Professional Documents
Culture Documents
by
Michael J. Benson, John K. Eaton
Thermosciences Division
Department of Mechanical Engineering
Stanford University
Stanford, CA 94305-3030
May 2003
ii
Abstract
Previous experimental research has shown that moderate loadings of small, dense particles
can have large impacts in attenuating the turbulence levels of the gas phase in particle-laden
flows. One parameter receiving little attention associated with such flows in fully-developed
channels is that of the wall roughness boundary condition. The objective of this research
was to investigate the nature of the eects of variations in the wall roughness on the particle
velocities.
The experiments were conducted in a well-documented apparatus, rebuilt for the current
work. The facility is the vertical, fully-developed channel flow in air, operated at a Reynolds
number of 13,800 used previously by Kulick et al, Fessler and Eaton and Paris. The final
1.7 m of the development section consisted of two dierent interchangeable sections. One
section, referred to as the smooth wall condition, matches the rest of the development length
and consists of clear acrylic walls. The rough wall section is similarly constructed to the
first, with an overlay of stainless steel wire mesh screen axed to its inner surfaces. The
flow was seeded with monodisperse, spherical 150 m diameter glass particles at a mass
loading of 15%. A single component LDA system was used to obtain flow tracer and particle
streamwise velocity profiles in the wall-normal direction.
The wall roughness condition does not have an impact on gas phase mean velocities in
the tunnel. Gas phase turbulence levels are also quite similar near the wall, while the rough
wall velocity fluctuations exceed those for the smooth wall in the region around the channel
centerplane. Particle velocity PDFs and profiles show a large dependence on the wall
boundary condition. Whereas particles lead the flow in the smooth wall case, they lag the
flow in the rough wall condition, except very near the wall. Throughout the profile, rough
wall particle mean velocities lag their smooth wall counterparts. Particle RMS velocities
are also dierent through the wall-normal profile. Smooth-wall RMS velocities exceed those
of the rough wall nearest the wall, while the opposite is true nearer the channel centerplane.
It was concluded that the rough wall boundary condition enhanced particle-particle and
particle-wall collisions near the wall. It is likely that the wall boundary condition will have
large eects on the gas phase turbulence levels in two-phase flow conditions.
iii
iv
Acknowledgments
The United States Army fully funded the first author through the completion of this work.
The research program was funded by the National Aeronautics and Space Administration
through Grant NAG3-2738.
Grateful acknowledgments go to Dr. Anthony Paris for his previous contributions to
this project, as well as senior PhD candidate Mr. Wontae Hwang, who illuminated many
facets associated with particle-laden flows. In addition, thanks go to Mr. Lakhbir Johal for
his assistance in the rebuilding of several large sections associated with the facility, and Mr.
Ryan Rogers, who provided valuable assistance in data collection.
v
vi
Contents
Abstract iii
Acknowledgments v
vii
4 Summary and Conclusions 41
4.1 Summary of Experimental Results . . . . . . . . . . . . . . . . . . . . . . . 41
4.2 Conclusions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 42
4.3 Recommendations for Future Research . . . . . . . . . . . . . . . . . . . . . 43
References 45
viii
List of Figures
ix
x
Nomenclature
Acronyms
Roman Symbols
CD coecient of drag
dp particle diameter
h channel half-width
m mass
m mass flow rate
np particle number density
N number of samples
O(n) terms of order n or smaller
P pressure
Reh channel Reynolds number
Rep particle Reynolds number
St particle Stokes number
t Time
U fluid velocity
V particle velocity
x1 streamwise direction
x2 transverse direction
x3 spanwise direction
y+ non-dimensional wall normal distance
Greek Symbols
volume fraction
particle mass loading
k Kolmogorov length scale
xi
fluid viscosity
fluid kinematic viscosity
material density
xii
Chapter 1
1.1 Introduction
Turbulent, particle-laden shear flows in air occur in a wide range of applications, includ-
ing industrial conveyance processes, pulverized coal combustion, pollution control systems,
fluidized beds, and the environment. As such, they represent an important class of mul-
tiphase flows, requiring detailed understanding to accurately model them. Over the past
several decades, much eort in research has focused on the nature of these complex flows.
To date, no single analytical model confidently predicts all the key properties of these flows.
The complex nature of the flows, and the various parameters influencing them, is not un-
derstood well enough to develop models usable in more general contexts. The very nature
of the interaction between particles and the carrier phase must be determined experimen-
tally for each new flow geometry and parameter set. As such, a great deal of work remains
in this area of turbulence research. To successfully develop models, therefore, two broad
categories of information must be reliably obtained from these two-phase flows. The first
regards the fluid motion. In this category, statistics such as the mean velocity profile and
any modification of turbulence must be captured. In the second, similar statistics regarding
the particle motion in wall-bounded flows is required. Particle mean velocity profile and
spatial concentration data must be collected. Finally, turbulence fluctuation information
must be gathered to fully account for the energy within the flow.
1
2
& Eaton (1989), Kulick found no change in the mean streamwise fluid velocity values in the
presence of particles. The fluctuating components of the air velocity decreased as particle
Stokes number, mass loading, and distance from the walls increased. Meanwhile, particle
velocity profiles flattened across the wall normal plane of the channel as the mass loading
increased. Of note, this was the first of the Eaton-group experimenters to notice the strange
eect that particles with larger Stokes numbers lagged the gas-phase in the center of the
channel, while leading the flow near the wall. This is despite the fact that gravity is acting
in the flow direction. Intuition suggests that particles would move more rapidly than the
mean flow.
Fessler and Eaton (1995) used the vertical channel studied by Kulick, modified with a
single-sided expansion. In their study, turbulence attenuation was found in specific regions
of the flow in the same manner as Kulick and Eaton (1993). However, other parts of the
flow, namely the separated shear layer and the redevelopment region behind the step, did
not exhibit any turbulence modification from the unladen state. In addition, they put
considerable eort into the study of preferential concentration eects. They showed that
particles are not uniformly dispersed in the fully developed region of the flow as is often
assumed.
Most recently, Paris and Eaton (2001) further investigated the same vertical channel flow
studied by Kulick (1993) using smaller measurement scales and Particle Image Velocimetry
(PIV) techniques in order to identify more clearly some of the mechanisms at work in the
turbulence dissipation process. Paris (2001) verified the same levels of turbulence attenu-
ation as Kulick (1993). The particle velocities measured by Paris consistently lagged the
streamwise gas phase flow at the channel centerline. In addition, the particle fluctuating
velocity profile in the streamwise direction was relatively flat, similar to data found by
Kulick, and exceeded the fluid phase turbulence levels throughout the profile except nearest
the wall.
Both Kulick (1993) and Paris (2001) found one common, yet surprising result. The
mean velocity of the particles in the streamwise direction at the channel centerplane actu-
ally lagged behind the mean velocity of the flow of the carrier phase. Despite the eects
of gravity, the particles used in those experiments moved slower than expected by an ap-
preciable amount. Numerical simulations, however, do not predict this counter-intuitive
result. Rouson and Eaton (1997) conducted direct numerical simulations on particle ve-
locities. They modeled the vertical channel flow studied by Kulick (1993), using specular
particle-wall interactions at smooth boundaries. The results predicted particle velocities,
using similar particles to those used by Kulick and Paris, that would exceed the mean fluid
flow velocity throughout the channel. Particle fluctuating velocities for copper particles in
the streamwise direction exceed those of the fluid phase near the wall, while these fluc-
tuating velocities are nearly equal in the channel centerline. This eect, which shows a
general decrease of fluctuating velocity as one moves from the wall to the channel center, is
a marked contrast from the flat profiles measured by Kulick and Paris.
As an attempt to explain this particle behavior, Paris (2001) suggested four potential
modes whereby particle mean velocities might be reduced. They include insucient chan-
nel development length, particle migration, particle-wall collisions, and particle-particle
Chapter 1. Introduction and Objectives 3
collisions. Of these four, they discount the first, due to the large particle residence time in
the upstream development section versus the particle time constant. The remaining poten-
tial contributors are all closely linked in terms of the physical processes by which they occur.
Increased transverse particle velocities due to particle-particle collisions were deemed the
most likely cause of reduced particle streamwise velocity because the particle concentration
seems to play a large role in the amount by which a particle lags the mean flow. Including
the influence of surface roughness eects in the overall flow analysis was one mechanism
suggested by Paris and Eaton to help explain some of the more surprising aspects of the
particle behavior.
on their turbulent flow, which was in the regime where carrier turbulence is attenuated by
particles. Of note in their experiment, they identified a correlation between the particle
concentrations with the behavior of flow over a rough wall. They were careful to identify
this eect only in terms of a non dimensional sand roughness. The overall eect of particle
concentrations in the flow act in such a way as to produce eects similar to that of a rough-
wall. However, the overall concentration eect, in terms of this equivalent wall roughness
parameter, remained quite small, and was still within the hydraulically smooth regime for
their experiments. Using particles ranging from 100 - 900 m, they found that larger parti-
cles increased this eect. Despite the resemblance of the two eects, they were unsure of the
similarity of the physical mechanisms causing the eect. Finally, they identified streakline
concentration behavior of the particles. Highest particle concentrations were found in low
velocity regions, such as near the walls, with concentration values exceeding those of other
regions by one to two orders of magnitude.
Recently, Tanaka, Maeda, and Hagiwara (2002) conducted numerical studies in a ho-
mogeneous turbulent shear flow. They performed their study using a particle loading in
the very dilute range, where inter-particle collisions had no eects, modeling their flow
with the relatively traditional two-way coupling analysis. They found that gravity plays
an important role in the particle interaction with the carrier phase. In weak gravity condi-
tions, particles actually increased turbulence, but in strong gravity, their overall eect was
to attenuate turbulence. They ascribed the dierence to the location of particle clusters
within the flow. In the weaker gravity conditions, the particle clusters were located in such
a way as to increase the downward flow and enhance Reynolds shear stress. However in the
stronger gravity regimes, particle clusters shifted to locations where drag dissipation eects
attenuated the flow turbulence. This analysis blended the importance of the particulate
interaction with the carrier phase with the eects of local concentrations of particles.
Important studies have also focused on how particles interact with the walls in turbulent
flows. In an experimental and numerical study conducted by Frank et al, (1993), a square
horizontal channel was used to shoot particles at walls of dierent consistencies. They
studied the resulting coecients of restitution and the dynamic friction coecients in an
attempt to better predict what happens to particles after contact with walls. Generally, they
found that particles in the presence of rough walls rebound at a fairly consistent mean value
of 65 degrees from parallel. Using these values, their numerical results agreed reasonably
well with their experiments. In each of their comparative profiles, the particles did lag the
mean flow by a slight percentage. They attributed this lag to three-dimensional flow eects
developed in their square cross-section.
Sommerfeld and Huber (1999) did a similar study of the particle-wall interaction eects.
Glass and quartz beads with diameters between 100 and 500 m were used. As in Frank et
al, they focused on particle rebound properties. Wall roughness eects were seen to depend
on particle size and geometrical eects of the roughness elements. In general, particles
were seen to rebound at increasingly large angles as the relative roughness of the surface
increased. Non-sphericity was also a factor in the rebound characteristics of the particles.
Using the coecients of restitution and friction from their experiments, they compared their
resulting rebound angles from a variety of impact angles for polished steel and rubber walls,
Chapter 1. Introduction and Objectives 5
dVi 1
= gi (Vi Ui ) (1.3)
dt p
In the case of non-Stokesian particles, particle inertial eects must be accounted for,
and the resulting dierential particle equation of motion then becomes (using the empirical
formula for the CD from above):
dVi 18
= gi [1 + 0.15Re0.687
p ](Vi Ui ) (1.4)
dt p d2p
This equation can be integrated forward in time until the terminal slip velocity (Vi Ui )
is reached. This terminal velocity, which neglects eects associated with any cross stream
mixing of the particles, represents the fastest the particle should travel for the given ex-
perimental configuration. For example, given a constant fluid velocity of 10.5 m/s, the
non-Stokesian terminal particle velocity for a 150m sperhical glass particle and the above
assumptions would be 11.48 m/s, reached in approximately 0.5 seconds.
are replaced with acrylic Plexiglas, optical access is improved, and components checked to
ensure everything operates in such a way as to make the entire experiment reproducible.
This included developing an LDA system capable of accurately measuring particle velocity
statistics. In addition, the final two meters of the development length are built with three
exchangeable wall plates including smooth walls made of plain acryclic, medium-rough walls
made of particleboard, and rough walls consisting of wire mesh screen attached to acrylic.
The second objective of this work is to measure particle mean velocities and streamwise
turbulence intensity of each of these three roughness regimes and compare and contrast
their statistics. It is estimated that this second objective will definitively answer whether
wall roughness is the primary reason by which mean particle velocities are reduced.
8
Chapter 2
This chapter discusses the experimental apparatus and the data measurement techniques
used in this study. Some qualifications of the techniques are presented, and estimates and
analysis of the sources of error are discussed. Many of the details contained in this section
are similar to works done by Paris (2001), Fessler (1995), and Kulick (1993), and thus
details are provided here only when there are significant dierences relative to the previous
works.
9
10
to the wooden platform. In this configuration, the blower can create tunnel velocities in
excess of 15 m/s.
The air exits the fan through a rectangular diuser and enters a 25.4 cm diameter sheet
metal duct. It travels 2.5 m through the duct, and enters a particleboard duct at the top of
the wind tunnel. Over a horizontal distance of 70 cm, this section turns the flow 90 degrees
from a 40 cm x 25.6 cm rectangular section, into a 40 cm x 45.7 cm rectangular cross
section, 0.87 m long. The header-type 90 degree bend replaces a flexible pipe connection
on the original apparatus. Flow-conditioning grids are placed in four locations throughout
this section. A clear acrylic 40 cm x 45.7 cm rectangular section, 0.82 m long, follows
the particleboard ductwork. Particles are introduced into the flow in this region. A flow-
conditioning grid follows the particle loading system. A wire mesh screen follows this grid,
and a second flow-conditioning grid follows the screen. The grids and screen act to uniformly
disperse the particles in the flow.
The particle loading system, as shown in figure 2.3, is identical to that used by Paris
(2001). It consists of a large number of drainage buckets connected to two sets of conveyer
belts that rotate across the breadth of this loading section. The two sets of drainage buckets
are filled through a particle hopper mechanism. Pressurized air starts the flow of particles
through drainage tubes into the rapidly moving buckets. The buckets then drain steadily
across the tunnel to the opposite side where a matching feeding mechanism refills them
for the return trip. The buckets are constructed with inlets that are much larger than the
outlets, allowing them to be filled rapidly while draining more slowly. The particle hoppers
each hold 17 L of particles, so a typical experimental run can last for nearly three minutes
with a constant mass flow rate. The belts are moved by a 1/12 hp Boston Gear DC motor
with an RB1B Beta II speed controller. Adjusting the speed of the controller can increase
or decrease the resultant mass flow rate of particles in the channel.
The two-phase flow next enters a 10:1 contraction that smoothly transforms the 40
cm x 45.7 cm cross section into the high aspect ratio channel cross section with interior
dimensions of 4.0 cm x 45.7 cm. A 5 cm long aluminum Hexcel honeycomb section is pressed
into the end of the contraction to minimize the cross stream velocity of the particles exiting
the contraction. After exiting the honeycomb, three newly constructed 1.27 cm thick acrylic
channel development sections follow. Each section is seamlessly connected with the previous
through a flange and gasket connection, and the total height is 5.2 m (260 channel half
heights), with constant interior dimensions of 4.0 cm by 45.7 cm. Support ribs throughout
the development sections assure that the overpressure in the tunnel does not significantly
bulge the channel walls. The transparent nature of the acrylic sections allows for one
to visually see the particles from the exit of the honeycomb cells at the contraction exit
through the test section. Static pressure taps are emplaced every 30.5 cm throughout the
development section.
The final of the three 1.7 m long development sections is an exchangeable section that
is relatively easy to replace. There were three versions of the replaceable section that could
be exchanged to investigate the eects of dierent wall boundary conditions. The first
version was made of smooth acrylic. It is smooth both to the touch and optically, with no
measureable roughness. The second version was the rough-wall development section. This
Chapter 2. Experimental Facility and Techniques 11
was another acrylic section, but square-mesh wire screen was bonded to both the 45.7 cm
wide walls to make a surface with large regular roughness. The screen was manufactured
from 0.25 mm diameter wire at a grid spacing of 1 mm. The screen was bonded to the
acrylic using a 3M 90 high strength spray adhesive. The third version had one of the 45.7
cm walls covered with the same screen, and the opposite was manufactured with 12.7 mm
thick particleboard.
Particle velocities in the development section were approximately 10 m/s near the chan-
nel centerline, and slower elsewhere. Therefore, even a fast moving particle took approxi-
mately 170 ms to traverse the length of the final development section. Most particles took
longer. Also, the length of the final development section corresponds to 85 channel half-
widths. Therefore, it was believed that the final development section was suciently long
for the flow to become fully developed prior to entering the test section.
Figure 2.4 shows the dimensions and layout of the newly constructed test section. It has
a total length of 63.5 cm, and is optically transparent. Regardless of the wall conditions
in the final development section, the test section retained its smooth acrylic walls required
for optical access. Based on the average particle speeds in this region, the particles have an
average residence time of less than 30 ms before reaching the measurement volume. The test
section has five static pressure ports, each 2.54 cm apart, as well as a column of instrument
ports 5 cm from the channel centerline for devices such as pitot probes and temperature
sensors. The test section has optically clear glass walls in the middle 20 cm of each side of
the 4 cm wide side walls. These walls are clearer than the acrylic, so oer improved optical
imaging and laser transmittance. In addition, a 20 cm square glass window in the center of
one 45.7 cm channel wall facilitates the performance of the collection optics. The opposite
side test section wall has a matching sized, removable acrylic window to allow for access
into the interior of the channel for purposes of laser alignment and any required cleaning.
Of note, the entire test section support frame was emplaced using a laser level device. The
entire tunnel was then secured to the frame ensuring that it was square with the fixed frame,
thereby validating the vertical orientation of the entire apparatus.
At the exit of the test section, a custom designed sheet metal contraction smoothly
transitions the rectangular duct flow into a 10 cm diameter PVC pipe, which turns through
a 90 degree elbow, travels horizontally for 1.4 m, goes through another 90 degree elbow,
travels vertically for 2.4 m, bends a third time, and enters an Environmental Industrial
Products model FK912D1 NF12 cyclone separator. The cyclone separator extracts the
particulate phase from the gas phase, exhausting the air to an exit 7 m above and 10 m
horizontally through the building roof, while dropping the particles into a stainless steel
tote directly beneath the cyclone. At the end of each experimental run, the mass inside
the tote was weighed. Dividing the mass by the duration of the particle flow gave the
particulate mass flow rate. The particles were reused several times, and were brought back
to the upstairs loading platform using a 9 m electric chain hoist.
12
TSI Inc. products used in Fessler (1995), and Kulick (1993). Figure 2.6 shows the basic
components of the LDA system, and Table 2.3 describes the primary components for the
complete system. The transmitting optics assembly was mounted on a flat rail system, which
in turn was controlled by a traverse. The traverse, oriented parallel to the Y component of
the tunnel (that is, in the wall-normal direction), was powered by a 1/6 hp APM916AT-1
Boston Gear DC motor, controlled by a Boston RB1R Beta II speed controller.
The elliptical measurement volume was estimated, based on fringe spacing and the
number of fringes in an oscilloscope trace of an average signal, at 75 x 500 m, with an
average depth estimated at 75 m. This fairly small measurement volume, compared to the
150 m glass particles, yielded excellent data rates during all experimental runs and for all
of the mass loadings tested in the experiment.
The laser used in the study was a Lexel Model 95 Argon-Ion laser. Initially, the laser
was operated in its standard multiline operational mode, where green and blue wavelengths
dominated. To increase the data aquisition rate the laser was placed into single line mode
by adding an etalon device into the laser head. The 488 nm wavelength (blue) used was
operated at a continuous power of 1 Watt.
Beams were oriented in the X Z plane of the tunnel, entering through the narrow channel
wall, and crossing near the center of the 45.7 cm wide channel dimension. This allowed for
the long axis of the measurement volume to be oriented in such a way as to provide the best
spatial resolution in the wall normal plane of the tunnel. Only streamwise components of
velocities were measured with this apparatus, and all profiles subsequently referred to are
in the wall-normal direction.
Once data were taken at one location, the transmitting optics were moved by the tra-
verse to the next profile location. There were 8 meaurement locations within the channel,
including centerline (20 mm from wall), 10 mm, 8 mm, 6 mm, 4 mm, 3 mm, 2 mm, and
finally 1 mm from the wall. After moving the traverse so that the beam crossing point
coincided with the appropriate profile location, the collection optics was then moved on a
Melles Griot optical platform until they were again refocused on the measurement volume.
For all experiments, the collection optics were focused such that the maximum data rate
was achieved when using a TSI model 9302 atomizer oriented into the measurement volume.
stored. The recording of data can also cease at the discretion of the experimenter. For
most two-phase experiments, approximately 5000 velocity measurements are aquired.
At the conclusion of the experimental run, the particle-feeder belts are switched o and
the stopwatch is stopped. The total time interval for the particle loading can be accurately
recorded. A Pelouze 4010 heavy duty shipping scale with a resolution of 0.05 kg is used
to weigh the particles collected in the steel tote. The final step in the experimental run is
to shut o the compressed air supply to the particle hoppers. The total elapsed time for an
experimental run is approximately 120 seconds. The size of the particle hoppers limit the
flow of particles to a maximum of 150 seconds. Over the course of this study, laboratory
temperatures varied by less than 5 degrees Celsius and atmospheric pressure readings varied
by less than 0.3 kPa.
the beams further supports this conclusion. Finally, the digital burst correlator thresholds
all Doppler burst records that are substantially dierent than their neighbors, a feature
which all but eliminates chances of this error.
(N 1)s2 2 (N 1)s2
x < (2.2)
2(N1); 2(N1);1
2 2
where x2 is the true variance and 2N; is the chi-squared distribution. For large values of
N, the chi-squared distribution is approximated by
" s #3
2 2
2N; = (N 1) 1 + z (2.3)
9(N 1) 9(N 1)
where z is a function of the confidence level for the chi-squared distribution (z =1.96 for
95% confidence as N ).
LDA velocity measurements used a minimum of 4000 samples. For a 95% confidence
interval for the particulate phase, this uncertainty varies from 0.6% at the centerline to 1.1%
near the wall. In the case of gas phase statistics, where 20,000 samples were taken, this
corresponds to uncertainties varying from 0.2% to 0.5% for the same confidence interval.
The corresponding uncertainty using the chi-squared distribution for the rms fluctuating
velocities in the particle phase was on the order of 2.5% at the channel centerline, and
reduces as the wall is approached.
The error associated with the digital burst correlator is extremely low, as the manufac-
turer provides a sample resolution accuracy on the order of 0.05%.
Perhaps a more likely source of error to consider is the actual positioning of the beams
within the channel. The beams were initialized at the channel centerline, which corresponds
to 20 mm from each wall. This distance was measured with a ruler with 0.5 mm accuracy.
Each subsequent wall normal distance was measured with this same technique and same
Chapter 2. Experimental Facility and Techniques 17
ruler. Each data point in the profile represents an average of at least four experimental runs,
which tends to reduce this type of location uncertainty. Since one individual measured all
wall normal distances of the beams for each experiment, the error in this measurement is
estimated as very low, on the order of 5% as a conservative estimate. As this error represents
uncertainty in the actual location within the profile for a given data point, each point has
an equal likelihood of actually representing a point less than one particle diameter on either
side of its stated location, which incurs very minimal eects on the analyses in this report.
acrylic. This change in turn aects the values of dependent flow parameters in the fluid
flow parameter table.
dP
w = h (2.4)
dx
and the shear velocity is
r
w
U = (2.5)
A least squares fit of the data yields a shear velocity of 0.47 m/s, in fair agreement with
the value of 0.49 m/s measured by Kulick and Eaton (1993). Paris and Eaton (2001) showed
an unladen wall pressure gradient of 12.2 m/s2 , as compared to the pressure gradient of
12.1 m/s2 measured in this report.
Chapter 2. Experimental Facility and Techniques 19
h 20 mm
Ucl 10.5 m/s
Reh 13,800
U 0.47 m/s
from
particle
hoppers
tunnel cross-section
overflow drainage
x3
x2
31.1
x1
63.5
glass
windows
45.7
Figure 2.4. Test section schematic, in cm (from Paris and Eaton, 2001)
Chapter 2. Experimental Facility and Techniques 23
1.0
0.8
0.6
U1 / U1 CL
0.4
Paris
Kulick
0.2 Smooth Wall
Rough Wall
0.0
0.0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1.0
x2/h
0.080
0.040
0.000
0.0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1.0
x2/h
0.6
B = 0.1 (Paris)
rough wall
0.4
PDF
0.2
0.0
5 6 7 8 9 10 11 12
0.8 Paris
Smooth Wall
0.6
DP/(0.5rfU1 2)
0.4
0.2
0
-250 -200 -150 -100 -50 0
x1/h
This chapter presents the experimental results from this research. It is divided into two
components, gas phase statistics and particle velocity measurements for the diering wall
roughness regimes. Profile data are normalized using the mean streamwise centerline gas
phase velocity of 10.5 m/s. Where appropriate, selected data from Paris and Eaton (2001)
are presented along with the present data for comparison. The particle board apparatus
used by Kulick, Fessler, and Paris can be expected to have a roughness level roughly cor-
responding to the 150 m glass particles. This is attributed to the many years of research
use in which glass was the prevalent particulate phase utilized. The interior of the duct was
coated with glass particles. As such, that earlier case should represent a condition that is
rougher than the present smooth wall, but not as rough as the mesh screen rough wall.
27
28
Paris, a larger 330x330 m2 area of the channel centerline was interrogated. Because of
the larger area in the Paris measurements, all turbulence fluctuations smaller than this
length scale cannot be captured by this technique Paris (2001). While this dierence is
likely to have only minimal eects, it does highlight one potential result of the dierent
measurement techniques.
It is curious to note, however, that despite the extreme dierences in the wall boundary
conditions in the three cases, the data for each of the four experiments depicted has its
best agreement nearest the wall. This similarity between data in the same region where
the greatest dierences in experimental conditions exist seems to imply that the physical
processes aected by the wall condition must exhibit their impact well away from their
source. A reasonable approach might consider the development of the change through
the final development section where the dierent boundary conditions exist to see if the
turbulent fluctuations of the gas phase diuse away from the wall. Likely in the region
where the roughness begins, significant dierences at the wall are found, which then tend
towards the center of the channel as the flow develops downstream.
For convenience, a table of the data used in this section is provided in table 3.1.
observable. Only one velocity is evident and the apparent bi-modal characteristic of the
centerline PDFs is no longer present.
seen. As has been the case through most of this discussion, the data from Paris (2001) most
closely resembles the case of the rough wall. The shape of the smooth wall particle fluctu-
ating velocity profile, however, parallels that of the gas phase fluctuating velocity profile, a
trait quite evident in figure 3.10.
For the smooth wall case, the highest particle RMS velocity occurs at the nearest mea-
surement location to the wall. Contrarily, the Paris and rough wall cases have their highest
fluctuating velocities at the channel centerplane. Probably, the particle mean velocity gra-
dient is larger near the wall in the smooth wall case because particles can remain there for
a longer time. They are not forced away from the wall by the obstructive nature of the
roughness elements. This near-wall residence time can produce large particle mean velocity
gradients, which produce relatively larger particle fluctuating velocities. It is apparent that
dierent processes are at work near the wall for the smooth and rough wall cases. The rough
wall conditions likely eject the particles towards the center of the flow rapidly, reducing the
residence time of particles near the wall, and thereby reducing the velocity gradient near
the wall. This reduced gradient exhibits itself in lower particle fluctuating velocities for the
rough case near the wall.
Also of note, the location in the profile where the gas phase and the particle fluctuating
velocities cross for the smooth and rough walls is quite similar. Figure 3.11 shows that at a
value of x2 /h of 0.2, which corresponds to a distance of 4 mm from the channel wall, there
is a change in the nature of the particle RMS velocity. Specifically, closer than 4 mm from
the wall, the smooth wall particles have higher RMS velocities than rough wall particles.
Closer to the centerline, however, the rough wall particles have fluctuating velocities that
exceed those of the smooth wall. A similar, albeit less identifiable, eect occurs for the gas
phase profiles. The rough wall gas phase has higher RMS velocities on the centerline side
of the 4 mm location. The RMS velocities for the rough and smooth wall in the gas phase
are quite similar as they get closer to the wall, with identical values at the location 1 mm
from the wall. The importance of this distance, which corresponds to 27 particle diameters
or 16 roughness heights from the wall, is not readily evident from these results. However,
even in the gas phase statistics, the data for the RMS velocities for smooth and rough wall
cases presented here begins to diverge from that of Kulick and Paris at this same region of
the channel. As a final note on this peculiar behavior, the data of Paris (2001) is presented
in figure 3.12 to indicate that indeed at this same profile location the characteristics of gas
and particle phases become significantly dierent in the wall normal velocity component.
One eect contributing to the dierences in the velocity PDFs is the likely dierence in
the particle-wall collision process. In the region nearest the wall where collisions are a near
certainty for the case of the rough wall, and presumably in the case of Paris, the narrower
PDFs, and hence smaller particle fluctuating velocities, indicate a sort of velocity mixing
eect close to the wall. In the smooth wall case where these collisions are not as certain, no
such mixing is assured, and greater particle RMS velocities are seen. As regards particle-
particle collisions, this same logic should hold, as particles rebounding from wall collisions
transport their energy across the channel through collisions with other particles. If particle-
particle collisions are important closer to the channel centerplane region, as discussed by
several authors mentioned in Chapter 1, then their eect in this experiment must be to
32
enhance particle fluctuating velocities in the rough wall and Paris cases.
Sommerfeld and Huber (1999) suggest that the eects of roughness are strongly de-
pendent on the ratio of the particle diameter to the roughness height. Referring again to
figure 3.10, it is interesting to note that in the region closest to the wall, and presumably
representing those particles that have had the most collisions with both wall and neigh-
boring particles, the data of Paris shows the lowest particle RMS velocties. In the work of
Paris, the particle size and the characteristic wall roughness size are assumed to be equal.
In the smooth wall case, the ratio of particle size to wall roughness size is largest, and in
the rough wall, this same ratio is the smallest. If there is to be a correlation, therefore, it
appears as though a particle size to wall roughness ratio nearest unity produces the greatest
eects of dispersing the particle velocities through increased collisions both with the wall
and with other particles. This is a substantially dierent conclusion than the one suggested
by Sommerfeld and Huber, which suggests that the smaller this ratio between particle and
roughness sizes, the larger the resulting eects.
When comparing the results with those in Kussin and Sommerfeld (2002), less similarity
can be seen in the RMS characteristics than in the mean velocity characteristics. Whereas in
the present report rough wall particles have higher RMS velocities in the channel centerline,
while lagging near the wall, the report by Kussin and Sommerfeld (2002) shows that the
RMS velocities are uniformly higher across the channel for the high roughness case. This
dierence may possibly be a function of either the dierent particle size or mass loading, or
even the higher tunnel velocities, generally run much faster than the 10.5 m/s used in this
report. In either case, a fundamental dierence between the two experiments remains. The
augmentation in the RMS velocity associated with the increase in wall roughness for the
case of Kussin was attributed to enhanced transverse dispersion of particles. While this is
well supported in the centerline region of the current report, the explanation near the wall
provided here, namely that increased wall collisions cause higher transverse velocities, and
particles do not remain near the wall as long, does not hold in the same fashion as before.
For convenience, a table of the data used in this section is provided in table 3.2.
Chapter 3. Velocity Measurements in the Wall Normal Plane 33
1.4
1.3 Paris
1.2 Smooth Wall
1.1 Rough Wall
1.0
0.9
PDF (m-1s)
0.8
0.7
0.6
0.5
0.4
0.3
0.2
0.1
0.0
8 10 12
U1 (m/s)
0.80
0.60
U1 / U1 CL
Paris
0.40
Smooth Wall
Rough Wall
0.20
0.00
0.0 0.1 0.2 0.3 0.4 0.5 0.6
x2/h
0.16
Paris
Kulick
0.12 Smooth Wall
Rough Wall
U1 rms/ U1 CL
0.08
0.04
0.00
0.0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1.0
x2/h
0.4
0.3
0.2
0.1
0.0
6 8 10 12
V1 (m/s)
0.8
0.6
PDF
0.4
0.2
0.0
8 10 12
V1 (m/s)
0.3
0.2
0.1
0.0
4 6 8 10 12
V1 (m/s)
Figure 3.6. PDFs of particle phase velocities near the channel wall
38
1.20
1.00
0.80
V1 / U1 CL
0.60
Paris (f = 0.2)
0.00
0.0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1.0
x2/h
0.10
0.08
0.06
(V1-U1) / U1 CL
0.04
0.02
0.00 Smooth Wall
x2/h
Figure 3.8. Slip velocity and still-air particle terminal velocity profile
Chapter 3. Velocity Measurements in the Wall Normal Plane 39
0.20
Paris
0.10
0.05
0.00
0.0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1.0
x2/h
0.20
0.10
0.05
0.00
0.0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1.0
x2/h
0.20
Smooth Wall Gas Phase
Rough Wall Gas Phase
0.15
Smooth Wall Particle Phase
V1rms/ U1 CL
0.10
0.05
0.00
0.0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1.0
x2/h
gas-phase
0.08
particle
V2 rms/ U1 CL
0.06
0.04
0.02
0.00
0.0 0.1 0.2 0.3 0.4 0.5 0.6
x2/h
Figure 3.12. Transverse velocity fluctuation intensity for 150 m glass particles at a mass
loading of 0.2 (From Paris)
Chapter 4
This chapter provides a brief summary of the results obtained, draws logical conclusions
from those results, and suggests further work that might be done to extend the base of
information presented.
41
42
channel profile due to the wall roughness condition. When comparing the magnitude of the
associated dierences with the work compiled by the previous experimenters, wall roughness
clearly plays a larger role in modifying particle velocities than mass loading for the regimes
studied. Smooth wall conditions show particle velocities that lead the flow as intuition
suggests. Rough wall particles lag the flow in all regions except very near the wall. Finally,
particle slip velocities increase as the wall is approached.
Particle RMS velocities dier substantially in two ways. Nearest the wall, the smooth
case suggests there are fewer particle-mixing eects, resulting in increased velocity gradients
between particles and a higher RMS velocity. After a distance from the wall of roughly 26
particle diameters, this eect is no longer seen. The rough wall case then shows steadily
increasing RMS velocities to the channel centerplane, while the smooth wall shows steadily
decreasing RMS velocities. This crossing location has additional importance in other data
sets. It represents the location in previous work where particle transverse RMS velocities
exceed their gas phase counterpart. In addition, gas phase characteristics in the streamwise
direction also show a divergence which begins at this same location.
The very flat nature of particle velocity profiles measured in previous work was only gen-
erally seen here in the rough wall case, indicating that rough wall eects likely contributed
to certain characteristics of the data presented in the earlier report. The current smooth
wall case shows behavior that more closely resembles the gas phase characteristics.
4.2 Conclusions
The results presented in this study are straightforward and clear. The consideration
of wall roughness is an important parameter in particle-laden channel flows. It has large
eects on the particle mean and RMS velocities in fully developed channel flow at all points
within a profile of the wall-normal plane. On the other hand, wall roughness seems to have
little impact on the mean velocity of the gas phase, a characteristic most clear in the region
nearest the wall. Wall roughness also slightly augments the gas phase turbulence in the
region nearest the centerplane of wall-bounded flows, while near the wall no dierences in
RMS velocities can be identified.
As regards the nature of the mechanisms behind the wall roughness eect, the particle
slip velocity may yield important information. In the present report, it was shown that
the particle slip velocity for the smooth wall case exceeded that of the rough wall case
throughout the flow, with highest values for each case in the near wall region. The influence
of the particle Reynolds number has already been established as an important parameter
in fully-developed channel flows. The ramifications of this slip velocity behavior indicates
that particle Reynolds numbers are higher in the smooth wall case than in the rough.
However, the particle slip velocities used in this report relied on average velocities, rather
than fluctuating velocities. Regardless, the implication remains that these smooth wall
particles are further from the Stokes regime than those with the rough wall boundary
condition. Particle wake eects will increase as the particle Reynolds number leaves the
very low Reynolds numbers associated with Stokes flow.
Chapter 4. Summary and Conclusions 43
Previous studies have clearly shown that particles attenuate gas phase turbulence levels
in fully developed channel flows. Increased particle wake eects were one characteristic
shown in the study by Paris (2001) to have increased the levels of gas-phase turbulence
attenuation. The mass loading was also seen to have strong eects regarding the magnitude
of that attenuation. In this report, the wall roughness condition impacts particle velocities in
a substantially greater way than the mass loading did in the study by Paris. The likelihood
that particle-particle and particle-wall collisions are increased in the rough wall regime is
quite high. As this eect is a likely contributor to gas phase attenuation, it seems reasonable
that wall roughness eects will tend to increase the attenuation of gas-phase turbulence for
the loading regime studied herein.
This project has built on the existing foundation of fully developed channel flow knowl-
edge established by a number of previous researchers. In each of those cases, the apparatus
consisted of development sections constructed of particle board wall material coated with
an accumulation of particles from years of experimentation. The recent study of Paris
and Eaton (2001) resembles the data associated with the rough wall case in the present
study. With the exception of the particle RMS velocity very near the wall, the rough wall
case matches the particle board study results in terms of particle velocity PDF shapes and
magnitudes, and general velocity profile trends.
Finally, taking measurements through the wall roughness development region to see the
growth of the eects away from the wall might provide substantial information regarding the
mechanisms by which particle velocities are aected in the centerplane region. While this
will provide some optical challenges for PIV or LDA techniques, continued improvement of
existing measurement techniques may provide an avenue by which this information might
be successfully obtained.
References
Bendat, J. and A. Piersol (1971). Random Data: Analysis and Measurement Procedures.
Wiley-Interscience, New York.
Fessler, J. R. (1995). Particle-Turbulence Interaction in a Backward-Facing Step Flow.
Ph. D. thesis, Stanford University, Stanford, CA 94305. Also available as Fessler, J.R.,
Kulick, J.K., & Eaton, J.K. Report No. MD-70, Mechanical Engineering Department,
Stanford University.
Fohanno, S. and B. Oesterle (2000). Analysis of the eect of collisions on the gravita-
tional motion of large particles in a vertical duct. International Journal of Multiphase
Flow 26, 267292.
Frank, T., K. Schade, and D. Petrak (1993). Numerical simulation and experimental in-
vestigation of a gas-solid two-phase flow in a horizontal channel. International Journal
of Multiphase Flow. 19, 187198.
Kaftori, D., G. Hetsroni, and S. Banerjee (1998). The eects of particles on wall turbu-
lence. International Journal of Multiphase Flow. 24, 359386.
Krogstad, P. A., R. A. Antonia, and L. W. B. Browne (1992). Comparisons between
rough- and smooth-wall turbulent boundary layers. J. Fluid Mech. 245, 599617.
Kulick, J. D. (1993). On the Interactions Between Particles and Turbulence in a Fully-
developed Channel Flow in Air. Ph. D. thesis, Stanford University, Stanford, CA
94305. Also available as Kulick, J.K., Fessler, J.R., & Eaton, J.K. Report No. MD-66,
Mechanical Engineering Department, Stanford University.
Kussin, J. and M. Sommerfeld (2002). Experimental studies on particle behaviour and
turbulence modification in horizontal channel with dierent wall roughness. Exp. Flu-
ids 33, 143159.
Paris, A. D. (2001). Turbulence Attenuation in a Particle-laden Channel Flow. Ph. D. the-
sis, Stanford University, Stanford, CA 94305. Also available as Paris, A.D., & Eaton,
J.K. Report No. TSD-137, Mechanical Engineering Department, Stanford University.
Perry, A. E., K. L. Lim, and S. M. Henbest (1987). An experimental study of the tur-
bulence structure in smooth- and rough-wall boundary layers. J. Fluid Mech. 177,
437466.
Perry, A. E., W. H. Schofield, and P. N. Joubert (1969). Rough wall turbulent boundary
layers. J. Fluid Mech. 37, 383413.
Raupach, M. R., R. A. Antonia, and S. Rajagopalan (1991). Rough-wall turbulent bound-
ary layers. Applied Mechanics Review 44 (1), 125.
Rogers, C. B. (1989). The Interaction Between Dispersed Particles and Fluid Turbulence
in a Flat-plate Turbulent Boundary Layer in Air. Ph. D. thesis, Stanford University,
Stanford, CA 94305. Also available as Rogers, Chris B. & Eaton, J.K. Report No.
MD-52, Mechanical Engineering Department, Stanford University.
45
46