-dimensional the number of full saddle points. A full saddle point exists
cavity flow contains four center points (the primary eddy between two adjacent cells.
plus three secondary vortices), six saddle points (two for The second pattern, Fig. 9(b), is a three-cell structure
each secondary vortex) on the no-slip walls, and four quar- where each ceil contains two stable foci and a no-slip
ter saddles at the corners. boundary saddle at the bottom border. There is a saddle
The analysis of a three-dimensional pattern is consid- and a no-slip boundary saddle between each cell, and a
erably more involved. It is usually more productive to con- no-slip boundary saddle on each side wall. Therefore the
sider the projection of a three-dimensional pattern onto a three-cell structure contains six stable foci, two saddles,
two-dimensional plane. Let us first examine the three-di- four corner saddles, and seven no-slip boundary saddles, as
mensional pattern of Fig. 9 (a). Here, the flow is symmetric illustrated in Fig. 11 (b). Similarly, the four-cell structure
with respect to the centerline of the cavity. The symmetry [Fig. 9(c)] contains eight stable foci, three saddle points,

2087 Phys. Fluids A, Vol. 3, No. 9, September 1991 Aidun, Triantafillopoulos, and Benson 2087
four corner saddles, and nine no-slip boundary saddle
points [Fig. 11 (c)J The foci in the four-cell structure are
not as clearly distinguishable as the ones in the other two
flow structures. Our experimental technique does not allow
visualization of the cross-sectional plane of the Aow pat-
terns in Fig. 9. However, the cross sections of the flow at
various spans could consist of a deformed primary cell and
a DSE, similar to the primary state or the flow shown in
Fig. 10 of Prasad et ~1.~’

IV. DISCUSSION
Since the stability properties of LDC flows have re-
mained unexplored, at this time we outline the expected
behavior of LDC flows based on fundamental principles FIG. 12. The state diagram for an ideal lid-driven cavity system with
applied to confined systems such as the circular Couette or free-slip end walls.
Rayleigh-B6nard convectian flows. Assuming that these
principles also apply here, this will provide a frame of
reference for discussion and interpretation of the flow vi- where Z = 2n(z - a)//Z, /z is the wavelength of the peri-
sualization results. odic cellular flow, and a represents an arbitrary phase of
the disturbance structure.
A. Expected qualitative behavior of flow in a Now consider a cavity with finite span, L, and free-slip
confined LDC end walls. The boundary conditions at the end wall planes,
Let us begin with the simplest case-a LDC with as- Xl,, are given by
pect ratio H/D = I and infinite span. The Navier-Stokes au au
and continuity equations given by w=O and z=z=o on do,. (2’)
u,~+ Re(u*Vu) =Vp + V2u, in 0, (14 Here also the steady two-dimensional solution (3) sat-
V*u=O, in 0, (lb) isfies the boundary conditions (2’). The bifurcating 3-D
solution at onset can be represented by (4) with the eigen-
govern the flow in the cavity domain defined by 0. The function modified to m-z/L, where n is the wave number of
velocity vector, u, pressure, p, and time, t, are scaled with the destabilizing disturbance structure inside the container,
the lid velocity, V, pressure scale, pvV/D, and time scale, and the phase angle, a, being no longer arbitrary, is set to
P/Y, respectively, and the cavity depth, 0, is used as the zero. Modes representing an even or odd number of cellu-
length scale. The Reynolds number, Re, defined as VWv, lar patterns can become critical and destabilize the 2-D
is the only parameter in this problem. The boundary con- base solution, as shown in Fig. 12. Here the vertical (hor-
ditions are no-slip walls, given by izontal) plane represents solutions with an odd (even)
ulan,= ( V,QO) and UIJO - ZL,= (QW), (2) number of cellular patterns.
Since the slip condition at the end wall given by (2’) is
where Xl is the boundary of a and XX, represents the top also a symmetry condition, the similarity mapping used by
surface. At a sufficiently small value of Re, the solution to Aiduns3 applies. Therefore, by calculating the stability
this system represents a two-dimensional (2-D) flow given boundary for a single cell pattern, R,‘(L), the entire sta-
by
U2-D= [~(-%~>&,~>,ol* (3)
This solution is available for creep flows in closed form
by Pan and Acrivos,* and in numerical form for a wide
range of Re by many investigators (for example, see Boz-
eman and Dalton;31 or Nallasamy and Krishna Prasad3*).
In practice, however, this solution becomes unstable at a
critical Reynolds number, R, and most likely gives rise to
a steady cellular flow through a pitchfork bifurcation. The
value of R, and the wavelength of the critical mode, h,,
depend on the aspect ratio, H/D. The three-dimensional
(3-D) solution near onset can be approximated by sepa-
rating variables in the linearized disturbance equation and
using trigonometric representation in the z direction. The
critical disturbance at onset is then given by FIG. 13. An exampIe of the stability boundary showing the similarity
relation between the corresponding critical modes as a function of the
id= [u’(x,y) cos Z,v’(x,y) sin Z,w’(x,y) sin Z], (4) span L.

2088 Phys. Fluids A, Vol. 3, No. 9, September 1991 Aidun, Triantafiliopoulos, and Benson 2088
time-periodic time-periodic
. .** , .*’
. .* .
.*

FIG. 14. Decoupling of a pitchfork bifurcation for a three-dimensional FIG. 15. An example of a possible state diagram for a lid-driven cavity
lid-driven cavity. with throughflow.

bility boundary for the initial bifurcation point can be boundary conditions is unique and unconditionally sta-
mapped using transformation ble;37 and (3) the Leray-Schauder (L-S) degree theory
applies, that is, except at bifurcation points, the total num-
R,“(L) =R,‘(L/n), where n= 1,2,... . (5) ber of solution branches at a given value of Re is odd and
The expected form of the stability boundary predicted by the summation of the L-S indices for stable ( - 1) and
Eq. (5) is shown in Fig. 13. unstable ( + 1) solutions will be unity.38
Let us next discuss the real LDC system where the Given these facts, and considering the addition of no
no-slip condition applies to the end wall, as well as any slip as a perturbation of the ideal system, we expect the
other solid boundary. The no-slip condition generates an pitchfork bifurcation to decouple in the standard manner
Eckman-type viscous layer at the end walls, which results represented in Fig. 14. The decoupling generates two stable
in a radially inward motion of the fluid toward the center steady states and one unstable branch through a turning
of the primary eddy near the wall. This can be demon- point satisfying the conditions outlined above. The first
strated by application of the radial component of Euler’s steady branch that represents the only solution near
equation to the circular streamlines near the side walls. Re = 0 is the primary state, and it represents a three-di-
The radial pressure gradient, ap/dr, is balanced by the mensional flow as soon as it forms at Re = 0 + . The other
centrifugal acceleration pU’/r, where U is the circumfer- stable solution, the secondary mode, represents a second
ential velocity of the fluid. For a fixed pressure gradient, steady flow that destabilizes through a turning point at
velocity U becomes smaller as the fluid particles approach Re = RTP. The secondary state can exist only at
the end walls. Therefore to keep the balance, fluid particles Re > RTP and is replaced by the primary state for lower
must follow a circular path with a smaller radius r. values. The other cellular branches of the ideal system of
[B&Iewadt’s34 solution (see Schlichting35) of a semi-infi- Fig. 12 also decouple and result in additional locally stable
nitely long eddy contacting a stationary solid surface modes. W e propose that the patterns in Fig. 9 represent the
shows an overshoot in tangential velocity near the wall, decoupled secondary modes resulting from perturbations
which has also been verified experimentally and would ap- of the cellular patterns of the ideal system.
proximate the flow here.] To keep the mass balance, the The bifurcation sequence and the stability characteris-
fluid at the center of the primary eddy near the wall is then tics of flow in a LDC proposed above remain to be proved
forced away toward the center symmetry plane where it or modified through rigorous analysis. To study problems
flows radially outward and completes a loop. The three- that are a subclass of the LDC system (e.g., coating and
dimensionality of the flow is vividly apparent, even at Re as other surface application systems) requires a fundamental
low as 100, where the Davis and Mallinson36 numerical understanding of flow in a LDC. At this time, we base the
solution shows this general flow pattern. Also, extensive discussion of our experimental results on the above theory.
particle track results of Freitas and Street I9 and Iwatsu ef
al.” clearly show the three-dimensional character of the
B. Discussion of the experimental results
primary state.
To construct the bifurcation diagram for the 3-D cav- To discuss the experimental results, we assume that the
ity flow, we outline the following principles: ( 1) the pitch- throughflow serves only as a perturbation of the LDC; that
fork bifurcation diagrams of the ideal flow in Fig. 12 are is, we assume that all of the principles outlined in Sec.
not generic and therefore almost any perturbation of the IV A also apply here. Furthermore, we note that the bifur-
system will change their qualitative appearance; (2) there cation diagram of Fig. 14 is generic and therefore pertur-
exists a positive real number E, where, for Re < e, the so- bations of it do not necessarily result in qualitative
lution to the governing equations along with the no-slip changes.

2089 Phys. Fluids A, Vol. 3, No. 9, September 1991 Aidun, Triantafillopoulos, and Benson 2089
Given the above facts and assumptions as a frame- The globally unstable feature of throughflow LDC sys-
work, we can identify the sequence of states in Fig. 4 as tems explains a practical difficulty that has puzzled the
transitions in the primary mode. Figure 4(a) represents users of short-dwell coaters for some time. It is well known
the steady-state primary mode, which consists of primary that short-dwell coaters do not always behave the same
and secondary corner eddies. This state, which we refer to under identical operating conditions. The possibility of
interchangeably as either the primary state or the one-cell multiple flow states in the pond of these coaters may ex-
pattern, is the LDC flow, which has been studied in the plain their nonunique behavior.
past, Our flow visualization experiments (which include
high-speed cinemaphotography) show that this state loses v. CQNCLUSIQNS
stability at Re - 825 and gives rise to a time-periodic flow Multiple steady states have been observed in a lid-
[Fig. 4(b)], as shown by the bifurcation diagram in Fig. 15. driven, throughflow rectangular cavity. The primary mode
It appears that the disturbance structure is in the form of a (one-cell pattern) that forms at Re = 0 + , contrary to the
spiral vortex that superimposes on the base state and is previous belief, loses stability to finite-amplitude distur-
responsible for this transition. bances at finite Reynolds number and competes with at
Possible mechanisms of transition to time-periodic least three other secondary steady states (two-, three-, and
flow include a loss of existence with concomitant loss of four-cell patterns). At Re-825, the primary mode be-
stability or a loss of local stability of the steady solution comes locally unstable and gives rise to a time-periodic
(Hopf bifurcation). The latter mechanism can be a super- state. This time-periodic state destabilizes at Re- 1000.
critical or a subcritical transition. Although we show a In principle, if the addition of flow through an other-
supercritical bifurcation in Figs. 14 and 15, we note that wise confined LDC is only a perturbation of the closed
flow visualization experiments cannot accurately deter- system, then the generic nature of Fig. 15 suggests that
mine the character of this transition. We have observed multiple stable states may also exist in a closed LDC.
some hysteresis in our experimental observations of this
transition; however, we are not convinced whether this is
physical or due to human visualization error.
Although flow visualization methods used in these ex- Portions of this work were used by NGT as partial
periments cannot provide the detailed information re- fulfillment of the requirement for the Ph.D. degree at the
quired to establish the higher-order transitions, some in- Institute of Paper Science and Technology.
formation can be extracted through observations. At This research was supported by an industrial consor-
Re- 1000, the interface between the primary and the DSE tium through the Institute of Paper Science and Technol-
becomes wavy and the spiral-shaped vortex on the DSE ogy. Support for the equipment provided by Beloit Corpo-
appears to fold over itself and develop spikes [Fig. 4(c)]. ration is gratefully acknowledged.
At higher Re, the flow changes character in the region of
DSE, a clear signal of higher-order transitions. In the ‘C. K. Aidun and N. G. Triantafillopoulos, in International Symposium
on Mechanics of Thin-Film Coating, Spring National Meeting of the
neighborhood of Re - 2000, mushroom-shaped structures AIChE, 18-22 March, 1990.
form and move rapidly in alternating positions. These 2G. L, Booth and N. Millman, TAPPI Ser. No. 28, New York, 1965.
structures are almost identical to the cross-sectional view 3J K. Carpenter and P. I% Steen, in International Symposium on Me-
of Goertler vortices observed and recorded in boundary chanics of Thin-Film Coating, Spring National Meeting of the AIChE,
18-22 March, 1990.
layer flow over concave surfaces (for example, see Peerho- 4N. G. Triantafillopoulos and C. K. Aidun, TAPPI J. 73, 129 (1990).
ssaini and Wesfreid39). These vortices in LDC systems ‘F. R. Pranckh, and L, E. Striven, TAPPI J. 73, 163 (1990).
were first observed by Koseff and Street,r4 who focus on ‘B. G. Higgins, Dynamics of Coating, Adhesion and Wetting, Status Re-
the transition to turbulence that they show occuring at port Project 3328, The Institute of Paper Chemistry (now Institute of
Paper Science and Technology), 24 March 1982.
Re> 6000. These and other studies have focused only on ‘0. R. Burggraf, I, Fluid Mech. 24, 113 (1966).
the primary state of flow presented schematically in Fig. 3. ‘F. Pan and A. Acrivos, J. Fluid Mech. 28, 643 ( 1967).
In fact, to our knowledge, secondary stable modes have 9L. Prandtl, NACA Tech. Memo. 452, 1904.
“‘G. K. Batchelor, J. Fluid Mech. 1, 177 (1956).
never been reported in this system. “U. Ghia, K. N. Ghia, and C. T. Shin, J. Comput. Phys. 48, 387 (1982).
Our experiments show that in a LDC system with lzR. Schreiber and H. II. Keller, J. Comput. Phys. 49, 310 (1983).
EI/D = 1 and S/D = 3, and with a small amount of ‘>M. C. Thompson and I. H. Ferziger, J. Comput. Phys. $2, 94 ( 1989).
throughflow, the primary stable steady state [Fig. 4(a)], t4(a) J, R. Koseff and R. L. Street, J. Fluid Eng. 106, 21 (1984); (b) J.
R. Koseff and R. L. Street, J. Fluid Eng. 106, 385 (1984); (c) J, R.
which is the only state that exists at Reynolds number Koseff and R. L. Street, J. Fluid Eng. 106, 390 (19843.
close to zero, becomes globally unstable and competes with “A. K. Prasad and J. R. Koseff, Phys. Fluids A 1, 208 (1989).
at least three other secondary steady-state modes, each “V. Ya. Bogatyrev and A. V. Gorin, Fluid Mech. Sov. Res. 7, 101
having a unique and qualitatively different flow pattern, as ( 19781.
“J. Kim and P. Mom, J. Comput. Phys. 59, 308 (1985).
shown in Fig. 9. All these patterns remain symmetric with “C. J. Freitas, R. L. Street, A. N. Findikakis, and J. R. Koseff, Int, J.
respect to the midplane by forming two-, three-, and four- Num. Methods Fluids 5, 561 ( 1985).
cell structures. We identify these states with the decoupled ‘9C. J. Freitas and R. L. Street, Int. J. Num. Methods Fluids 8, 769
n-cell branches of the ideal system (Fig. 12) that may form (1988).
“C.-Y. Perng and R. L. Street, Int. J. Num. Methods Fluids 9, 341
and mutate in various ways. An example of one of the (1989).
possible arrangements is shown in Fig. 15. *‘R. Iwatsu, K. Ishii, T. Kawamura, K. Kuwahara, and J. M. Hyun,

2090 Phys. Fluids A, Vol. 3, No. 9, September 1991 Aidun, Triantafillopoulos, and Benson 2090
Fluid Dvn. Res. 5. 173 (1989). CP, 1988, p. 288.
“H. C. Ku, R. S. Hirsh,‘andT. D. Taylor, J. Comput. Phys. 70, 439 “5 D Bozeman and C. Dalton, J. Coniput. Phys. 12, 348 (1973).
(1987). 32M. Nallasamy and K. Krishna Prasad, J. Fluid Mech. 79, 391 ( 1977).
23M. D. Neary and K. D. Stephanoff, Phys. Fluids 30, 2936 ( 1987). 33C. K. Aidun, Bifurcation Phenomena in Thermal Processesand Con-
“N K Ghaddar, K. Z. Korczak, B. B. Mikic, and A. T. Patera, J. Fluid vection, edited by H. H. Bau, L. A. Bertram, and S. A. Korpela
t&h. 163, 99 (1986). (ASME-AMD, New York, 1987), Vol. 89, p. 31.
“C. H. Amon and A. T. Patera, Phys. Fluids A 1, 2006 (1989). 34U. T. Bodewadt, Z. Angew. Math. Mech. 20, 241 (1940).
26D. R. Carlson, S. E. Windall, and M. F. Peeters, J. Fluid Mech. 121, 35H. Schlichting, Boundary-Layer Theory, 7th ed. (McGraw-Hill, New
487 (1982). York, 1979), p. 225.
“G K. Batchelor, An Introduction to Fluid Dynamics (Cambridge U.P., 36G de Vahl Davis and G. D. Mallinson, Comput. Fluids 4, 29 ( 1976).
Nkw York, 1967). “J. ‘Serrin, Arch. Rat. Mech. Anal. 3, 1 (1959).
‘sH, S. Rhee, J. R. Koseff, and R. L. Street, Exp. Fluids 2, 57 (1984). 38T B. Benjamin, Math. Proc. Cambridge Philos. Sot. 79, 373 ( 1976).
29A. E. Perry and M. S. Chong, Annu. Rev. Fluid Mech. 19, 125 (1987). “H Peerhossainiand J. E. Wesfreid, in Propagation in Systems Far from
30A. K. Prasad, C. Y. Pemg, and J. R. Koseff, AIAA Paper No. 88-3654- E&librium (Springer-Verlag, New York, 1988).

2091 Phys. Fluids A, Vol. 3, No. 9, September 1991 Aidun, Triantafillopoulos, and Benson 2091

You might also like