Experimental Investigation of Particle Migration in Suspension

Flow Through Bifurcating Microchannels

Bhaskar Jyoti Medhi, Vipin Agrawal, and Anugrah Singh
Dept. of Chemical Engineering, Indian Institute of Technology Guwahati, Guwahati 781039, India

DOI 10.1002/aic.16084
Published online January 22, 2018 in Wiley Online Library (wileyonlinelibrary.com)

Experimental measurements of velocity and concentration profiles were carried out to study transport of non-colloidal
suspension in bifurcating micro channels for both diverging and converging flow conditions using a combination of
mirco-particle image velocimetry and particle tracking velocimetry techniques. Migration of particles across the stream-
line was observed and symmetric velocity and concentration profile in the inlet branch becomes asymmetric in the
daughter branches. Further migration of particles toward the center of the channel in the outlet branch make the pro-
files again symmetric. The evolution of velocity and concentration profiles was observed to be different in the symmetric
and asymmetric bifurcation channels. The comparison of the streamlines for the fluid and the particles showed signifi-
cant deviation near the bifurcation region. This may explain why there is unequal flow and particle partitioning during
flow of suspension in asymmetric bifurcating channels as reported in many previous studies. V C 2018 American Institute

of Chemical Engineers AIChE J, 64: 2293–2307, 2018

Keywords: bifurcation channels, micro particle image velocimetry, suspension, shear induced migration

Introduction of bifurcation networks especially in the lung, aorta, and brain.

Flow of suspension through microchannels has drawn sig-
nificant attention in recent years due to its immense scientific
and engineering applications in flow focusing, separation, and
fractionation devices. Due to comparable size of the particles
and channel width, the flow behavior in microchannels is sig-
nificantly different from those of larger dimensions. In the
pressure driven flow under the conditions of small but finite
Reynolds number the particles in suspension undergo lateral
migration.1,2 Another mechanism of particle migration is
hydrodynamic shear induced migration caused by velocity
gradients.3 Shear induced particle migration has been observed
even for the conditions of very low Reynolds number where
the effect of inertia may be neglected. In pipe or channel flow
the shear induced migration causes the particles to move from
the high shear rate regions to the low shear rate regions. This
lateral migration can be more pronounced in microchannels
since the mean velocity gradient can be large due to small
width, and also the distance over which the migration has to
take place is considerably less.4–6 This effect has been utilized
in separation of biological particles, such as cells and proteins,
and focusing of bacteria and DNAs in microchannels.7 Most
of the research work on particle migration has been devoted to
straight channels and tubes.8,9 In several applications involv-
ing microfluidic devices suspension transport in channels with
bifurcations is commonly encountered. Suspension transport
in bifurcating channels is also encountered in physiological
systems. Micro vascular system in human body has a number
Correspondence concerning this article should be addressed to A. Singh at
anugrah@iitg.ernet.in.
C 2018 American Institute of Chemical Engineers
V Xing20 studied the pressure driven flow of RBC suspensions
In the human blood circulatory system the bifurcation region
has been detected as the region prone to aneurysm, Reynaud,
coronary artery, carotid disease, etc. For this reason the subject
of red blood cells (RBC) transport in bifurcating channels has
drawn significant attention in the literature.10–12 Most of these
studies have focused on the investigation of velocity profiles
in the region close to the bifurcations and confluences with an
objective to infer hematocrit profiles of human blood flowing
through bifurcating microcapillaries. Therefore, the study of
suspension transport in bifurcating channel is a relevant
problem.
Another set of studies have investigated fractional flow of
particles in bifurcating channels dividing into two daughter
branches. It was observed that the particle fraction generally
increases in the branch with higher flow rate compared to the
one with lower flow rate. This phenomena is commonly
known as Zweifach-Fung effect.13,14 Roberts and Olbricht
reported that in a bifurcation network, the partitioning of par-
ticles is dissimilar from the partitioning of fluid.15,16 In a Y-
shape bifurcation they observed particles preferentially enter-
ing the daughter branch which has the higher volumetric flow
rate. The partitioning of particles in the daughter branches
depends on the local and instantaneous flow rate.17 Utilizing
the principles of Zweifach-Fung effect Aarts et al. proposed a
method known as “plasma skimming” which is used to sepa-
rate RBC from whole blood by flowing it through a network
of well-designed bifurcations.18 They also observed significant
asymmetry of the velocity profiles in the downstream
branches. Rong and Carr19 observed that for circular tube
bifurcations, the separating plane is a function of the ratio of
branch width and the fractional flow at the junction. Wang and

AIChE Journal June 2018 Vol. 64, No. 6 2293

through axisymmetric channels using numerical simulations
and observed that the velocity profile was markedly blunt in
comparison to the parabolic profiles of pure plasma flow. The
transport of RBC and cancer cells in symmetric bifurcations
and confluences has also been studied to ascertain that the
behavior of RBC and cancer cells was different near the bifur-
cations and confluences.21 Wang et al.11 carried out numerical
simulation in diverging and converging channel to investigate
the effects of RBC shape, the deformability of membrane and
hematocrit on rheological nature and the hemodynamic. They
also reported the lateral migration of RBC in the inlet branch
and blunted velocity profiles in the inlet as well as daughter
braches.
The phenomena of particle migration in straight tubes and
channels have been studied extensively, but less attention has
been given to bifurcating channels. Although, in the past some
studies have been conducted for suspension flows in micro
bifurcations network, many aspects of it is yet to be under-
stood. Much less is known about the redistribution of particles
near the bifurcation junctions which leads to nearly equal par-
ticle partitioning but unequal flow partitioning even though
the channel is asymmetric. In this study we have tried to
understand the cross streamline migration of particles near the
bifurcation region via flow visualization experiments. Dinther
et al. have presented a good review of the experimental techni-
ques to study concentration and velocity profiles of suspension
flow in microfluidic devices.22 Leble et al. investigated the
velocity profiles of RBC through divergent and convergent
symmetric bifurcation channels using micro-particle tracking
velocimetry (l-PTV) technique and reported asymmetric pro-
files after the bifurcation.10 Lima et al. studied migration of
RBC using a confocal l-PTV system and calculated the lateral
dispersion coefficients of particles in a microchannel.23 The l-
PIV technique has also been used to measure the velocity pro-
file, wall shear stress of non-Newtonian fluid through bifurcat-
ing microchannels.24,25 For the case of two phase flow where
the interest is to know the velocity profiles of both the phases,
hybrid particle image velocimetry (PIV)–particle tracking
velocimetry (PTV) method is widely used.26,27
In most of the studies on bifurcating microchannels it was
assumed that the particles follow the fluid motion. This ena-
bles one to predict the final distribution of particles from the
knowledge of trajectories in the vicinity of the bifurcations.
Any cross streamline migration of inertial origin or hydrody-
namic shear induced diffusion would alter this behavior. In
symmetric bifurcation with equal widths of daughter branches
one would still expect equal flow and particle partitioning.
However, in asymmetric bifurcation the concentration profile
in the inlet branch would influence the distribution of particles
in the daughter branches.28–30 Lateral migration in the inlet
branch would cause the particle rich center of the inlet section
to divide unequally in the daughter branches of asymmetric
bifurcations. As reported by Xi and Shapley28 the low effi-
ciency of microfluidic devices used for separating the particles
from fluid could be due to nearly equal partitioning of the par-
ticles in the daughter branches of asymmetric channel even
though the fluid partitioning is unequal. The shear induced
redistribution of particles is believed to be the main reason for
this behavior. What is the effect of lateral migration on the
asymmetry of velocity and concentration profile near the bifur-
cation has not been reported so far in any experimental meas-
urements. The particle distribution in the inlet branch can be a
major factor in determining how the fluid and particles divide
2294 DOI 10.1002/aic Published on behalf of the AIChE
at the bifurcation. The past experimental works on shear
induced migration were conducted in straight micro-channels
but to the best of our knowledge the problem of particle migra-
tion in bifurcating micro-channels has not been addressed.
This has been the main motivation behind carrying out the pre-
sent experiments. An important issue related to particle migra-
tion that has been addressed in this work is the nature of fluid
and particle streamlines in the bifurcation region. Simulta-
neous measurement of the velocity and concentration profiles
has been carried out for the case of converging as well as
diverging flow to highlight the irreversibility of shear induced
migration. To achieve this we have conducted experiments
using a combination of PIV and PTV techniques. Another nov-
elty of our work is the use of single camera system as opposed
to two camera systems used in the previous works on two
phase flow measurements with PIV/PTV techniques. The
images of tracer and suspension particles were effectively sep-
arated using in-house developed MATLAB code for image
processing.
Experimental Method
The bifurcation geometry signifies a structure where parent
branch divides into two daughter branches with an angle (h).
We have conducted experiments to study velocity and par-
ticles concentration profiles in both symmetric and asymmetric
bifurcation channels and their sketch is shown in Figure 1.
The locations at which these profiles were measured are also
shown in this figure. Such bifurcation channels are commonly
used in several microfluidic devices. In a symmetric bifurca-
tion the two daughter branches make equal angle with the inlet
branch, whereas in asymmetric channel one of the side branch
bifurcates obliquely from the inlet branch. The symmetric
channels considered in our experiments were of two types: T-
shape, and Y-shape with 1208 bifurcation angle. The three dif-
ferent asymmetric channels had oblique side bifurcation with
angles of 908, 608, and 308 between the inlet and side branch.
The width of the daughter branches (500 lm) in all the chan-
nels was same as that of the parent branch and the length of
each section was equal to 7 cm. Microchannels of rectangular
cross section (500 lm 3 500 lm) with various branching
angles were fabricated from stainless steel by electric dis-
charge machine (EDM) cutting. After EDM cutting the top
and bottom surfaces of the microchannels were covered with
transparent glass plates.
Neutrally buoyant suspension having 5% volume concentra-
tion of PMMA particles in glycerol–water mixture were pre-
pared. A mixture of 74% glycerol and 26% water matched the
density of PMMA particles (1.18 gm/cc). Suspension of two
different sizes of particles (mean diameter 6 and 10 lm) were
prepared. To set up the flow, the inlet of the microchannel was
connected to the syringe pump through silicon pipe. The flow
visualization experiments were conducted for converging as
well as diverging flow conditions. In diverging flow the fluid
was pumped through one inlet and it exited from the two out-
lets, while in the case of converging flow the two bifurcating
branches were connected to the syringe pump and the fluid
exited from a single outlet. The speed of the pump could be
adjusted to get the desired flow.
Figure 2 shows the schematic diagram of our flow visualiza-
tion setup. It consists of an inverted epi-fluorescent micro-
scope (IX 83, Olympus), fluorescence illumination light
source (U-HGLGPS, Olympus) and a continuous acting
syringe pump (DUEL-NE-1000, New Era Pump System). The
June 2018 Vol. 64, No. 6 AIChE Journal
Figure 1. Schematic diagram of bifurcating channels used in our experiment: (a) symmetric Y-shape bifurcation, (b)
asymmetric bifurcation.
[Color figure can be viewed at wileyonlinelibrary.com]

test channel was placed on the inverted microscope and the
images were captured using a 12 bit, 1376 3 1040 pixel CCD
camera (Sensicam qe, PCO) and were transferred via frame-
grabbing board (PCI interface 520/525) to the computer. Two
hundred particle image frames were recorded for each set of
experiment at a particular flow condition. Our primary objec-
tive was to simultaneously measure the fluid and particle
phase velocities. To measure the fluid phase velocity very
small amount of tracer (0.025% by volume of 1lm polysty-
rene particles) was added in the suspending fluid. It was
desired that these tracer particles should move with the local
fluid velocity without disturbing the flow. Any shear induced
migration of these tracer particles is undesirable. Therefore,
tracer particles small enough to have fast response time (low
Stokes number) but large enough to produce good quality of
images should be chosen. Considering various optical con-
straints in any micro-PIV experiments, Lindken et al. recom-
mended the tracer particle size to be in the range of 0.5–1.5
lm, with particle image diameter equal to 2–3 pixels.31
Smaller particles (<0.5 lm) can exhibit Brownian motion and
generate measurement uncertainties, whereas larger particles
may not meet the criteria of low Stokes number. In our experi-
ments the maximum (centerline) velocity in the channel varied
from 0.0161 to 0.0284 cm/s. The flow Reynolds number was

Figure 2. Schematic diagram of experimental set up for micro particle image velocimetry.
[Color figure can be viewed at wileyonlinelibrary.com]

AIChE Journal June 2018 Vol. 64, No. 6 Published on behalf of the AIChE DOI 10.1002/aic 2295
O(1022) and for 1 lm tracers the Stokes number was found to
be O(1029). The particle Reynolds number and Stokes number
for the larger sized particles (6 and 10 lm) that make up the
bulk suspension phase was O(1024) and O(1027), respec-
tively. To capture the fluorescent images of particles all the
1lm fluid tracer and a small fraction (0.005% by volume) of
suspended particles (6 and 10 lm) were colored in a dye (rho-
damine 6G from Sigma Aldrich). These particles were first
dissolved in a solution of 0.06 g of rhodamine 6G in 200 mL
of pure ethanol. The solution was vigorously stirred at 358C
for about 45 min. This was done to ensure that the dye impreg-
nated the particles. The particles were dried before mixing in
the glycerol–water mixture. These fluorescently labeled micro-
spheres has an emission peak of about 570 nm. The emitted
light passes through the CCD camera via barrier filter, where
the scattered light is filtered out.

Post Processing of Images

In the PIV method one of the most important step is proc-
essing of the images in the required manner to get correct
velocity and concentration profile. In case of micro-particle
image velocimetry (l-PIV) the image filtering (removing
noise from the images) is required as the raw images suffer
from a range of imperfection and noises such as non-uniform
contrast and low frequency background noise. In the captured
images the larger suspension particles needed to be identified
so that it can be separated from the smaller fluid tracer par-
ticles to calculate particle velocity and concentration. Since
volume illumination is used in l-PIV experiments, separated
image also contains out-of-focus particles which gives rise to
wrong vectors during analysis. Particles which are more dis-
tant from the focal plane are more defocused. Various steps of
image processing to improve the accuracy of l-PIV and l-
PTV analysis are describe below.
Separation of large particles
There are many techniques which can be used to sepa-
rate particles based on size. “Dilation” and “Erosion” is
one of the most common methods which are being used.
One of the main disadvantages of this method is the diffi-
culty to define a structure element for spherical particles
with small radius. This problem can be overcome by
“Circular Hough Transform” which is a feature extraction
technique for detecting circles. Figure 3a illustrates the
working principle of Circular Hough Transform to locate
circles in defective image inputs. The circle entrants are
created by “voting” in the Hough parameter space and
thereafter selection of the local maxima is performed in a
so-called accumulator matrix.
In a two dimensional space a circle of radius “r” whose cen-
ter is located at (a, b) can be represented as:
x5a1rcos h (1)
y5b1rsin h (2)
When the angle h sails through the full 3608 range the points
(x, y) trace the periphery of a circle. If image of a particle is
made up of several pixels, some of which fall on the periph-
ery, one needs to find parameter triplets (a, b, r) to recreate a
circle representing the spherical particle. Since the radius of
suspended particles is known, the problem of finding parame-
ter triplet reduces to finding the locus of (a, b) which would
fall on a circle of known radius “r.” The true center point
Figure 3. (a) Representation of particle in image and
Hough space; (b) sample image of particles;
(c) image of 1lm tracers; (d) image of 10 lm
particles; (e) focused image of 10 lm par-
ticles.
[Color figure can be viewed at wileyonlinelibrary.com]
will be common to all parameter circles and can be found by
using “Hough Accumulation Array.” The first step is to
assign a radius range (rm < r < rh), where r is the radius of
actual large particle in the image. However, taking a radius
range instead of a constant radius increases the search time
but ensures calculating almost every circle. This is required
since in the captured images the size of any particle may vary
due to noise as well as its distance from the focal plane. By
choosing values of rm and rh carefully the defocused particles
can also be removed. Generally it is expected that radius of
defocused particles has greater radius than the focused parti-
cle. Therefore, greater the difference between the actual par-
ticle radius and image radius, greater the degree of
defocusing. rm and rh should be chosen such that particles
which appear of different radius due to noise should not be
missed and particles which are defocused should not be
selected. Once the center of every circle is known it is an
easy task to separate that circle from the image by knowing
the fact that every point in a square of side 2r need to be
selected with center of square being the same as the center of
circle and it should be checked whether the point is inside the
circle or not. After every large sized particle is separated
from the image the left out image would be that of 1 lm fluid
tracer particles.

2296 DOI 10.1002/aic Published on behalf of the AIChE June 2018 Vol. 64, No. 6 AIChE Journal
Removal of overlapped particles
A simple method has been used to remove overlapped par-
ticles. For any particle pair if the distance between their cen-
ters is less than the sum of their radius then only one particle
should be kept and the other one should be removed. Selection
of particle could be done by using the information of differ-
ence in actual and expected radius. Particles with large differ-
ence should be removed.
For n particles this method takes O(n2) time to check each
and every pair but this algorithm could be modified to improve
the search time to O(n), where n is the number of pixels in the
image. Starting the loop from the first pixel and for every cen-
ter of circle to be found, the neighbor pixel at a distant r from
that center is marked. In the next loop, the particle which
comes in the marked area should be removed by using the
information of difference between the actual radius and
expected radius.
x0 1m2
P P 
i5 x0 2m j5 y0 2n
2 2
r ðx0 ; y0 Þ5 sffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
x0 1m2 y0 1n2 
P P
I ði; jÞ2 Ib
i5 x0 2m2 j5 y0 2n2
I(i,j) is the brightness pattern of image and Im(i,j) is the
brightness of particle mask at pixel location (i,j); m and n are
the coefficients for interrogation area with value equal to 1.5r.
Ib and Ibm are the spatial average of respective brightness pattern
over the interrogation area.
It could be easily seen that co-relation coefficient does not
depend on peak brightness but only on the brightness pattern
that any particle has. This co-relation coefficient can be con-
verted into binary form by using some threshold to remove
undesired particles. In this study the value of this threshold is
taken as 0.7 as suggested by Takehara and and Etoh.33 The
defocused particles have less brightness, and therefore, can be
eliminated by using this threshold value.
A sample image recorded by the CCD camera is shown in
the Figure 3b. This image contains fluorescence tagged par-
ticles representing the 10 lm suspension particles and 1 lm
fluid tracer particles. We have separated the image shown in
Figure 3b based on the particle size using an in-house devel-
oped MATLAB code whose algorithm was described above.
Figures 3c, d displays the separated 1 and 10 lm particles,
respectively. The image shown in Figure 3d includes both
focused and defocused particles. Therefore, it requires further
processing to separate the focused and defocused particles.
Focused images of 10 lm particles extracted from the Figure
3d is shown in Figure 3e. Finally l-PTV analysis was carried
out for the images of larger and focused particles to get the
particle velocity and concentration field using PTV lab Beta
software34 and l-PIV analysis was performed for 1 lm tracer
image (Figure 3c) using PIVlab_1.32 software35 to measure
the fluid phase velocity.
In the PIV and PTV analysis we have used smoothing,
outlier-rejection, and interpolation filter to remove erroneous
vectors which show up due to poorly illuminated regions in
the image or any out-of-plane flow. PTV method directly
tracks single particles, whereas the PIV method uses the cross-
correlation analysis to derive the local motion statistically.
AIChE Journal June 2018 Vol. 64, No. 6
Filtering of particle based on particle mask co-relation
method
A particle mask is typically a type of brightness pattern
which is expected in the images of particles. In this study two
dimensional Gaussian distribution has been used for particle
mask as shown by Ushijima et al.32 The intensity of any point
in the image of a particle can be expressed as:
!
ðx2x0 Þ2 1 ðy2y0 Þ2
I ðx; yÞ5 a exp 2 (3)
2r2
In the above expression a is the peak intensity, r is the repre-
sentative radius, and (x0, y0) is the location of the center of the
particle. In the calculation of co-relation value particle mask is
kept fixed. As shown by Takehara and Etoh.33 co-relation
coefficient can be calculated as:
y0 1n2  
I ði; jÞ2 Ib Im ði; jÞ2 Ibm
ffisffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi (4)
2 xP 012
m
y0 1n2  2
P
Im ði; jÞ2 Ibm
i5 x0 2m2 j5 y0 2n2
Therefore, the possible accuracy of the PTV algorithm is given
by the accuracy of the detection of particle positions, whereas
the PIV method is limited by the accuracy of locating the cor-
relation peak. For the PIV analysis the images are divided into
grid known as interrogation windows (IW). The size of the IW
should be small enough for good spatial resolution of velocity
vectors, but large enough so that it contains sufficient number
of particles. In our PIV analysis the IW size was 32 3 32 pix-
els with 75% overlap. The fluid velocity vector maps for sus-
pension of 10 lm particles in diverging flow and converging
flow for symmetric T-shape and asymmetric (h 5 608) bifurca-
tion channels are shown in the Figure 4. Since both l-PIV and
l-PTV are optical flow diagnostic techniques, measurement
uncertainties are associate with the images. Thickness of the
measurement plane is one of the important parameter which is
normally expressed in terms of DOC (depth of correlation).
The DOC depends on the depth of field of the objective and on
the size of the particles. In our experiments DOC was different
for 1 and 10 lm particles. Because of varying particle sizes
present in the same medium, it is impossible to simultaneously
focus all of them. The thickness of the measurement problem
was minimized by focusing more on the larger particles. Since
the number of smaller tracer particles were much more than
the larger particles, some defocused smaller particles should
not affect the result significantly.
Results and Discussion
Velocity field
First we report the results from the diverging flow experi-
ments in which the fluid coming from a single inlet bifurcates
into two daughter branches. The mean velocity profiles for
pure suspending fluid and suspensions of two different particle
sizes (6 and 10 lm) at various locations in the inlet section of
the bifurcating channels are plotted in the Figure 5. The veloc-
ity magnitude (U) is normalized relative to the maximum
Published on behalf of the AIChE DOI 10.1002/aic 2297
Figure 4. Velocity vector map of suspending fluid phase for 10 lm suspension in symmetric T-shape and asymmet-
ric (h 5 608) bifurcation channels for diverging flow (a, b) and converging flow (c, d).
[Color figure can be viewed at wileyonlinelibrary.com]

velocity (Umax) for each case. Figure 5a shows the velocity
profiles for converging flow in the symmetric T-bifurcation
channel at three positions in the inlet branch (locations 1, 2,
and 3). These locations can be identified in the schematic dia-
grams of the channels. At location 2 and 3 the velocity profiles
are symmetric and fully developed. It was observed that at
location 3, the velocity profiles in the inlet branch are similar
for the suspension as well as suspending fluid and close to the
parabolic profile. This observation was same for all the bifur-
cation channels. Near the bifurcation the nature of velocity
profile depends on the bifurcation angle.
Figure 5b shows the velocity profile at location 4 (bifurca-
tion position) for symmetric T-shaped, and Figure 5c shows
the profile for the symmetric Y-shaped bifurcation. Since
beyond the location 4 the flow has to divide into left and right
branches the profile near the center becomes smeared for pure
fluid. On the other hand we observed peak-valley-peak pattern
for suspension flow. This kind of profile is an indication of
deceleration of particles that are near the center of the channel.
It is expected that migration will cause the particles to concen-
trate in the center increasing the local viscosity there, which in
turn retards the flow. This creates the peak-valley-peak type of
profile. The retardation of velocity near the center is more for
the suspension of 10 lm particles compared to that of 6 lm.
This is due to relatively higher migration for larger sized
particles. The velocity profiles at the location 4 in the asym-
metric bifurcating channels are shown in the Figure 6. Unlike
the symmetric channels no peak-valley-peak pattern was
observed. In all the three asymmetric channels a small shift in
the direction of the side branch was observed for both pure
fluid as well as suspension flow. The skewness coefficients for
908, 608, and 308 channels were found to be 0.15, 0.13, and
0.09, respectively. It can be observed that as the bifurcation
angle decreases the velocity profile not only become less
blunted but the skewness toward the side branch also
decreases. After the bifurcation the flow enters into the daugh-
ter branches. The velocity profiles at the entrance and exit
locations of the daughter branches of two symmetric (T- and
Y-shape) and two asymmetric bifurcation (908 and 308) chan-
nels are shown in the Figure 7. For the clarity of profiles in the
two daughter branches we have shown the outer wall of the
left branch at x/B 5 21, and that of the right branch at
x/B 5 11. It can be observed that in symmetric channels
(Figures 7a, b) the velocity profile at location 5 in the left
branch is almost mirror image of the profile at location 8 of
the right branch. In both the daughter branches of T- and Y-
bifurcation the peak in velocity profiles are shifted toward the
outer walls. The skewness coefficient for T- and Y-bifurcation
was found to be 0.35 and 0.21, respectively. For asymmetric
channel (h 5 908) the velocity profile (Figure 7c) in the main

2298 DOI 10.1002/aic Published on behalf of the AIChE June 2018 Vol. 64, No. 6 AIChE Journal
Figure 5. Velocity profiles at locations 1, 2, and 3 in the inlet branch (a), and at the bifurcation (location 4) in sym-
metric bifurcation channels of T-shape (b) and Y-shape (c).
[Color figure can be viewed at wileyonlinelibrary.com]

Figure 6. Velocity profiles at the location 4 in the asymmetric bifurcation channels: (a) h 5 908, (b) h 5 608, (c) h 5 308.
[Color figure can be viewed at wileyonlinelibrary.com]

AIChE Journal June 2018 Vol. 64, No. 6 Published on behalf of the AIChE DOI 10.1002/aic 2299
Figure 7. Velocity profiles at the beginning (locations 5 and 8) and exit of the daughter branches (locations 7 and
10) for diverging flow in bifurcation channels: (a) T-Shape; (b) Y-Shape; (c) h 5 908; (d) h 5 308.
[Color figure can be viewed at wileyonlinelibrary.com]

Figure 8. Velocity profiles at the end of inlet branches (locations 5 and 8) in converging flow through bifurcation
channels: (a) symmetric T-shape; (b) symmetric Y-shape; (c) asymmetric (h 5 908); (d) asymmetric (h 5 308).
[Color figure can be viewed at wileyonlinelibrary.com]

2300 DOI 10.1002/aic Published on behalf of the AIChE June 2018 Vol. 64, No. 6 AIChE Journal
Figure 9. Velocity profiles at the confluence (location 4) and exit (location 0) of the outlet branch in converging
flow through bifurcation channels: (a) location 4 of symmetric T-shape, (b) location 4 of asymmetric (308)
bifurcation, (c) exit of symmetric T-shape, and (d) exit of asymmetric (308) bifurcation.
[Color figure can be viewed at wileyonlinelibrary.com]

Figure 10. Streamlines of fluid phase velocity (obtained from PIV analysis of tracer particles) and the particle phase
velocity (obtained from PTV analysis of suspended particle of 10 lm size) in various channels for diverging
flow: (a) symmetric T-shape, (b) symmetric Y-shape; asymmetric channels: (c) h 5 908, (d) h 5 608, (e) h 5 308.
[Color figure can be viewed at wileyonlinelibrary.com]

AIChE Journal June 2018 Vol. 64, No. 6 Published on behalf of the AIChE DOI 10.1002/aic 2301
Figure 11. Streamlines of fluid and particle phase velocity in various channels for converging flow: (a) symmetric
T-shape, (b) symmetric Y-shape; asymmetric channels: (c) h 5 908, (d) h 5 608, (e) h 5 308.
[Color figure can be viewed at wileyonlinelibrary.com]

branch is less skewed (skewness coefficient 5 0.23) compared
to the side branch (skewness coefficient 5 0.40), i.e., a shift
toward the outer wall is observed. The skewness coefficients
for 308 asymmetric channels (Figure 7d) in the main and side
branch were found to be 0.11 and 0.27, respectively. The pro-
files were much less skewed for 608 asymmetric bifurcation
(not shown here) and the skewness coefficient for main and
side branch was measured to be 0.06 and 0.13, respectively.
As we move to further downstream locations the skewness in
the velocity profile decreases. The locations 7 and 10 are near
the outlet of daughter branches and as expected the profiles
are symmetric at these locations for all the channels.
Next we have performed experiments for converging flow
conditions. In these experiments the two outlet sections of the
diverging flow experiments were made as inlets and after the
confluence the flow exited from a single outlet. The velocity

Figure 12. Velocity vector map of particle phase (suspension of 10 lm particles) in the various bifurcation channels
for diverging flow: (a) symmetric T-shape, (b) symmetric Y-shape; asymmetric channels: (c) h 5 908, (d)
h 5 608, (e) h 5 308.
[Color figure can be viewed at wileyonlinelibrary.com]

2302 DOI 10.1002/aic Published on behalf of the AIChE June 2018 Vol. 64, No. 6 AIChE Journal
Figure 13. Velocity vector map of particle phase (suspension of 10 lm particles) in the various bifurcation channels
for converging flow: (a) symmetric T-shape, (b) symmetric Y-shape; asymmetric channels: (c) h 5 908, (d)
h 5 608, (e) h 5 308.
[Color figure can be viewed at wileyonlinelibrary.com]

profiles at the end of the inlet branches are shown in Figure 8.

It was observed that in both the symmetric T- and Y-
bifurcation the velocity profiles are skewed toward the outlet
section and resemble with the profiles of the corresponding
diverging flow case. For asymmetric channels the velocity
profile is asymmetric in the side inlet branch but nearly sym-
metric in the main inlet branch. The asymmetry of velocity
profile is more pronounced in 908 channel compared to 308.
Figure 9 shows the velocity profiles at the confluence of
both the inlet streams, which is also the beginning of the outlet
section (location 4) and exit of the outlet branch (location 0).
The profiles are symmetric but slightly blunted at location 4 of
the symmetric T-shaped channel (Figures 9a). However,
unlike the diverging flow no peak-valley-peak type of profile
exists. In the case of the asymmetric 308 channel (Figure 9b)
the profiles are skewed toward left. The skewness coefficient
for 908 (profile not shown here) and 308 channel was measured
to be 0.04 and 0.24, respectively. The decrease in the skew-
ness with bifurcation angle is expected since for asymmetric
channel the ratio of side branch to main branch opening in the
confluence region decreases with the bifurcation angle. In the
Figures 9c, d which show the profiles at the exit locations we
notice that the velocity profiles are again symmetric and
resembles the fully developed parabolic profile of the diverg-
ing flow case.
Streamlines can be constructed from the velocity fields
obtained from the PIV and PTV analysis. In our study the
streamlines for the carrier fluid was obtained from the velocity Figure 14. (a) Sample image for determining the con-
vectors of 1 lm fluid tracer particles using the cross- centration of particles, (b) schematic repre-
correlation PIV analysis. The velocity vectors from the PTV sentation of 11 spatial bins (B1 to B11)
analysis were used to construct the streamlines for the particle created across the width of the channel to
phase. Since the positions of the velocity vectors from PIV count the number of particles in each bin,
and PTV analysis were not same, the vectors from PTV analy- (c) profiles of the particle fraction at the
sis were interpolated over the same grids that was used in the inlet sections of symmetric T-shape channel
PIV analysis. Figure 10 shows these streamlines for suspen- in diverging flow.
sion of 10 lm particles for diverging flow in various [Color figure can be viewed at wileyonlinelibrary.com]

AIChE Journal June 2018 Vol. 64, No. 6 Published on behalf of the AIChE DOI 10.1002/aic 2303
Figure 15. Profiles of the particle fraction in the left and right branches in diverging flow through bifurcation chan-
nels: (a) T-shape, (b) Y-shape, (c) h 5 908, (d) h 5 308.
[Color figure can be viewed at wileyonlinelibrary.com]
bifurcation channels. The streamlines for converging flow con-
ditions are shown in the Figure 11. The streamlines shown in
red and blue color are for the carrier fluid and suspension par-
ticles, respectively. From the knowledge of the position of
dividing streamlines the fluid and particle partitioning in the
daughter branches can be known. For symmetric T- and Y-
shape the position of the dividing streamline is almost at the
center of the channel indicating equal flow and particle parti-
tioning. The fluid and particle streamlines in the inlet section
are nearly same but differ significantly in the bifurcation
region. This shows that the larger particles which make the
bulk suspension do not follow the fluid motion in the bifurca-
tion region. The position of the dividing streamlines for both
the fluid and particle phase was shifted more toward main
branch in case of 908 and 608 asymmetric bifurcations. On the
other hand for 308 bifurcation channel it is shifted more
toward the side branch. It should be noted that the width of the
inlet, main and side branches in all the bifurcation channels
were the same. Therefore, the junction part leading to side
branch have unequal opening for the channels of 608 and 308
angle. As the bifurcation angle decreases the ratio of side
branch to main branch opening for the flow increases. For 308
bifurcation the opening for the flow is more compared to
908and 608 bifurcation channels. This is why both the fluid
and particle phase streamlines are shifted more toward the side
branch. Similar to the diverging flow, significant deviation in
the fluid phase and particle phase velocities are also observed
near the bifurcation for converging flow.
2304 DOI 10.1002/aic Published on behalf of the AIChE
Particle migration
The l-PTV analysis was based on the tracking of particles
that make up the bulk suspension. The exact locations of a par-
ticle in two consecutive frames is used to obtain the velocity
vector. Figure 12 shows velocity vector map of 10 lm par-
ticles in all the five bifurcation channels for diverging flow.
For the clarity of vectors we have shown only a small region
near the bifurcation. From the vector map it can be clearly
observed that more number of particles are located at the cen-
ter of the channel in the inlet branch. The concentration of par-
ticles near the center of the channel is due to the particle
migration. It can also be observed that the particle concentra-
tion is relatively higher in the region close to the inner walls
of the daughter branches. This is due to the fact that the parti-
cle rich central core of the inlet section moves toward the
inner wall of the side branches. The particle velocity vector
map for converging flow case is shown in the Figure 13. We
notice that for symmetric T- and Y-shape channel the two
streams of particles coming from the inlet branch merge in
such a way that the maximum concentration is shifted toward
either side of the channel center near the entrance of the outlet
section. This gives rise to peak-valley-peak pattern in concen-
tration profile which will be discussed next.
The qualitative measurement of particle concentration was
carried out using a particle counting program written in MAT-
LAB. Figure 14a shows a sample raw image of 10 lm par-
ticles distributed across the width of the channel. The
June 2018 Vol. 64, No. 6 AIChE Journal
Figure 16. Profiles of the particle fraction at the end of inlet section (location 5 and location 8) of bifurcation chan-
nels for converging flow: (a) T-shape, (b) Y-shape, (c) h 5 908, (d) h 5 308.
[Color figure can be viewed at wileyonlinelibrary.com]
rectangular section of the images that were chosen for the par-
ticle counting measured 380 pixels in the velocity gradient
direction and 65 pixels in the flow direction. To count the par-
ticles, the whole image was divided into 11 bins of equal
width as shown in the Figure 14b. The relative concentration
of particles in each bin can be represented as Np/NT where, Np
denotes the number of particles in a bin and NT is the total
number of particles present in the whole image. The average
value of Np/NT was obtained by performing average over 100
consecutive image frames. The normalized particle count in
the inlet section of symmetric T-channel is shown in Figure
14c. The particles move from wall (higher shear rate) to the
center of the channel (lower shear rate) which is clear from
inverted V-shape of the profile. As we move from location 1
to location 4, the peak concentration also increases. The parti-
cle concentration profiles at various locations in the inlet chan-
nel indicates that even though the profiles at locations 3 and 4
are close to each other, fully developed concentration profiles
may not have been achieved. Figure 15 shows the normalized
particle count in the daughter branches of the symmetric and
asymmetric bifurcation channels. We would like to mention
that for the clarity of profiles x/B 5 21 represents inner wall
of the left branch and x/B 5 11 represents the inner wall of
the right branch. At location 5 of the left branch and the corre-
sponding location 8 of the right branch, the particle concentra-
tion is highest near the inner wall. The profile of the left
branch is similar to the right branch for all the locations. As
we move to the downstream locations the particle concentra-
tion near the wall gradually decreases and due to particle
AIChE Journal June 2018 Vol. 64, No. 6 Published on behalf of the AIChE
migration the peak concentration shifts to the center of the
channel. Similar behavior was also observed for asymmetric
bifurcations except for the fact that the profile at any location
in the main branch was not the same as that of the right
branch. For 908 bifurcations the concentration near the inner
wall of the left branch is slightly more compared to that of the
right branch. However, for 308 channel reverse was observed.
Figure 16 shows the comparative particle concentration pro-
file at the end of both the inlet sections in converging flow.
We observe that at these locations the profiles in both the inlet
sections shift toward the outlet direction. The skewness in con-
centration profile is more for the symmetric channels com-
pared to the asymmetric channels. We have also observed that
for asymmetric channel (h5308) the concentration profile
(Figure 16d) is symmetric and identical for both the inlets.
The position of dividing streamlines (Figure 10) can be used
to compute the flow partitioning; similarly from the particle
count fractional flow of particles in daughter branch can be
evaluated. For asymmetric channel (h 5 908) the fraction of
particles and fluid entering the main branch was calculated as
0.576 and 0.50, respectively. This in qualitative agreement
with the observations of Koh and Kim.36
Figure 17 shows the concentration profiles at various loca-
tions in the outlet section of symmetric and asymmetric chan-
nels for converging flow condition. At location 4 and 2 we
observe peak-valley-peak pattern in all the channels. As we
move further downstream locations in the outlet section, the
peak-valley-peak pattern in concentration profile gradually
vanishes and only a single peak emerges. At location 1 and 0
DOI 10.1002/aic 2305
Figure 17. Profiles of the particle fraction at different locations of the outlet section in converging flow through
bifurcation channels: (a) T-shape, (b) Y-shape, (c) h 5 908, (d) h 5 308.
[Color figure can be viewed at wileyonlinelibrary.com]
the profiles are fully symmetric with maximum concentration
at the center of the channel. These results clearly establish that
due to irreversible nature of particle migration the profiles
near the bifurcation region are not the same for the converging
and diverging flow conditions.
Conclusion
Simultaneous measurements of fluid and particle phase
velocities were carried out for pressure driven flow of non-
colloidal neutrally buoyant suspension in bifurcating micro-
channels. A combination of l-PIV and l-PTV was used to
study the velocity and concentration profiles in diverging and
converging flow conditions. It was observed that in the inlet
branch the particles migrate toward the center of the channel.
This gives rise to peak-valley-peak type of velocity profile
near the bifurcation for diverging flow in symmetric channels.
In the daughter branches the peak in the velocity profile shifts
toward the outer wall whereas the maxima in particle concen-
tration was observed near the inner wall. This causes asymme-
try in the velocity and concentration profiles in the daughter
branches. As we move toward the downstream positions the
maxima in velocity and concentration profile again shifts
toward the center of the channel. For converging flow condi-
tions we have observed peak-valley-peak pattern in concentra-
tion profile near the confluence region. The velocity and
concentration profiles were found to be strongly influenced by
the bifurcation geometry. The comparison of the streamlines
2306 DOI 10.1002/aic Published on behalf of the AIChE
for the fluid and particle phase showed significant deviation
near the bifurcation region. Due to irreversible nature of parti-
cle migration the profiles were not the same near the bifurca-
tion/confluence for diverging/converging flow conditions.
Future experiments at higher particle concentration is expected
to give more insight into the role of particle migration in the
flow and particle partitioning in the asymmetric bifurcation
channels.
Acknowledgments
The authors would like to thankfully acknowledge funding
from the Department of Science and Technology, India (Pro-
ject no. SB/S3/CE/088/2013).
Literature Cited
1. Segre G, Silberberg A. Radial particle displacements in Poiseuille
flow of suspensions. Nat Commun. 1961;189(4760):209–210.
2. Han M, Kim C, Kim M, Lee S. Particle migration in tube flow of
suspensions. J Rheol. 1999;43(5):1157–1174.
3. Leighton D, Acrivos A. Measurement of shear-induced self-diffusion
in concentrated suspensions of spheres. J Fluid Mech. 1987;177(1):
109–131.
4. Frank M, Anderson D, Weeks ER, Morris JF. Particle migration in
pressure-driven flow of a Brownian suspension. J Fluid Mech. 2003;
493:363–378.
5. Gao C, Xu B, Gilchrist JF. Mixing and segregation of microspheres
in microchannel flows of mono and bi-dispersed suspensions. Phys
Rev E. 2009;79(3 Pt 2):036311–036317.
June 2018 Vol. 64, No. 6 AIChE Journal
 6. Nott PR, Brady JF. Pressure-driven flow of suspensions: simulation
and theory. J Fluid Mech. 1994;275(1):157–199.
7. Kang K, Lee SS, Hyun K, Lee SJ, Kim JM. DNA-based highly tun-
able particle focuser. Nat Commun. 2013;4:1–8.
8. Kim YW, Yoo JY. The lateral migration of neutrally-buoyant spheres
transported through square. J Micromech Microeng. 2008;18:1–13.
9. Kim YW, Yoo JY. Two-phase flow laden with spherical particles in
a microcapillary. Int J Multiph Flow. 2010;36(6):460–466.
10. Leble V, Lima R, Dias R, Fernandes C, Ishikawa T, Imai Y, Yamaguchi
T. Asymmetry of red blood cell motions in a microchannel with a diverg-
ing and converging bifurcation. Biomicrofluidics. 2011;5(4):1–15.
11. Wang T, Rongin U, Xing Z. A micro-scale simulation of red blood
cell passage through symmetric and asymmetric bifurcated vessels.
Sci Rep. 2016;6(1):1–13.
12. Sherwood JM, Kaliviotis E, Dusting J, Balabani S. Hematocrit, vis-
cosity and velocity distributions of aggregating and non-aggregating
blood in a bifurcating microchannel. Biomech Model Mechanobiol.
2014;13(2):259–273.
13. Svanes K, Zweifach BW. Variations in small blood vessel hematoc-
rits produced in hypothermic rats by micro-occlusion. Microvasc
Res. 1968;1(2):210–220.
14. Fung YC. Stochastic flow in capillary blood vessels. Microvasc Res.
1973;5(1):34–48.
15. Roberts BW, Olbricht WL. Flow-induced particulate separations.
AIChE J. 2003;49(11):2842–2849.
16. Roberts BW, Olbricht WL. The distribution of freely suspended par-
ticles at microfluidic bifurcations. AIChE J. 2006;52(1):199–206.
17. Davis JM, Pozrikidis C. Hydrodynamic instability of a suspension of
spherical particles through a branching network of circular tubes.
Acta Mech. 2012;223(3):529–540.
18. Aarts PA, Van den Broek SA, Prins GW, Kuiken GD, Sixma JJ,
Heethaar RM. Blood platelets are concentrated near the wall and red
blood cells in the center in flowing blood. Arteriosclerosis. 1988;
8(6):819–824.
19. Rong FW, Carr RR. Dye studies on flow through branching tubes.
Microvasc Res. 1990;39(2):186–202.
20. Wang T, Xing Z. Characterization of blood flow in capillaries by
numerical simulation. J Modern Phys. 2010;01(06):349–356.
21. Ishikawa T, Fujiwara H, Matsuki N, Yoshimoto T, Imai Y, Ueno H,
Yamaguchi T. Asymmetry of blood flow and cancer cell adhesion in
a microchannel with symmetric bifurcation and confluence. Biomed
Microdevices. 2011;13(1):159–167.
22. Van Dinther AMC, Schroen CGPH, Vergeldt FJ, Van Der Sman
RGM, Boom RM. Suspension flow in microfluidic devices: a review
AIChE Journal June 2018 Vol. 64, No. 6 Published on behalf of the AIChE
of experimental techniques focussing on concentration and velocity
gradients. Adv Colloid Interface Sci. 2012; 173:23–34.
23. Lima R, Ishikawa T, Imai Y, Takeda M, Wada S, Yamaguchi T.
Measurement of individual red blood cell motions under high hemat-
ocrit conditions using a confocal micro-PTV system. Ann Biomed
Eng. 2009;37(8):1546–1559.
24. Anastasiou AD, Spyrogianni AS, Koskinas KC, Giannoglou GD,
Paras SV. Experimental investigation of the flow of a blood ana-
logue fluid in a replica of a bifurcated small artery. Med Eng Phys.
2012;34(2):211–218.
25. Balan CM, Broboana D, Balan C. Investigations of vortex formation
in microbifurcations. Microfluid Nanofluid. 2012;13:819–833.
26. Korevaar MW, Padding JT, Deen NG, Kuipers JAM, Wang J, Wit
M, Schutyser MAI. Hybrid PIV/PTV measurements of velocity and
position distributions of gas-conveyed particles in small, narrow
channels. AIChE J. 2015;61(11):3616–3627.
27. Ha H, Nam KH, Lee SJ. Hybrid PIV-PTV technique for measuring
blood flow in rat mesenteric vessels. Microvasc Res. 2012;84(3):
242–248.
28. Xi C, Shapley NC. Flows of concentrated suspensions through an
asymmetric bifurcation. J Rheol. 2008;52(2):625–646.
29. Yezaz Ahmed GM, Singh A. Numerical simulation of particle
migration in asymmetric bifurcation channel. J Non-Newtonian Fluid
Mech. 2011;166(1–2):42–51.
30. Reddy MM, Singh A. Flow of concentrated suspension through obli-
que bifurcating channels. AIChE J. 2014;60(7):2692–2704.
31. Lindken R, Rossi M, Große S, Westerweel J. Micro-particle image
velocimetry (mPIV): recent developments, applications, and guide-
lines. Lab Chip. 2009;9(17):2551–2567.
32. Ushijima S. Particle tracking velocimetry using laser beam scanning.
J Hydraul Coastal Environ Eng. 1996;35:99–107.
33. Takehara K, Etoh T. A study on particle identification in PTV parti-
cle mask correlation method. J Vis. 1999;1(3):313–323.
34. Thielicke W, Stamhuis EJ. PIVlab—towards user-friendly, affordable
and accurate digital particle image velocimetry in MATLAB.
J Open Res Soft. 2014;2:1–10.
35. Brevis W, Ni~ no Y, Jirka GH. Integrating cross-correlation and relax-
ation algorithms for particle tracking velocimetry. Exp Fluids. 2011;
50(1):135–147.
36. Koh MG, Kim C. Separation of particles of different sizes from non-
Newtonian suspension by using branched capillaries. J Chem Eng
Japan. 2007;40(11):964–972.
Manuscript received June 25, 2017, and revision received Dec. 16, 2017.
DOI 10.1002/aic 2307