Wyss 2006

Download as pdf or txt
Download as pdf or txt
You are on page 1of 4

PHYSICAL REVIEW E 74, 061402 共2006兲

Mechanism for clogging of microchannels

Hans M. Wyss,1 Daniel L. Blair,1 Jeffrey F. Morris,2 Howard A. Stone,3 and David A. Weitz1
1
Department of Physics & DEAS, Harvard University, Cambridge, Massachusetts 02138, USA
2
Levich Institute at the City University of New York, New York, New York 10031, USA
3
DEAS, Harvard University, Cambridge, Massachusetts 02138, USA
共Received 18 April 2006; published 11 December 2006兲
We investigate clogging of microchannels at the single-pore level using microfluidic devices as model
porous media. The process of clogging is studied at low volume fractions and high flow rates, a technologically
important regime. We show that clogging is independent of particle flow rate and volume fraction, indicating
that collective effects do not play an important role. Instead, the average number of particles that can pass
through a pore before it clogs scales with the ratio of pore to particle size. We present a simple model that
accounts for the data.

DOI: 10.1103/PhysRevE.74.061402 PACS number共s兲: 82.70.⫺y, 47.55.Kf, 47.56.⫹r, 47.61.Jd

The extraction of solids from fluids is a common and with the ratio of the pore to particle size and we present a
important process in industry and technology; deep-bed fil- simple model that correctly accounts for this scaling.
tration in oil recovery, ground water treatment, catalysis, and We flow particle suspensions through a microfluidic de-
liquid chromatography all depend on the capture of solid vice that consists of a single wide channel followed by an
particles inside a porous medium. However, a serious limita- array of parallel, narrow channels 关see Fig. 1共a兲兴; this design
tion to the performance of all filters is ultimately the clog- allows us to investigate a large number of clogging events
ging of the porous medium. Clogging can also lead to cata- simultaneously. Each channel consists of a set of constric-
strophic results in technologies which rely on the complete tions along their length. The height of the channel is held
suppression of any kind of blockage, such as ink-jet printing constant and is the same as that of the large incoming chan-
or any microfluidic system where particles are present either nel. We control the flow rate of the suspension in the chan-
intentionally or unintentionally in the form of dust or other nels by setting the pressure difference of the fluid across the
contaminants. However, despite their importance, the physi- channels. This approach ensures that the flow rate in any
cal mechanisms that lead to clogging of porous media are given channel does not depend on the state of the other chan-
still not well understood. Physically, filters are porous media nels. Clogging events are independently characterized under
with a distribution of pore sizes that determines the smallest identical conditions, allowing us to acquire statistical infor-
filtrate to be removed from suspension. The simplest possible mation about the random clogging process. The ratio of the
picture of clogging is one of pure size exclusion, where pore to particle size is varied over an order of magnitude: the
blockage events occur only if particles enter a pore that is width of the pores is varied over tens of micrometers and the
smaller than their own diameter. However, in many instances particle diameter is on the order of micrometers. Controlled
clogging is observed even when the pores are much larger and well-defined pore structures are produced by fabricating
than the suspended particles. poly-共dimethylsiloxane兲 共PDMS兲 microfluidic devices using
The process of clogging is exceptionally complicated and soft lithography 关4,5兴. Our devices have four PDMS walls
not well understood. Model descriptions of clogging often and are fabricated without plasma activation to ensure the
focus on different length scales than do typical experimental surface wetting properties remain constant in time 关6兴. We
investigations. However, a sound theoretical description of use aqueous suspensions of uniform polystyrene spheres that
clogging ultimately must be based on an understanding of are electrosterically stabilized by sulfate and carboxyl sur-
the blockage mechanism at the level of a single pore. Unfor- face groups 共IDC兲.
tunately, experimental studies typically focus on macro- We use digital video microscopy to record movies of the
scopic clogging in realistic porous materials such as packed microfluidic channels and measure the dynamics of clogging
grains 关1,2兴, or porous membranes 关3兴. Although these ex- at the single-pore level. The typical transit time of a particle
periments provide important insight into the dynamics of through the entire microfluidic device is ⬃0.1 s. Thus, at
clogging in the pore network as a whole, clogging on the standard video rates, we cannot image individual particles
single-pore level cannot be accessed. Without experimental passing through pores. Instead, we capture images of the
data on clogging at the single-pore level, a full interpretation average particle concentration. The intensity of a point in the
of clogging phenomena in macroscopic porous media will image reflects the average local number concentration of par-
remain elusive. Thus, experimental measurements of clog- ticles; the darker the region, the higher the local particle
ging at the single-pore scale are essential. concentration. A typical experimental image is shown in Fig.
In this article, we report experimental results of filtration 1共a兲. The gray regions on the left correspond to a volume
and clogging at the single pore level. We observe that the fraction of ␾ = 4 ⫻ 10−2, while the lightest regions are either
formation of clogs is determined solely by a critical number PDMS walls or pure water. The regions that appear darkest
of particles that flow through single pores irrespective of are areas of highly concentrated particles near a clog. A time
both flow rate and particle volume fraction. Moreover, we series of the clogging process for three channels is shown in
show that this critical number scales in an unexpected way Fig. 1共b兲. The darkest shades of gray 共red online兲 indicate

1539-3755/2006/74共6兲/061402共4兲 061402-1 ©2006 The American Physical Society


WYSS et al. PHYSICAL REVIEW E 74, 061402 共2006兲

come clogged at their inlets and the central channel is still


open to particles and fluid. In the last frame, t = 38 s, all three
channels are clogged and a region of densely packed par-
ticles spans their inlets. After each clogging event, the poros-
ity of the clog remains nonzero and therefore fluid continues
to flow. As a consequence, a “filter cake” of packed particles
builds up on the upstream side of the clog location.
To quantify the clogging process, we measure the ratio of
clogged to open channels as a function of time 关see Fig.
1共c兲兴. For these data, we fix the pore geometry and volume
fraction. The diamonds correspond to the experiment dis-
played in Fig. 1共a兲. At higher pressures clogging occurs
faster, as shown by the circles. We normalize the data by the
ratio of the applied pressure to the lowest applied pressure,
⌬p / ⌬p0, where ⌬p0 = 2.0 pounds per square inch 共psi兲. This
rescaling eliminates the dependence on flow rate, as shown
in Fig. 1共c兲, inset, indicating that a fixed volume of fluid
must pass though the pore prior to clogging.
One possible mechanism for clogging is the random oc-
currence of a local particle concentration large enough to
result in a physical blocking of pores 关7兴 or arch formation
within a channel 关8兴, even though the typical number of par-
ticles in a pore is very small; for example, at ␾ = 0.01, there
are typically only ⬃4 particles passing through a pore of
cross section W ⫻ H = 25⫻ 20 ␮m2 at any given moment. In
a stable suspension, fluctuations in the particle concentration,
at length scales comparable to pore sizes, will be determined
by Poisson statistics. As the volume fraction is increased, it
becomes more probable that a concentration fluctuation will
be large enough to fill a pore with particles, thus blocking it.
As a consequence, clogging should be strongly dependent on
␾.
To test the dependence on such local concentration fluc-
tuations, we vary the volume fraction and measure the
change in the average clogging time t* from the curves in
Fig. 1共c兲, while holding all other experimental parameters
fixed. We normalize t* by the relative applied pressure and
plot the result as a function of ␾ 共in Fig. 2兲. As ␾ varies
between 4 ⫻ 10−4 and 4 ⫻ 10−2, t* varies by two orders of
FIG. 1. 共Color online兲 共a兲 Microscope image of a microfluidic
magnitude from seconds to hours. Within experimental error,
device during a clogging experiment. A particle suspension of vol- t* is not only inversely proportional to the flow rate Q of
ume fraction ␾ = 4 ⫻ 10−2 is pumped from left to right through a suspension flowing through a channel, but also on the par-
series of channels in the device. The pressure difference between ticle volume fraction ␾:
inlet and outlet is held constant at ⌬p = 2 psi; the total length of the
channels is 1.2 mm. The particle diameter is D = 2.9 ␮m; the height 1
of the device is H = 25 ␮m. Inset: close-up of the channel exit zone; t* ⬀ . 共1兲
Q␾
the smallest width at the bottleneck of each channel is W = 20 ␮m.
共b兲 Close-up of three channels in the device as a time series, While the clogging time is strongly ␾ dependent, the ef-
t = 0 s , 8 s , 38 s. 共c兲 Dependence on applied pressure for the same fects of fluctuations will be most apparent in the total num-
geometry. The ratio of clogged channels as a function of time at ber of particles that must flow through a channel prior to
different applied pressures. ⌬p = 2 psi 共red diamonds, results of ex- clogging,
periment shown on top兲; ⌬p = 4 psi 共blue circles兲. Inset: plot of the
same data with time normalized by the relative pressure difference N* ⯝ t*Q␾/V p . 共2兲
⌬p / ⌬p0 共⌬p0 = 2 psi兲.

regions where particles are densely packed, whereas particle- Due to the inverse dependence of t* on Q and ␾, N* is
free fluid appears light gray 共yellow online兲; the channel side independent of ␾. Thus, fluctuations due to Poisson statistics
walls are shown in white. At t = 0 s the areas inside the chan- cannot describe the clogging mechanism since the marked
nels appear evenly gray, demonstrating that particles are ho- insensitivity of N* to both ␾ and Q indicates that pore clog-
mogeneously distributed. At t = 8 s two channels have be- ging must be a single-particle effect.

061402-2
MECHANISM FOR CLOGGING OF MICROCHANNELS PHYSICAL REVIEW E 74, 061402 共2006兲

FIG. 2. 共Color online兲 Dependence of the average clogging time


t* and the number N* of particles required to clog a channel on the
particle volume fraction ␾. Clogging times of experiments taken at
pressure differences ⌬p ⫽ ⌬p0 = 2 psi are normalized by the relative
pressure difference ⌬p / ⌬p0. The dotted line is a power-law fit to
the data with a resulting slope of −1. The solid line is a constant
value.
FIG. 3. 共Color online兲 共a兲 Number of particles, N*, that pass
through a pore before clogging as a function of W / D 共circles兲.
The same data rescaled according to the prediction of our simple
To elucidate the mechanism of clogging, we vary W / D, model 共diamonds兲. Solid symbols: D = 5.2 ␮m. Open symbols:
the ratio of the constriction width W, to the particle diameter D = 2.9 ␮m. Inset: schematic pore cross section. Shown in blue are
D, while keeping the channel height fixed at H ⯝ 25 ␮m. We the “sticking areas” of width ⑀ and height H. 共b兲 Schematic illus-
observe a very strong dependence of N* on W / D, as shown tration of fluid streamlines 共blue兲 and a particle trajectory 共red兲 at a
in Fig. 3. The dependence is well described by a power law, constriction.
as shown by the solid line which is a fit to the data, N*
⬀ 共W / D兲m, with an exponent m = 4 ± 0.1. Qstick ⑀H vstick
To interpret this scaling, we propose a model based on pstick = = , 共3兲
Qtotal WH vtotal
successive particle buildup through irreversible sticking
events near the constriction. We assume that a sticking event
where vstick / vtotal is the ratio of the average velocity in the
is determined by the initial streamline on which a particle is
sticking region to that in the overall constriction. With each
traveling prior to reaching any constriction. The fluid follows
sticking event the effective channel area decreases, while the
streamlines, which must be compressed at constrictions. Far width of the sticking area, ⑀, remains approximately the
away from a constriction, particles follow the same stream- same. After n sticking events have already occurred, the
lines; however, at constrictions solid particles are forced to sticking probability pstick共n兲 becomes
move relative to the fluid. This movement must result from
physical forces, such as viscous lubrication and electrostatic
␣DH vstick共n兲
surface forces, which are repulsive and drive particles away pstick共n兲 = , 共4兲
from the sidewalls. Therefore, for particles to stick at con- WH − nD2 vtotal共n兲
strictions, their distance of approach to the sidewall must be
comparable to the length scale where attractive surface where ␣ = ⑀ / D; the additional nD2 term accounts for the re-
forces dominate and some fraction of these particles will duction in the total cross-sectional area through the addition
interact strongly with the sidewalls and stick irreversibly. We of particles stuck to the wall. The number of particles that
assume that all particles initially found on streamlines that pass through the pore before complete clogging occurs is
pass within a characteristic distance ⑀ Ⰶ D from the sidewall then N* = 兺n=0
N
1 / pstick, where N = WH / D2 is the number of
will stick irreversibly to the wall; this region is shown sche- particles needed to fill the channel cross section. Assuming a
matically within the inset of Fig. 3共a兲. The probability of parabolic velocity profile in the channel the velocity ratio
sticking for any particle is thus given by the fraction of becomes vstick共n兲 / vtotal共n兲 ⯝ 2␣DH / 共WH − nD2兲, where we
streamlines that pass through the region associated with irre- assume that ␣D is small compared to the effective width of
versible sticking events. Consequently, the sticking probabil- the channel Weff共n兲 ⯝ 共WH − nD2兲 / H. Therefore, the average
ity of a particle is given by the ratio of the flow rate Qstick number of particles that must flow through a channel con-
through the sticking region to the total flow rate Qtotal, striction before clogging occurs is

061402-3
WYSS et al. PHYSICAL REVIEW E 74, 061402 共2006兲

N
1 共WH − nD2兲2 W 3H
N ⯝ 兺
*
⯝ . 共5兲
2 n=0 ␣2D2H2 6 ␣ 2D 4

We plot N*D4 / W3H as a function of W / D in Fig. 3共a兲; within


experimental error, this quantity is independent of the ratio
between pore and particle size, as illustrated by a linear fit to
the data, shown by the dotted line.
The trajectory of a particle, shown schematically in Fig. FIG. 4. 共Color online兲 Characteristic clogging time scale t* of
3共b兲, is determined by the balance of the repulsive forces and single pores versus the NaCl concentration. The error bars represent
the drag force towards the wall, induced by the fluid. In our the standard deviation of t*.
model, the occurrence of a sticking event is determined only
by the initial streamline a particle is traveling on prior to any
constriction. We assume that there exists a limiting fluid low this value, we observe distinctly faster clogging dynam-
streamline 关9,10兴 that is located a distance ⑀ = ␣D from the ics with increasing salt concentration; both the average
sidewall. Thus, the critical distance of minimum approach clogging time t* and the corresponding length scale ⑀ in-
should correspond to a length that is comparable to the at- creased by a factor of 3 within the range of salt concentra-
tractive well of the interaction potential that stabilizes par- tions studied. Thus, the characteristic length scale of particle
ticles against aggregation. An order-of-magnitude estimate of capture, ⑀, which is independent of the flow rate, is closely
forces acting on a particle shows that the dominant repulsive linked to the forces that regulate particle stability.
mechanism is viscous lubrication. This suggests that particle Using a simple macroscopic technique for tracking the
trajectories are independent of fluid flow rate, as the domi- blockage times of microfluidic channels, we characterize the
nant repulsive and attractive forces are both proportional to physical processes that govern single-pore clogging dynam-
fluid velocity. This is also in agreement with the observed ics at the nanometer scale. The dynamics of clogging and the
flow-rate independence of the clogging. length scale ⑀ may provide an alternate measure of the sta-
Remarkably, ⑀ is determined directly from our macro- bility of particle suspensions. Our clogging picture is not
scopic experimental data 关see Fig. 3共a兲兴. By rescaling the limited to the dependence on colloidal interaction forces. It
data in Fig. 3共a兲, we obtain ⑀ ⬇ 10 nm, a length scale not captures the dependence of clogging dynamics on the rel-
directly discernible in our measurements. However, this evant experimental parameters, thus providing a surprisingly
length scale is comparable to the typical width of an electro- general picture for the dynamics of clogging processes at the
static double layer at low ionic concentrations. single-pore level. This single-pore clogging mechanism is
A consequence of our model is that the time scale for thus an essential first step in understanding and predicting
clogging should depend sensitively on the length scale of the blockage of real porous materials that consist of intercon-
repulsive interactions. To test this, we perform experiments nected pore networks.
in the presence of salt 共NaCl兲, over a broad concentration
range from 10−3 to 0.1 mol/ L 共see Fig. 4兲. The addition of We thank A. Ajdari, E. Dufresne, and E. Weeks for dis-
counter-ions screens the surface charge, eventually leading to cussions and Halliburton, the NSF 共Grant No. DMR-
spontaneous aggregation of the colloidal particles at salt con- 0602684兲, and the Harvard MRSEC 共Grant No. DMR-
centrations ⬎0.1 mol/ L. However, at salt concentrations be- 0213805兲 for support.

关1兴 C. Ghidaglia, L. de Arcangelis, J. Hinch, and E. Guazzelli, 关6兴 M. A. Unger, H.-P. Chou, T. Thorsen, A. Scherer, and S. R.
Phys. Rev. E 53, R3028 共1996兲. Quake, Science 288, 113 共2000兲.
关2兴 R. Narayan, J. Coury, J. Masliyah, and M. Gray, Ind. Eng. 关7兴 G. H. Goldsztein and J. C. Santamarina, Appl. Phys. Lett. 85,
Chem. Res. 36, 4620 共1997兲. 4535 共2004兲.
关3兴 S. Hong, R. Faibish, and M. Elimelech, J. Colloid Interface 关8兴 K. V. Sharp 共unpublished兲.
Sci. 196, 267 共1997兲. 关9兴 W. B. Russel, D. A. Saville, and W. R. Schowalter, Colloidal
关4兴 Y. N. Xia and G. M. Whitesides, Annu. Rev. Mater. Sci. 28, Dispersions 共Cambridge University Press, Cambridge, En-
153 共1998兲. gland, 1989兲.
关5兴 T. Squires and S. Quake, Rev. Mod. Phys. 77, 977 共2005兲. 关10兴 L. A. Spielman, Annu. Rev. Fluid Mech. 9, 297 共1977兲.

061402-4

You might also like