You are on page 1of 10

The effect of small particles on fluid turbulence

boundary layer in air

in a flat-plate, turbulent

C. B. Rogers@ and J. K. Eaton


Department of Mechanical Engineering, Stanford University, Stanford, California, 943053030

(Received 11 April 1990; accepted 7 January 1991)


This paper describes the result of both an experimental and an analytical investigation of the
response of a two-dimensional, turbulent boundary layer in air to the presence of particles.
Copper shot, 70 ,um in diameter, were uniformly introduced into a vertical boundary layer, at a
momentum thickness Reynolds number of about 1000. The particle mass flux was set at 20%
of the fluid mass flux, and all measurements were made using a single-component, forwardscatter laser Doppler anemometer. The measurements clearly demonstrated that the particles
damped fluid turbulence, apparently affecting all scales equally. The measurements further
showed a strong correlation between the degree of damping and the particle concentration in
the log region of the boundary layer.

I. INTRODUCTION
Though common throughout industry, particle-laden
flows have received relatively little attention in the laboratory until the past decade. The two-phase environment has
been traditionally hostile to the presence of even,simple velocity probes and has therefore hindered accurate velocity
measurements. The advent of the laser Doppler anemometer
allowed two-phase data to be taken with comparative ease
and has greatly enhanced the volume of experimental data.
Even with these new data, a coherent understanding of turbulence attenuation by particles allowing accurate flow prediction has not emerged.
Turbulence attenuation (i.e., the reduction of the gasphase turbulent stress levels) has been very difficult to investigate experimentally. One problem is the requirement to
measure the gas-phase velocity statistics in the presence of at
least moderate particle concentration. Another problem is
that nonuniform particle loading will modify the fluid mean
flow. This can in turn modify the turbulence level, an effect
that cannot be attributed to particle/turbulence interaction.
Because of these difficulties, there are only two flows, fully
developed pipe flow and axisymmetric jet flow, where turbulence modification has been extensively studied. It is the objective of this work to extend the understanding to another
simple shear flow, the flat-plate boundary layer.
One of the problems encountered in investigating particle-laden flows is the large parameter space. An incompressible, single-phase flow can be defined by a flow Reynolds
number and geometric parameters. However, the addition of
a particulate phase adds several new parameters. In particular, the particle Reynolds number, particle time constant,
particle gravitational drift velocity, particle mass loading,
particle loading distribution, particle diameter, and particleto-fluid density ratio may all play an important role in determining both particle and fluid behavior. An experiment can
only fill a small niche in the parameter space, but possibly
will illuminate the flow behavior over a wider range of parameters. In the following paragraphs, we define where the
a) Present address: Department of Mechanical Engineering, Tufts University, Medford, Massachusetts 02155.
928

Phys. Fluids A 3 (5). May 1991

present experiments fit within the parameter space. We also


examine the presumed effects of varying each parameter in
hopes of extending the usefulness of the data.
The particle Reynolds number is defined as
Rep = Ureld,/vp

(1)

where Cr,, is the particle relative velocity ( V - U), dP is the


particle diameter, and vf is the fluid kinematic viscosity. For
the remainder of this paper, we will use the customary notation of U for the fluid velocity and V for the particle velocity,
and the subscripts f and p will refer to fluid and particle
properties, respectively. This Reynolds number characterizes the flow around the particle; small Reynolds numbers
correspond to attached laminar flow and large Reynolds
numbers corresponding to fully turbulent particle wakes.
Particles with small Reynolds numbers are expected to act
only as point sources of momentum exchange between the
phases while large particles with high Reynolds numbers are
expected to increase turbulence levels by producing turbulent wakes. The particles in the present study had a particle
Reynolds number ranging from 4 to 6. Therefore, the results
are expected to be representative of the broad class of problems where the particles act only as point sources of momentum exchange. Stokes showed that for Reynolds numbers
less than 0.1, the particle drag is linearly related to the relative velocity as
CD = 24/Re,,
(2)
where C, is the particle drag coefficient. While the present
particles had larger Reynolds numbers, Rogers and Eaton
showed that the assumption of linear drag was adequate to
model the response of the particles to a turbulent flow.
The particle time constant, or particle relaxation time,
characterizes particle inertia. For particles with Stokes drag,
the time constant rp is defined as
7;, = ppd ;/18/q,
(3)
where pp is the particle density and ,u~ is the fluid viscosity.
Rogers and Eaton showed that using Eq. (3) for particles
with Reynolds numbers greater than 0.1, however, leads to
overestimation of the particle drag. Therefore, the time con-

0899-8213/91/050928-10$02.00

@I 1991 American Institute of Physics

928

Downloaded 11 Apr 2005 to 132.77.4.129. Redistribution subject to AIP license or copyright, see http://pof.aip.org/pof/copyright.jsp

stant for these non-Stokesian particles can be estimated from


the measured particle drag. Using Stokes formulation of the
particle transport equation,

dV,- 1 v,, - g>


dt -c
and evaluating it at the particle terminal relative velocity
U,, one can estimate a corrected time constant to be
9;, = u,/g.

(5)
Conventionally, the time constant is presented nondimensionally as the Stokes number which is defined as the
ratio of the particle to fluid time scale. Particles with small
Stokes numbers ( < 0.01) will follow the flow exactly and
will not affect the turbulence except to the degree that they
modify the fluid properties. Particles with large Stokes
numbers ( > 100) will not respond significantly to turbulent
velocity fluctuations. The particles in the present experiment
had Stokes numbers between 1 and 10, implying that the
particles would follow some, but not all, of the fluid turbulence. Therefore, this experiment is representative of a broad
class of situations where the particles respond to turbulent
fluctuations but there is a significant relative velocity
between the particles and the flow. Unfortunately, it cannot
be determined a priori how variation of the Stokes number
may change turbulence attenuation.
Gravitational force, too, keeps particles from exactly
following the path of a fluid point. Combined with the inertial effects described above, these forces will pull the particle
through a series of different fluid neighborhoods, making it
difficult to predict the behavior of the fluid surroundings in
the particle reference frame. This effect is called the crossing-trajectories effect and was first reported by Yudines3
The crossing trajectories effect acts to reduce the fluctuation
levels of the particles.4 The related continuity effect5 reduces the particle fluctuations even more in directions normal to the direction of the gravitational drift. Possible dimensional parameters describing this effect are the ratio of
the particle terminal velocity to the mean free-stream velocity or a characteristic turbulent velocity. In the present case,
the ratio of the particle terminal velocity to the free-stream
velocity is about 10%. More significant, perhaps, is the ratio
of the particle terminal velocity to the fluid streamwise turbulence, u,/m),
which varies across the boundary layer but is close to 1 at the position of peak turbulenceintensity. The presumed effects of varying this ratio are described in
the analysis section.
The particle mass loading, that is the ratio of the total
mass flux of particles to the mass flux of the fluid, characterizes the influence of the particles on their surroundings. Low
mass loadings imply that the total particle drag is small compared to other forces involved and therefore the fluid behavior remains unaffected by the particle presence. The results
presented here were at a mass loading of about 20%. Based
on previous experiments, such loadings should have a minor
effect on the mean flow behavior. However, as will be shown
below they do have a significant effect on the turbulence. For
moderate particle loadings, we may guess that the particle
loading affects the turbulence linearly; that is if a 10% parti-

cle loading causes a 5.% turbulence attenuation, a 20% loading would cause a 10% attenuation. Such a linear relation
would probably not hold if the attenuation became large or if
the particle concentration became so large that particle-toparticle interactions were significant.
Nonuniform particle loading will cause mean flow variations leading to changes in the turbulence. A dramatic example would be if all the particles collected in the center of
an upward pipe flow. The mean fluid velocity would thus be
retarded in the pipe center making a dip in the velocity profile. The velocity gradient would serve as a source of turbulence production. A key objective of the present experiment
was to have a uniform particle loading. Nonuniformities are
treated as sources of uncertainty. It is assumed that calculation of mean-velocity variations and turbulence modification produced only by nonuniform loading can be estimated
analytically.
Variations in particle diameter, with respect to the fluid
length scale, will also affect the particle and flow behavior. If
the particle diameter is substantially larger than the Kolmogorov length scale of the surrounding fluid, then the simple
presence of the particle will affect the energy distribution of
the 0ow. Turbulent eddies will be locally strained in the vicinity of each particle. In addition, the particle will be exposed to a turbulent rather than laminar approach flow. The
particles used in this experiment were smaller than the Kolmogorov length scales of the flow. Therefore, the present
results should be representative of all cases where individual
particles do not interact directly with turbulent eddies.
Last, the ratio of the particle-to-fluid density is important in determining which forces on the particle are important and which are not. The particle transport equation
above [ Eq. (4) ] neglects a large number of forces. The full
transport equation would be the one first put forth by Tchen,
with the addition of acceleration forces, the Magnus force,
spin forces, the Saffman lift force, and electrical forces. For
particles with low Reynolds numbers and densities much
greater than the fluid, one can show that the affect of these
forces is small and therefore Eq. (4) closely approximates
the particle behavior.
Once having defined our parameter space, the objective
of this research was to examine the interaction between the
fluid turbulence and the particle motion in fluid shear. This
interaction was isolated from changes in the fluid turbulence
due to variations in the fluid mean behavior. For the remainder of this paper, we will refer to fluid turbulence simply as
turbulence, since particle turbulence is an unrealistic concept for these low-volume loadings. Using these results, we
examine the concept of modeling this interaction as an increased viscosity.
II. RELATED EXPERIMENTAL

WORK

The problem of turbulence modification by dispersed


particles is receiving increasing attention in recent years.6
The majority of the experimental work has concentrated on
two flows, the axisymmetric jet and the axisymmetric pipe
flow. Gore and Crowe recently reviewed the available experimental data for these flows. They identified the ratio of
the particle diameter to a characteristic length scale of the

929
Phys. Fluids A, Vol. 3, No. 5, May 1991
C. B. Rogers and J. K. Eaton
929
Downloaded 11 Apr 2005 to 132.77.4.129. Redistribution subject to AIP license or copyright, see http://pof.aip.org/pof/copyright.jsp

turbulence (d,/l, ) as a key parameter and found that turbulence was attenuated for small values of d/Z, and amplified
for large values. The critical value of d,/l, separating regions of attenuation and amplification was of the order of
0.1. This critical value varied with radius innthe pipe flows
but not in jet flows. It is surprising that this simple correlation works at all since the particle diameter-to-length scale
ratio is just one of the many relevant parameters of the problem. The Gore and Crowe correlation cannot be used to predict the level of turbulence modification. However, it is clear
that the present experiment is well within the size range expected to cause turbulence attenuation.
In the remainder of this paper, we will concentrate only
on the pipe flow studies since they are directly relevant to the
present work on wall-bounded turbulent flows. The pipe
flow studies have used a variety of fluid and particle types the
most relevant ones being those examining solid particles dispersed in gas flows. In many cases, the particles were introduced nonuniformly, causing mean-velocity variations that
overwhelmed the direct effects of particles on turbulence. As
a result, there is little uniformity in the data and no generally
accepted correlation describing turbulence modulation by
particles.
Key papers describing early work in vertical pipe flows
are those by Soo et a1.,8 Doig and Roper, and Reddy and
Pei. lo Reddy and Pei also provided an extensive review of all
of the work prior to 1969. Most of the experiments examined
vertical pipe flows with spherical particles at mass loadings
ranging up to about 5. Typical instruments used were special
pitot probes to measure the mean gas velocity and photographic techniques to measure particle velocities. Most of
the results showed that the mean gas velocity profiles were
flattened by the presence of particles for loadings above
about 1. The particle concentration profile was uniform in
some experiments while other experiments showed a higher
concentration near the wall. Soo (quoted in Reddy and
Pei) reasoned that nonuniform distribution was caused by
electrostatic effects. Direct measurements of the gas-phase
turbulence were impossible, but Soo et al. inferred the gasphase turbulence behavior based on helium diffusion measurements. Their results, valid only near the centerline of the
pipe, showed that the intensity of the turbulence was not
a&cted by the particles, but the Lagrangian integral scale
was decreased.
Boothroyd and Walton12 examined the displacement of
CO, -He tracers in the presence of 0-40pm zinc powder in a
vertical pipe flow of air. Using a Kathometer to measure
species concentration they found that for a mass loading of
300%, the particles suppressed turbulent diffusion, especially near the wall. More recent investigations of particle-laden
pipe flows have used modified laser anemometers to make
measurements of both the particle and gas-phase velocities.
In one of the earliest such studies, Carlson and Pesl&3
measured particle velocities in a 7.62 cm square duct. The
flow was laden with either 44- or 2 14pm glass beads at mass
loadings less than 1. They found that the bulk of particle
velocity fluctuations could be attributed to particle size variations, suggesting that such variation may be responsible for
increases in gas-phase turbulence when the mass loading is
930

Phys. Fluids A, Vol. 3, No. 5, May 1QQl

substantial.
Lee and Durst14 made axial velocity measurements in a
2 cm diam pipe laden with 100, 200, 400, or 800 pm glass
beads. Mass loadings increased from about 1 for the smallest
particles to about 2.5 for the largest. They found that the
largest particles had a nearly uniform velocity profile across
the pipe and caused a flattening of the gas velocity profile
and increases in the turbulence intensity. They also found
that smaller particles damped the turbulence. Finally, they
documented particle mean velocities exceeding those of the
surrounding flow near the wall, implying that perhaps the
particles were still accelerating. They proposed a simple
model in which the particles either totally ignore the fluid
turbulence, or follow it exactly. Dividing their flow into
these two regimes, they used their model to predict their
results with a fair degree-of accuracy. :
Arnason and Stock injected 57 ym glass beads on the
centerline of a vertically downward pipe flow. They noticed
an increase in the flow turbulence as they increased the particle loading. However, the nonuniform particle distribution
would create a distorted mean-velocity profile which may
have been responsible for the increase in the turbulence.
The most complete experiment for particle-laden pipe
flows was performed by Tsuji et al. and examined the flow in
both horizontal6 and vertical pipes. They examined the
fluid turbulence in the presence of particles ranging from 200
to 30qOpm and for mass loadings up to 610%. They found
that the small particles tended to suppress the iluid turbulence whereas the large particles enhanced the fluid turbulence, The. intermediate particles enhanced the fluid turbulence near the pipe centerline and reduced the fluid
turbulence near the wall. They also showed that the large
particles added broadband energy, increasing the energy
content at all frequencies equally, whereas the small particles increased the high-frequency energy content of the ffow
and decreased the low-frequency- content.
Analytical and numerical studies of turbulence attenuation have been restricted by a lack of fundamental understanding of particle-turbulence interactions. Elghobashi and
co-workersi9 modified the transport equations for the kinetic energy and the dissipation rate in the k-epsilon model
to account for damping of turbulence by particles, The model worked well for the jet flow but may be quite difficult to
generalize to different flows. Leems2and Lee and Borne?
developed a model based on theory and empirical input from
the pipe flow experiments of Lee and Durst and Tsuji and
Morikawa. The model divided the flow into regions where
the dominant flow frequency was above or below the particles cutoff frequency. In the low-frequency region, particles were assumed to diffuse like 0uid points while in highfrequency regions, they respond only to the mean-velocity
field. Another element of the model was an empirical correlation for an effective turbulent viscosity seen by the particles. The model represented the pipe flow data well but cannot be generalized to other flow geometries easily. In
particular, Rogers and Eaton showed that the model did not
.correctly predict the present data.
In a sister study to the present work, Squires and Eaton23*24mhave
used direct numerical simulation of the incomC. B. Rogers and J. K. Eaton

930

Downloaded 11 Apr 2005 to 132.77.4.129. Redistribution subject to AIP license or copyright, see http://pof.aip.org/pof/copyright.jsp

pressible Navier-Stokes equations coupled with Lagrangian


tracking of up to one million particles to study turbulence
damping in homogeneous, isotropic turbulence. The particle
Stokes number ranged from 0.14 to 1.5 for the computations.
Their work was restricted to relatively low-turbulence Reynolds numbers, but showed significant turbulence damping
for a particle mass loading of 10%. At 100% particle mass
loading the turbulence kinetic energy was reduced by nearly
one-half.
III. EXPERIMENTAL

1.0 CIll

1.3 cm

Floi

SETUP

In order to examine a particle-laden boundary layer, we


built a low-turbulence air boundary layer wind tunnel with a
unique particle feed mechanism for uniform particle loading. Figure 1 is a schematic of the tunnel itself, showing the
vertical orientation of the test section, minimizing the effect
of gravity on the local particle concentrations. Air was
drawn through a filter into a seeding chamber where talcum
powder Aow tracer was dispersed using an air brush. The
talcum powder was suspended inalcohol which evaporated
before the seed reached the test section. The flow was drawn
into a centrifugal blower then passed through a diffuser, a
flow conditioning section, and a turn to the vertical orientation.
The particle phase was added downstream of the flow
conditioning using a mechanism designed for optimal uniforrnity.and steadiness of the mass loading. The particles
were dispersed into the flow by 1.3 cm tall, 1.0 cm diam
buckets (see Fig. 2)) having a 1.O cm opening on the top and
a 0.15 cm hole in the bottom. This geometry allowed the
buckets to fill rapidly on either side of the wind tunnel before
slowly draining their contents as they crossed the flow.
These buckets were traversed across the wind tunnel by two
sets of belts. Since the draining rate of the particles out of the
bucket remained constant as the bucket was drawn across
the tunnel, the particle flux proved to be extremely uniform
under each belt. Grids downstream of the feeder enhanced
the lateral mixing of the particles. Two hundred buckets
were used to get the required mass loading rate of 20%. The

Bucket Cross-section:

EXHAUST DUCT

Direction

FIG. 2. Particle feeder schematic.

loading uniformity was ensured through visual checks using


a laser sheet.
The flow then passed through a 3:1 area ratio, two-dimensional contraction. A section of honeycomb was installed at the exit of the contraction to eliminate the small
normal velocity of the particles. In the absence of the honeycomb, the contraction curvature accelerated the particles toward the tunnel center. With the honeycomb in place, the
concentration was uniform at the test section entrance. The
free-stream turbulence level (of 1%) was higher than desired but apparently had little effect on the boundary-layer
development.
The boundary layer developed in a 7.6 by 46 cm rectangular duct that was 114 cm long. A 0.8 mm high by 13 mm
long rectangular boundary-layer trip was mounted at the
test section entrance to ensure a uniform, two-dimensional
transition in the relatively low Reynolds number boundary
layer.
The particles used in this experiment were nominally 70
pm diam spherical copper beads with a material density of
8800 kg[m3. They were commercially sized classified to a
t 5 ,um window to limit,the effects due to particle diameter
variations. Microscopic examination showed that most of
the particles were nearly spherical with some particles more
elongated. The actual size distribution and the corresponding aerodynamic time constants are shown in Fig. 3. These
data were measured using a Coulter counter. An actual photograph of the particles is shown in Fig. 4.
Velocity statistics were obtained using a single-compo-

CYCLONE

PARTICLE

HOPPERS

PARTICLE

FEED

B5

0
40

a0

60

Diameter [pm]
FIG. 1. Wind tunnel schematic.
931

Phys. Fluids A, Vol. 3, No. 5, May 1991

FIG. 3. Diameter distribution of particles.


C. B. Rogers and J. K. Eaton

Downloaded 11 Apr 2005 to 132.77.4.129. Redistribution subject to AIP license or copyright, see http://pof.aip.org/pof/copyright.jsp

931

FIG. 4. Photograph of the particles.

nent, forward-scatter laser Doppler anemometer which was


used to measure the gas flow without particles (i.e., unladen
flow), the gas flow in the presence of particles (i.e., laden
flow), and the particles themselves. Measurements of the gas
phase required flow tracer particles that were talcum powder with a nominal diameter of 1 pm. Measured velocity
statistics for the unladen flow agreed to those taken by a hot
wire to within l%, implying that the tracers followed the
flow exactly for the velocities used in this experiment.
The local density approximation LDA system used a
low-power, helium-neon laser, TSI transmitting optics, and
on-axis forward-scattering, collection optics, with a TSI
model 1980B counter-type signal processor. Table I shows
the specific LDA parameters used for thedata-taking process. The key parameters are a measuring volume diameter
of approximately 0.5 mm and no Bragg shifting .used for
axial velocity measurements.
The LDA transmitting and receiving optics were rigidly
coupled on a computer-controlled traverse system. The measuring position was read using an Accurite miniscale linear
encoder with better than 13 ,om resolution. The traverse was
controlled by an IBM PC which also collected the data digitally from the LDA signal processor and controlled the wind
tunnel speed.
Discrimination between gas-phase tracers and the copper beads was done on the basis of signal amplitude. This
method was effective due to the distinct size gap between the

TABLE I. Operating conditions for LDA measurements.


Laser settings

Counter settings

Laser beam Wavelength


Diameter
Power
Measuring
Beam half-angle
volume
Diameter
Fringe spacing
Number of fringes
Streamwise Number of cycles
velocities
Shift frequency
Comparison setting
Exponent setting
Filter settings
Data transfer

largest tracer particles and the smallest copper particles. To


measure the particle phase, the preamplifier gain on the
counter processor was reduced until flow tracers could not
be detected. Therefore, there was no contamination of particle-phase measurements. To measure the gas-phase in the
presence of copper particles the preamplifier gain was increased and an amplitude discriminator circuit used. Highamplitude signals were not processed. Cross talk can occur
in such a system when a large particle grazes the measuring
volume producing a small signal amplitude. This effect is
minimized by requiring a relatively large number of fringe
crossings for a vali-d measurement. In adjusting the system,
we relied on the fact that the mean-velocity difference
between the particles and gas in the free stream was larger
than the standard deviation of the velocity. Therefore, cross
talk was easily detected. Extensive qualification experiments
are reported in Rogers and Eaton. These experiments
showed that the effect of cross talk was minimal for particle
mass loadings up to 20%.
Fluid power spectra were measured using the method of
Gaster and Roberts. Due to the random arrival of a velocity measurement (i.e., a tracer particle), one cannot use the
conventional method of performing a fast Fourier transform
on incoming data if the data rate is close to the frequencies
present in the flow. Therefore, we performed a direct Fourier
transform on autocorrelations formed from 60 000 instantaneous velocity measurements, separated into 1000 time bins,
each 0.0005 set wide. Figure 5 favorably compares the results of this method with conventionally processed, hot-wire
measurements.
.
The experimental uncertainty m the mean-velocity statistics presented in this paper is close to 6%. Measurement
uncertainty arose from inaccurate phase discrimination,
slight loading nonuniformity, small signal-to-noise ratio,
and from limitations inherent in finite data records. The
mass loading of 20% used in this experiment was the maximum loading obtainable before the increase in background
noise due to the copper particles became too large for accurate data validation.
Finally, the problem of particle loading uniformity becomes somewhat more complex in a shear layer. Although
both the particle flux and particle concentration were uniform across the test section entrance, as the boundary layer
grew, the particle flux initially remained constant, therefore

632.8 nm (red)
480.5 mm
4mW
2.98
0.48 mm
6.1 pm
80
16
OHz
1%
Automatic
0.3-3 MHz
Digital IO
FIG. 5. Spectral check: power spectrum at X = 55 cm and y+ = 300.

932

Phys. Fluids A, Vol. 3, No. 5, May 1991

C. B. Rogers and J. K. Eaton

932

Downloaded 11 Apr 2005 to 132.77.4.129. Redistribution subject to AIP license or copyright, see http://pof.aip.org/pof/copyright.jsp

increasing the local particle concentration in the low-velocity regions. The particle concentration eventually became
uniform across the boundary layer farther downstream due
to turbulent mixing. Thus initially in the boundary layer the
particle flux is constant, and farther downstream the particle
concentration approaches uniformity. Figure 6 compares
constant particle flux to constant particle concentration
drag profiles of a cloud of particles as a function of position
in the boundary layer. These curves result from simple arguments assuming that the particles were at their terminal velocity and that the total drag force of the particles is simply
the product of the particle concentration and the weight of
an individual particle. Therefore, for the case of uniform
particle concentration, the total drag of the particles is uniform across the boundary layer and for uniform particle
flux, the drag varies like the inverse of the local particle velocity. At the measuring locations presented in this paper,
the particle concentrations were close to uniform, as evidenced by the similarity between the laser anemometer data
rate profile and the mean-velocity profile.

TABLE II. Fluid parameters (dissipation from Murlis et ol., Ref. 28:
Re, = 1089).
Flow parameters

X = 55 cm

X = 85 cm

IV. RESULTS
Boundary-layer profiles were measured at two streamwise locations: 55 and 85 cm downstream from the trip. Each
profile consisted of 25 measurement points and each point
was the result of 4000 individual velocity samples. Integral
boundary-layer parameters are compiled in Table II. In order to establish that the boundary layer measured in this
experiment is similar to the others appearing in the literature, we compared our velocity profiles with those of Purtell
and Klebanoe6 and our velocity spectra with those taken
by Johnson and Johnston.7 Figure 7 compares the mean
fluid velocity profiles, showing excellent agreement between
our data and that of Purtell and Klebanoff. The streamwise
turbulence intensity (Fig. 8) measured in our boundary layer agrees well with the data of Purtell and Klebanoff in the
boundary layer but the effect of the honeycomb at the contraction exit can be seen in the higher free-stream turbulence
intensity values measured in our tunnel. The comparison of
velocity spectral data between our boundary layer and that
of Johnson and Johnston (Fig. 9) shows that the turbulence
in the boundary layer is characteristic of a-low Reynolds
number turbulent boundary layer.
Figures 10 and 11 show the laden and unladen fluid

Flow scales

Parficle Flux
Partlcle Concentration
3

y/S*
FIG. 6. Drag variation across the boundary layer.

10

8.0 m/set
3.0 m m
2.1 m m
1.4
1550
1090
0.38 m/set
0.0022
24 m m
8.2 trdsec
3.8 m m
2.6 m m
1.4
2020
1410
0.37 m/set
0.0020
5.0 m/set
0.16 m m
7.3 m m
3.8 m m

0.9

0.8

Dissipation
Kolmogorov
leligth scale
Integral scale
(streamwise)
Integral scale
(normal)

20 m m

mean-velocity profile at both X locations downstream of the


boundary-layer trip. The tunnel was readjusted in the case of
the laden flow to account for the bulk drag of all the particles
in the tunnel. That is, the free-stream fluid velocity at the
tunnel inlet was set to be the same in both the laden and
unladen flows. Since the boundary layers at the inlet are thin,
this is equivalent to holding the mass flow of air constant.
Due to the high experimental uncertainty, it is difficult to
ascertain whether the particles actually caused a change in
the shape of the fluid mean-velocity profile. Close examina-

-Uniform
-.-_-Uniform

X= 55 cm,
yf =300

Boundary-layer
thickness (S,, )
Free-stream
velocity
Displacement
thickness (a*)
Momentum
thickness (6)
Shape factor (H)
Rep
R%
UT
Skin friction coefficient ( Cf/2)
Boundary-layer
thickness (S,, )
Free-stream
velocity
Displacement
thickness (S*)
Momentum
thickness (0)
Shape factor (H)
Re,
Ree
cr,
Skin friction coefficient (C/2)

__
0

Purtall Data
Present Experiment

0.7

12

YIP
FIG. 7. Qualification: mean-velocity profile comparison.

933
933
C. 6. Rogers and J. K. Eaton
Phys. Fluids A, Vol. 3, No. 5, May 1991
Downloaded 11 Apr 2005 to 132.77.4.129. Redistribution subject to AIP license or copyright, see http://pof.aip.org/pof/copyright.jsp

FIG. 8. Qualification: fluctuating velocity profile comparison.

FIG. 10. Mean-velocity profiles: X = 55 cm.

tion of the local probability density function where the


curves appear to deviate suggests that the difference results
from experimental uncertainty. If the presence of the particles changes the turbulent shear stress, then the mean-velocity profile must also change. Such changes, however, develop slowly and would probably not be apparent in the
relatively short test section.
Since changes in the mean flow velocity statistics appear
to be minor, any decrease in fluid turbulence is most likely
due to a turbulence-particle interaction. The comparison of
the laden and unladen flow streamwise turbulence intensities across the boundary layer at X = 55 cm [Fig. 12) shows
no evidence of turbulence-particle interaction, whereas the
downstream velocity profile at the X = 85 cm location (Fig.
13 ) shows up to a 35% decrease in the measured fluid turbulence intensities. This substantial damping of the fluid turbulence cannot result solely from the high measurement uncertainty because the possible sources of uncertainty would
tend to increase rather than decrease the measured turbulence fluctuations. The increase in the measured free-stream
turbulence, however, does fall within the range of this measurement uncertainty.
The turbulence/particle interaction seen in the present
experiment had many similarities to that documented by
Tsuji et al. for a vertical fully developed pipe flow. Their data
show the same trend of larger turbulence suppression by the
particles in the end of the log region and the beginning of the
wake region of the boundary layer. This uneven suppression

is most likely a result of the uneven drag loading discussed


previously and possibly variations in the particle initial conditions, fluid length scales, and particle/wall interactions.
The fluid power spectra of both the present study and the
work of Tsuji et al., too, showed similar behavior. Figures 14
and 15 show the fluid power spectra at two positions in the
boundary layer at theX = 85 cm location. Both experiments
clearly showed particle-induced turbulence attenuation of
the low-frequency content and an increase in the high-frequency content of the streamwise component of the fluid
energy.
The increase in the high-frequency content of the flow is
most likely not a real phenomenon but rather a measurement
phenomenon. One would expect to measure energy at frequencies corresponding to the reciprocal of the time between
particles crossing the measuring volume; about IO3 Hz. Most
of the fluid velocity readings measured in the vicinity of the
particle would not be validated since both the particle and
the tracer would be scattering light simultaneously and the
resulting beat frequency would inhibit signal validation.
Cases would exist, however, where the particle would be
outside the measuring volume and the tracer would be inside. It is these cases that most likely increase the high-frequency content of the measured fluid energy. Furthermore,
any contamination of the fluid signal by particles would
show up in the high frequencies. Tsuji et al. noticed that as
the particle concentration was increased, this high-frequency content correspondingly increased, possibly a result of an

0.08

Bk5
10-l7
sz
I
E
-

0.06

4.J

0.04

$
-Johnson
- - -Current

Data
Experiment

lo?
0.02

10

12

y/S

FIG. 9. Qualification: power spectra comparison.


934

Phys. Fluids A, Vol. 3, No. 5, May 1991

FIG. 11. Fluctuating velocity profiles: X= 55 cm.


C. EL Rogers and J. K. Eaton

934

Downloaded 11 Apr 2005 to 132.77.4.129. Redistribution subject to AIP license or copyright, see http://pof.aip.org/pof/copyright.jsp

1.0

3.

0.0

0.8

--

0.7 ~
I-

i.

Ll-LJ-L-0

d
2

10 .~.".. 102
_=

1o.J--

1 1 1 1 1 1 I 1 1 I
8
10
12

1 1 1.111

-----JJ

f [Hz]

y/S'

103

.-

FIG. 12. Mean-velocity profiles: A= 85 cm.

FIG. 14. Flow power spectra: X = 85 cm, y + .= 100.

increased number of particles skirting the measuring volume


with a tracer inside. Full numerical simulations. (e.g.,
Squires and Eaton) that resolve all scales of fluid turbulence
but do not resolve the local perturbations of the flow around
the particles, do not measure any increase in the high-frequency content of the fluid power spectrum. In fact, they
measured broadband attenuation across all frequencies due
to the particles.

The Reynolds decomposition, in which all quantities are


expressed as the sum of mean and fluctuating parts, is applied to develop the transport equation for the turbulent kinetic energy. W e assume that the flow is homogeneous so we
can neglect transport and diffusion terms that will complicate the equation and obscure the. turbulence damping
terms. Since the particles will only indirectly affect these
terms; by varying the fluid velocity, the bulk of.the particle/
turbulence interaction will appear in an additional term in
the transport equation for kinetic energy. Because the flow is
incompressible and the particle volume is negligible, the fluid density is constant. The transport equation for the turbulent kinetic energy is then

._
V. ANALYTICAL

.-

EXAMINATION

The momentum equation for an incompressible fluid


bearing fine solid particles is
?f!L + upi,, = - -LpJ + vfui,ii - LFi,
at
Pf
._ ff
where indicial notation is usedand commas imply derivatives. Fi in this equation is the instantaneous drag force per
unit volume applied by the particles to the fluid. Assuming
that the particles occupy a negligible volume fraction and do
not react with the fluid (i.e., the present case), the continuity
equation for the fluid remains unchanged by the addition of
particles. Further limiting the scope of this analysis to particles obeying a linear drag law, the drag force can be:expressed as
F, = tWrp)(ut
-oil,
(7)
where Q, is the local particle concentration (mass of particles
per unit volume). Note that this assumption does not require
the use of the Stokes drag coefficient.

B q
-irtT=P-E- $-[.@ (&m,.

rP

+ (Q --Fi) q
___
~
+ ( wu;u; - wu;q,

.L.
(8)
where the turbulence production term, P and the viscous
dissipation term E are identical to the production and dissipation terms in a single-phase flow. This equation can be
further simplified for the conditions of the present experiment. Using direct numerical simulation, Squires and Eaton
showed that W is very small for Stokes numbers greater than
one. Therefore the triple correlations terms are negligible
compared to the other terms. Also, from the data in the present experiments, fi, - 7, is about the same as ( z) ,
n2 - p2 is very small, and g3 - F3, is identically zero (for
1,

O.OT I ai I I hI 1A
0.08

8
$
3

0 Unladen Flow ($=O)


0 Laden f%w (cpJ.2)

-----.--LB
2
?:

0.06

g
a

0.04

lo.3 ?
I

---Unladen
F6.Q (t$=O.O)
-- Laden Flow ($=O.i?)

I.
0.02

_.
lo.4

0
0

10

12

10

YIP

FIG. 13. Fluctuating velocity profiles: X = 85 cm.

1
102

103

f WI

FIG. 15. Flow power spectra: X=

85 cm, .J = 300.

935
C. El. Rogers and J. K. Eaton
935
Phys. Fluids A, Vol. 3, No. 5, May 1991
Downloaded 11 Apr 2005 to 132.77.4.129. Redistribution subject to AIP license or copyright, see http://pof.aip.org/pof/copyright.jsp

the purposes of discussion, we took X, as the streamwise


coordinate, X, as the normal coordinate, and X, as the spanwise coordinate). Therefore the term ( ni - F{ ) @ u; is of
the same order as the triple correlation terms and can be
ignored. The appropriate transport equation for the turbulent kinetic energy in the present conditions is then
DC
:y=P-e-ut 2

'"r(&y

, I -

-7-T
UiVi).

(9)

PfTP

This equation is identical. to the single-phase equation with


the addition of the last term due to particle drag. It should be
noted that the form of the viscous dissipation term, E is identical to the equivalent term in the single-phase equation. The
transport equation for dissipation, however, will be affected
by the particles as it is a function of the fluid turbulence. Any
change in the fluid turbulence due to the presence of the
particles will affect the amount of viscous dissipation. Under
the present conditions, the greatest change in the fluid turbulence is of the order of 10% and therefore the major cause of
turbulence modification will be the drag term in Eq. (9)
because it appears directly in the transport equation of the
turbulent kinetic energy.
The modification term can be expressed alternately as
(@/pf~p)[

- RBi(

-$)""(

$,"'I,

(10)

where R,,, is the correlation coefficient between fluid and


particle velocity fluctuations. In stationary, homogeneous
turbulence, ( e) 12 must be less than or equal to ( z) 12.
Therefore, particles that obey a linear drag law and do not
produce turbulent wakes must cause turbulence attenuation
in such flows. In inhomogeneous flows, however, the particle velocity fluctuation levels could possibly exceed the fluid
turbulence intensity (e.g., resulting from wall interactions).
In the present case, the streamwise particle fluctuation
levels are nearly the same as the streamwise turbulence intensity. This does not imply that the fluid and particle velocities are perfectly correlated but we might expect that u; u;
is a large fraction of u; u; . Previous measurements of a more
lightly laden flow in the same facility (Rogers and Eaton)
showed that the normal particle velocity fluctuations are
much less than the corresponding fluid velocity fluctuations
so the terms u; u; and u; v; must be small relative to -u; u;
and u; u;, respectively. Thus U;LJ; is less than ufu; and
therefore the particle-associated term in Eq. (9) should lead
to turbulence damping.
In order to be more quantitative, the fluid/particle velocity correlation ulvf must be estimated. Following the
discussion of the previous paragraph, we can guess that uiv;
is between 20% and 50% of U;ul. Evaluating the production and turbulence kinetic energy using the data of Murlis et
ai. for a similar single-phase boundary layer at y/S = 0.4
yields a production level of 11 m/sec and a particle damping term between 1.2 and 0.8 mz/sec3. In fact, if the particles
are unwrrelated with the fluid, the damping term would be
1.5 m2/sec3, or 14% of the production. Thus the particle
damping term is on the order of 10% of the production term,
corresponding roughly to the observed level of turbulence
damping.
936

Phys. Fluids A, Vol. 3, No. 5, May 1991

Using the analysis above and the present data we can


estimate the effects of varying the particle parameters on
turbulence attenuation. First we consider varying the particle terminal velocity independent.of other parameters. This
could be done, for example, by varying the particle material
density. Rogers and Eaton showed that particle response to
turbulence fluctuations is controlled by the ratio of the particle relative velocity to the gas velocity. As the terminal velocity increases, the spectrum of turbulence fluctuations observed by the particles shifts to a higher frequency. This
decreases the particle/fluid
velocity correlation and increases the turbulence attenuation.
The analysis is not so simple when the particle time constant is varied, holding the terminal velocity constant. Increasing the time constant reduced the particle/fluid velocity correlation, which should increase the turbulence
attenuation. However, the particle time constant also appears in the denominator of the turbulence attenuation term.
Thus, the variation of the turbulence attenuation with time,
constant cannot be predicted without a model of the fluid
turbulence spectrum, as observed by the particles. Squires
and Eaton showed that the turbulence attenuation is insensitive to the particle time constant for homogeneous flows
with Stokes numbers between 0.14 and 1.5. Further experimental and numerical studies are needed to determine how
the attenuation will vary over a larger range of Stokes number and in inhomogeneous flows.
VI. CONCLUSION
The results from this work showed four important
points. First, the measurements presented here are, as far as
we knoti, the first measurements that isolate particle/turbulence interaction in a flat-plate boundary layer. Second, for
Stokes numbers of order 1 and Reynolds numbers around 5,
mass loadings of 20% will cause a significant suppression of
the fluid turbulence. Third, the suppression of the turbulence appears to be a strong function of the local particle
concentration in the near-wall regions of high turbulent kinetic energy of the fluid. And fourth, the particles appear to
take energy from all fluid scales equally.
ACKNOWLEDGMENTS
The authors are grateful for sponsorship by a Presidential Young Investigator Award from the National Science
Foundation (Grant No. MEA-83-51417). The grant was
generously matched by the General Motors Corporation.

G. G. Stokes, Trans. Cambridge Philos. Sot. 9,8 (1851).


C. B. Rogers and J. K. Eaton, Report No. MD-52, Thermosciences Division, Stanford University, Stanford, California;l989.
3 M. I. Yudine, Adv. Geophys. 6, 185 (1959).
4 M. R. Wells and D. E. Stock, J. Fluid Mech. 136,31 (1983).
G. T. Csanady, J. Atmos. Sci. 20,201 (1963).
b E. E. Michaelides and D. E. Stock, Turbulence Modification in Dispersed
MultiphaseFlows (ASME FED, San Diego, CA,-l989), Vol. 80.
R. A. Gore and C. T. Crowe, in Ref. 6, p. 31.
S. L. Soo, H. K. Ihrig, and A. F. El Kouh, J. Basic Eng. 82, 609 (1960).
1. D. Doig and G. H. Roper, Ind. Eng. Chem. Fundam. 6,247 (1967).
OK. V. S. Reddy and D. C. T. Pei, Ind. Eng. Chem. Fundam. 8,490 ( 1969).
C. B. Rogers and J. K. Eaton

936

Downloaded 11 Apr 2005 to 132.77.4.129. Redistribution subject to AIP license or copyright, see http://pof.aip.org/pof/copyright.jsp

S. L. Soo, Gus-Solids Flow, Engineering Research Publicatioii (Rutgers


University, Piscataway, NJ, 1965). p, 45.
IR. G. Boothroyd and P. J. Walton, Ind. Eng. Chem. Fundam. 12, 75
(1973).
I3 C. R. Carlson and R. L. Peskin, Int. J. Multiphase Flow 2, 67 ( 1975).
14S. L. Lee and F. Durst, Int. J. Multiphase Flow 8, 125 ( 1982).
Is G. Arnason and D. E. Stock, ASME Vol. FED-lo, 25 (1984).
16Y. Tsuji and Y. Morikawa, J. Fluid Mech. 120,385 (1982).
Y. Tsuji, Y. Morikawa, and H. Shiomi, J. Fluid Mech. 139,417 (1984).
Ia S. E. Elghobashi and T. Abou-Arab, Phys. Fluids 26,93 1 ( 1983).
S. E. Elghobashi, T. Abou-Arab, M. Rizk, and A. Mostafa, Int. J. Multiphase Flow 10,697 (1984).

OS. L. Lee, Int. J. Multiphase Flow 13, 137 (1987).


*IS. L. Lee, Int. J. Multiphase Flow 13,247 (1987).
22S. L. Lee and I. Borner, Int. J. Multiphase F~QW13,233 (1987).
23K . D Squires and J. K. Eaton, Report No. MD-55, Thermosciences Division, Stanford University, Stanford, California, 1990.
K. D. Squires and J. K. Eaton, Phys. Fluids A 2, 119 1 ( 1990).
M. Gaster and J. B. Roberts, Proc. R. Sot. London, Ser. A 354, 27
(1977).
rb L. P. Purtell and P. S. Klebanoff, Phys. Fluids 24, 802 ( 198 1) .
r P. Johnson and J. P. Johnston, Report No. MD-53, Thermosciences Division, Stanford University, Stanford, California, 1989.
28J. Murlis, H. M. Tsai, and P. Bradshaw, J. Fluid Mech. 122, 13 ( 1982).

937
Phys. Fluids A, Vol. 3, No. 5, May 1991
C. 8. Rogers and J. K. Eaton
937
Downloaded 11 Apr 2005 to 132.77.4.129. Redistribution subject to AIP license or copyright, see http://pof.aip.org/pof/copyright.jsp

You might also like