Download as pdf or txt
Download as pdf or txt
You are on page 1of 13

Journal of Membrane Science 582 (2019) 322–334

Contents lists available at ScienceDirect

Journal of Membrane Science


journal homepage: www.elsevier.com/locate/memsci

Deposition and reentrainment of colloidal particles in disordered fibrous T


filters under chemically and physically unfavorable conditions
Handol Leea, Seungkoo Kanga, Seong Chan Kima,∗, David Y.H. Puia,b
a
Particle Technology Laboratory, Mechanical Engineering, University of Minnesota, 111 Church St., S.E., Minneapolis 55455, USA
b
School of Science and Engineering, The Chinese University of Hong Kong, Shenzhen (CUHKSZ), Guangdong 518172, China

A R T I C LE I N FO A B S T R A C T

Keywords: This study is to establish computational fluid dynamics (CFD) simulations of disordered fibrous filters under
Derjaguin-Landau-Verwey-Overbeek (DLVO) unfavorable conditions to investigate the effects of chemical and physical factors on filtration performance. We
theory employed a discrete phase model (DPM) to track individual particles, and user-defined functions were im-
Disordered fibrous filter plemented to consider the particle deposition to the collector surface via interception and interaction energy.
Single sphere efficiency
During the DPM process, calculations based on interactional and fluid hydrodynamic effects were proceeded for
Single fiber efficiency
Collision efficiency
all injected individual colloids. The results of filtration and collision efficiencies of the single collectors obtained
by our CFD simulations showed a very good agreement with the existing expressions and data, accurately
predicting the trends of particle deposition and reentrainment. After the validation of the developed particle
tracking method, the disordered fibrous filters with randomly distributed monodisperse fibers were studied. The
results showed that the reentrainment of deposited particles was enhanced with increasing fluid velocity and
solid volume fraction and decreasing fiber size due to the hydrodynamic effects. Moreover, the combined effect
with ionic strength, Hamaker constant and zeta potential determined particle deposition behavior by altering
particle/filter interactions. The developed simulation method for predicting filtration performance does not
contain any free parameters or empirical factors.

1. Introduction efficiency often does not correlate well with the sieving characteristics
of small nanoparticles. This discrepancy usually occurs because the
Enormous benefits from nanomaterials spur the development of transport of small nanoparticles is led by diffusion, which significantly
nanotechnology in many industries, e.g., semiconductor, pharmaceu- enhances the rate of contact to filter surfaces [21,22]. Besides, the
tical, chemical engineering, food and environmental applications [1–4]. complexity of possible interactions between a particle and filter sur-
As the significant investments are being encouraged and made for the faces increases uncertainty in estimating the quality of membrane fil-
research and development of nanotechnology for many years, serious ters. Therefore, the accurate prediction of filtration efficiency, espe-
hazards from nanosized materials have been accompanied, which may cially for small nanoparticles, i.e., < 100 nm, by understanding particle
create risks in human health and environment [5–8]. Therefore, efforts transport and successful deposition onto surfaces is necessary [23–27].
to remove nanomaterials discharged from various sources should be To investigate the deposition behavior of colloidal particles in
made for preventing the abovementioned issues. For instance, many porous media, people have been commonly using the classical colloid
researches on release of engineered nanoparticles and its control for filtration theory (CFT) originally developed by Yao et al. [23]. Ac-
cleaning diverse aqueous environments have been investigated [9–14]. cording to CFT, in terms of successful deposition of colloidal particles,
Especially, careful consideration has been given to nanoparticle re- the particles must overcome the energy barrier and attach in the pri-
moval for purification of wastewater and drinking water [15–17]. mary minimum of the Derjaguin‐Landau‐Verwey‐Overbeek (DLVO)
Micro- and ultrafiltration using membrane filters have been con- interaction energy profile, which is constructed by sum of two inter-
sidered as one of the energy-efficient water treatment strategies for action energies, van der Waals and electrical double layer energy. The
colloidal nanoparticles [18–20]. Quality of the membrane filters is ty- deposition behavior is rated by collision efficiency in CFT, which re-
pically rated by their mean pore sizes, assuming the dominant removal presents the fraction of collisions leading to the successful attachment
mechanism is sieving. It was found that, however, the filtration at energy minima. On the basis of CFT, the interaction force boundary


Corresponding author.
E-mail address: sckim@umn.edu (S.C. Kim).

https://doi.org/10.1016/j.memsci.2019.03.092
Received 26 January 2019; Received in revised form 30 March 2019; Accepted 30 March 2019
Available online 10 April 2019
0376-7388/ © 2019 Elsevier B.V. All rights reserved.
H. Lee, et al. Journal of Membrane Science 582 (2019) 322–334

layer (IFBL) analysis has been developed to investigate the single col- conducted single collector simulations obtaining filtration and collision
lector efficiency [28]. However, deposition characteristics predicted by efficiencies under favorable and unfavorable conditions to validate the
IFBL approximation under unfavorable conditions, i.e., like-charged accuracy of our simulation method. The particle tracking was done with
interactive surfaces in the presence of repulsive interaction, have shown the discrete phase model (DPM) modified by user-defined functions
significant deviations from experimental observations [29–33]. A (UDFs) using a C programming language in order to consider the effects
number of studies have been conducted to find the reasons for these of interaction energies and hydrodynamic drag on particle deposition
discrepancies between CFT and the experimental results and proved behavior. The Maxwell approach with the aid of the DLVO theory was
that the discrepancy is due to the ignorance of colloidal particle de- utilized to incorporate the secondary and primary minimum deposi-
position at the secondary minimum region [34–37]. To include the tions. After assuring the accuracy of our numerical simulations on fil-
secondary minimum deposition, which occurs at a shallow attractive tration performance of the single collectors, we investigated the effects
well, the Maxwell approach has been employed for particle behaviors at of chemical and physical factors, i.e., ionic strength, Hamaker constant,
the vicinity of collector surfaces [26,38,39]. Hahn and O'Melia [38] zeta potential, fluid velocity, fiber diameter and solid volume fractions
used a Brownian Dynamic/Monte Carlo model to simulate the move- (SVF), on collision efficiency of disordered fibrous filters. It should be
ment of colloidal particles near a collector surface and highlighted the noted that existing analytical or numerical studies based on trajectory
importance of the secondary minimum deposition. The further study by calculations, e.g., Happel's and Kuwabara's models, built the foundation
Shen et al. [26] proposed a priori analytical method utilizing the of current filtration theories using collectors in ordered positions by
Maxwell approach that estimates collision efficiency for both secondary assuming a unit cell of filter geometry arrayed uniformly. This approach
and primary minima. Their results showed that the collision efficiencies might need empirical correction factors, valid for a specific range of
obtained by the Maxwell approach were higher than ones predicted by collector, particle diameter and fluid velocity in order to predict the
IFBL approximation due to the secondary minimum deposition and performance of real filter media. The disordered fibrous filter simula-
gave the better agreement with the experimental data. However, their tions in air filtration by Hosseini and Tafreshi [47] demonstrated that
study overestimated the collision efficiency for large particles due to they did not require any empirical correction factors in estimating
the absence of hydrodynamic effects caused by the fluid flow. particle deposition behavior and the 2-D simulations of fibrous filter
Several studies demonstrated that, unlike the primary minimum media provided filtration efficiency predictions successfully with much
deposition in the presence of the strong attraction force, colloidal less computational resource compared to 3-D simulations. Therefore,
particles attached in the weak secondary minima are strongly affected our findings from numerical simulations of liquid filtration are expected
by the hydrodynamic force [40–42]. This explains the deviations in to provide important insights and valuable guidelines to the deposition
collision efficiency appeared in both Hahn and O'Melia [38] and Shen mechanisms of colloidal particles on real fibrous filter media.
et al. [26] because they only considered Brownian diffusion kinetic
energies and excluded the hydrodynamic effects. The effects of the 2. Flow fields
hydrodynamic drag on particle deposition behaviors were significant
findings for colloidal particle filtration study since traditional theories In this section, we describe the different calculation domains em-
had mainly considered chemical factors, e.g., ionic strength, pH and ployed for single collectors and disordered fibrous filters, and the nu-
zeta potential, as dominant parameters that affect the collision effi- merical schemes for solving equations using the CFD code from ANSYS
ciency [23,34,43]. Studies on hydrodynamic factors on the deposition Fluent are introduced.
behaviors revealed that the collision efficiency decreased as the fluid
flow velocity increased [41,42,44,45]. This was proved by Torkzaban 2.1. Single sphere and single fiber simulations
et al. [45] who developed the analytical approach for calculating ad-
hesive and hydrodynamic torques acting on colloidal particles attached Fig. 1 shows the 2-D calculation domains for single collectors and
to collector surfaces. However, all of the abovementioned filtration boundary conditions considered for the single sphere and single fiber
studies have only focused on spherical collectors in packed bed filtra- simulations are shown in Fig. 1a and b, respectively. Fluid enters the
tion systems and there is a clear lack of studies on other types of filter simulation domain through the velocity inlet and exits through the
media such as non-woven and composite types of membrane filters, pressure outlet. The axis boundary condition was applied at the side to
which contain cylindrical fibrous collectors with much smaller dimen- the semicircle for simulating the spherical collector while a symmetric
sions than the granular collectors, e.g., sand and glass bead. In our boundary condition was used for both upper and lower sides of the
previous study, we employed the Maxwell approach to consider the fibrous collector domain. It should be noted that for filtration studies on
effects of secondary and primary minimum deposition and performed granular (or spherical) and fibrous media, fluid flow in the media is
torque analysis for track-etched membrane filters with straight-through often described with an assemblage of unit cells. Each cell contains a
pore structures, which is limited to this filter structure with the aid of a spherical or fibrous collector and a fluid envelope, and the envelope
well-defined aerosol filtration model, e.g., capillary tube model [22]. dimension is determined by the fluid characteristics due to the collector
Most recent study by Rastegar et al. [46] conducted the computational shape, e.g., sphere or cylindrical fiber, and porosity of the filter media.
simulations on a single fibrous collector but did not consider the hy- Happel's and Kuwabara's model are most widely applied for the flow
drodynamic effects on the attached colloidal particles. Moreover, unlike characterization of spherical and fibrous media, respectively
the packed bed filtration system (fairly ordered structure), the fibrous [24,48–50]. In Fig. 1, bH and bK are the radii of the fluid envelopes
membrane filters are comprised of fibers distributed in random places determined by Eqs. (1) and (2), which are adopted from Happel's and
without ordered arrangement, where the flow characteristics differ Kuwabara's cell models, respectively, and they are used as the heights
depending on locations due to the disordered fiber structure. Therefore, of the domains in this single collector simulation study, representing a
it is very difficult to accomplish the accurate predictions of particle half cell.
deposition onto fibers. To the best of our knowledge, existing mathe-
ac
matical models and other analytical approaches cannot be easily ap- bH =
(1 − ε )1/3 (1)
plied to modeling the whole filter media due to the abovementioned
reasons. and
The objective of this study is to model the particle deposition be- ac
havior in disordered fibrous filters with randomly distributed mono- bK =
(1 − ε )1/2 (2)
disperse fibers using the computational fluid dynamics (CFD) code.
Prior to the investigation of disordered fibrous filters, we first where ac is collector radius and ε is porosity. Table 1 shows the

323
H. Lee, et al. Journal of Membrane Science 582 (2019) 322–334

2.2. Disordered fibrous filters

To understand the effects of chemical and physical factors on the


filtration performance of fibrous filters, we conducted 2-D simulations
comprised of randomly distributed monodisperse fibers. It should be
noted that the simulation domain containing the fibers should be suf-
ficiently large enough to produce the consistent results independent on
the domain size. Therefore, we monitored pressure drops across the
filter as we increased the domain size with the same SVF, and we found
that the pressure drop was converged and finally became independent
of the domain size. The final domain size for fibers was chosen as
40Dc × 40Dc, where Dc is the fiber diameter, in terms of effective
computational effort and reliable simulation results. We developed a
Java code to generate the fibrous filters with different SVFs, 5, 10 and
15%. Fibers with a diameter Dc were treated as circles and sequentially
added in the random place of the designated square domain with
40Dc × 40Dc. The locations of the most recently added fiber and the
existing fibers were recorded continuously. If the newly added fiber
overlapped with the existing ones or the distance between centers of the
fibers was smaller than the minimum gap of 1.5Dc to avoid mis-
calculation caused from two fibers being too close each other, the fiber
was regenerated in other random location. This procedure was repeated
until the desired SVF was reached.
Fig. 1. Calculation domains for (a) single spherical and (b) single fibrous col- Fig. 2 represents the calculation domains for disordered fibrous
lector. Each domain size was determined by the relation between porosity and
filters with the different SVFs of 5, 10 and 15%. The generated geo-
collector size based on Happel's and Kuwabara's cell model, using Eqs. (1) and
metries were exported to Gambit software for setting boundary condi-
(2).
tions and generating mesh. The boundary conditions are denoted in
Fig. 2b as a representative. A symmetry boundary condition for the
Table 1 upper and lower sides of the computational domain was employed. A
Information of calculation domains and flow conditions used in the CFD si- velocity inlet and pressure outlet were applied for the introducing and
mulations of single collectors.
escaping flow, respectively.
Collector Fluid envelope Porosity ε Fluid velocity U
diameter Dc radius bH or bK [m/s] 2.3. General information of flow simulations
[μm] [μm]

Single sphere 328 187.42 0.33 1.2 × 10−5 The calculation domains were meshed using triangular elements
600 355.69 0.4 2.8 × 10−3 with the sufficient number of grid points around a fiber. To determine
Single fiber 0.5 0.56 0.8 5.0 × 10−4 the optimized number of meshes, we increased the number of grid
2 2.24
points around each fiber, and the pressure drop across the filter domain
20 22.36
was monitored. It was found that the pressure drop increased with in-
creasing mesh density and reached a plateau. In terms of the effective
simulation conditions for single spherical and fibrous collectors. To computational effort, 60 grid points were used for 1 μm fibers and the
validate the accuracy of the CFD simulations of fluid flow and particle same approach was applied to determine the number of gird points for
tracking method in the presence of particle-filter interactions, we per- all different sized fibers. For all simulations, i.e., single collectors and
formed two cases of a single spherical collector and three cases of a disordered fibrous filters, ANSYS Fluent v14.0 software was used to
single fibrous collector under a broad range of parameters, predicting solve the continuity, momentum and energy equations. The fluid flow
filtration and collision efficiencies of the single collectors. In brief, the was assumed to be steady, laminar and incompressible. The SIMPLE
filtration efficiency is the fraction of particles captured by collectors, algorithm and double precision solver were used for flow field calcu-
and the collision efficiency is the fraction of effective collisions between lations, and no slip at collector surfaces were set. The general criterion
particles and collectors that result in successful attachments. The ob- for convergence was set as 10−6 for all equations. Another criterion for
tained efficiencies were compared to the existing theoretical models, the judgment of successful convergence for calculations is to achieve
correlations or experimental data under the same conditions. The or- the stable values of local parameters, e.g., pressures and velocities.
ders of fluid velocity, in this study, are ranged from 10−5 to 10−3 m/s
for representing the conditions of typical colloidal transport through 3. Theoretical considerations
the subsurface environments (low flow rate) and engineered filtration
systems (high flow rate) [51,52]. The spherical collector size and por- 3.1. Transport mechanisms and particle tracking methods
osity information were adopted from the studies by Tufenkji and
Elimelech [25] and Shen et al. [26] to directly compare our results with General transport mechanisms of suspended particles to filter media
their data. For the fibrous collector simulations, we arbitrarily chose the are known as diffusion, interception, gravitational settling and inertial
micro-sized fibers and the porosity based on a polypropylene mem- impaction. However, it should be noted that unlike aerosol filtrations,
brane filter as shown in Fig. S1 of Supplementary Material 1 (SM 1), the gravitation and impaction are typically not considered as main
which is a widely used membrane in micro- and ultrafiltration processes filtration mechanisms for nanosized particles in liquid filtrations due to
[53]. Generally, fibrous membrane filters have porosity ranging from the high fluid density comparable to the particles [54]. The particle
0.7 to 0.99. For parametric studies on the disordered fibrous filter, we tracking was simulated in a Lagrangian reference frame by using the
further investigated the various porosity cases, i.e., ε = 0.85, 0.9 and DPM in the ANSYS Fluent software. The DPM estimates the particle
0.95. trajectory by integrating the forces, F, acting on particles, which are
balanced with the acceleration of the particles as follows:

324
H. Lee, et al. Journal of Membrane Science 582 (2019) 322–334

3.2. DLVO theory and considered energies

The classical DLVO theory is represented by van der Waals (VDW)


energy, φVDW, as an attraction force at the particle/filter interaction and
electrical double layer (EDL) energy, φEDL, as repulsive force under
unfavorable conditions. In this study, the Hamaker approximation ex-
pression for the retarded VDW interaction energy was employed be-
cause the expression is usually valid for small separations between
surfaces but the VDW interaction is weak anyway for large separations
[57]. Therefore, this expression has been widely used due to its sim-
plicity and adequacy:
Aap 14H −1
φVDW = − ⎛1 + ⎞
6H ⎝ λ ⎠ (4)

where A is the Hamaker constant and ap is the colloidal particle radius;


H is the separation distance between the particle and filter surfaces; λ is
the characteristic wavelength of the dielectric, which is usually taken as
100 nm [27,38,58].
The linear superposition approximation (LSA) as the expression of
the EDL energy was used in this study because the LSA does not require
any assumption of constant potential or constant charge on surfaces and
is considered the most suitable approximation that expresses the rea-
listic EDL interaction [59–61]:

kB T 2
( ) tanh ( )
zeψc zeψp
φEDL =
64πε0 ε
κ ( )
ze
tanh 4kB T 4kB T

× {(κap − 1) e−κH + (κap + 1) e−κ (H + 2ap) } (5)

where ε0 and ε are the permittivity of vacuum and the relative per-
mittivity of the solution, respectively; kB is the Boltzmann constant; T is
the absolute temperature, 298 K; κ is the reciprocal double layer
thickness, which is a function of ionic strength; z is the valence of the
ions, which is 1, assuming the monovalent electrolyte used throughout
this study; ψc and ψp are the zeta potentials of the collector and particle,
respectively.
The classical DLVO energy, i.e., sum of VDW and EDL energies,
presents an infinite energy well where the primary minimum occurs,
which is irreversible attachment. However, many experimental studies
revealed that the release of colloids from the primary minimum was
observed with decreasing ionic strength and increasing pH of solutions
[62,63]. In this study, therefore, the repulsive term of the interatomic
Lennard-Jones potential as an expression for the Born repulsion, φBorn,
was included to have the finite depth of the primary minimum:

Aσ 6 ⎡ 8ap + H 6ap − H ⎤
φBorn = +
7560 ⎢
⎣ (2ap + H )
7 H7 ⎥ ⎦ (6)

where σ is the Born collision parameter and the value is usually taken as
Fig. 2. Calculation domains for the CFD simulations of disordered fibrous filters 0.5 nm [27,38]. The total interaction energy, φTotal, between particle
with SVFs of (a) 5%, (b) 10% and (c) 15% and boundary conditions. and filter surfaces under unfavorable conditions is modeled as a sum of
three energy components, i.e., Eqs. (4)–(6):

d→
vp 1 →
φTotal = φVDW + φEDL + φBorn. (7)
dt
=
mp
∑F
(3) The total DLVO energy profile described by Eq. (7) is generally
characterized by a deep attractive energy well at the primary minimum
where vp is the particle velocity and mp is the particle mass. The in- (Φpri), a shallow attractive energy well at the secondary minimum (Φsec)
cluded forces are Stokes drag and the Brownian diffusion, which was and a maximum energy barrier (Φmax), depending on solution chemis-
modeled by a Gaussian white noise process [55,56]. The standard DPM, tries, e.g., ionic strength, Hamaker constant and zeta potential. It
however, treats a particle as a point mass, resulting in a lack of the should be mentioned that the estimated primary minimum distance was
interception effect from the particle size. In this study, therefore, we 0.3 nm according to Eq. (7); however, it is highly improbable that the
developed a UDF using the C programming language to evaluate par- kinetic energies bring two surfaces at this close separation distance,
ticle deposition via interception, which was incorporated by con- which is unrealistic due to the hydration layer [64,65]. Also, many
tinuously tracking the distance between the center of a particle, i.e., studies observed that there was a hydration effect in water and aqueous
location of the point mass, and the surface of the closest filter media, salt solutions, which was originated from the presence of a thick layer
i.e., circle, in every particle time step. If the distance was smaller than of hydrated counterions around colloidal particles at relatively high
or equal to the particle radius, the particle was considered to collide ionic strength, e.g., I > 10−4 M [22,47,66]. The short-range repulsion
with the filter media. region typically exists at separations approximately below 1–3 nm

325
H. Lee, et al. Journal of Membrane Science 582 (2019) 322–334

[28,66]. Our previous study revealed that experimental data agreed the secondary and primary minimum well, respectively, which means
very well with models when considering a hydration layer thickness of the particle can escape to bulk suspension with the 7% probability. A
1–1.5 nm, which means that within this separation distance the col- generated random number ranging from 0 to 1 is designated to each
loidal particles experience very strong repulsion that prevents deposi- tracked particle and compared with αsec and αpri for its fate of the de-
tion [22]. Therefore, in addition to the Born repulsion as a short-range position behavior as shown in Fig. S2 in SM 2. From this approach the
repulsion, we took a hydration effect with the 1 nm hydration layer as tracked particles in a Lagrangian reference frame can have the statis-
the primary minimum distance in our developed UDFs, which has been tically reliable deposition characteristics based on the Maxwell dis-
considered as the possible closest separation distance between two tribution.
surfaces in the presence of the hydration layer [65].
3.3.2. Torque analysis
3.3. Determination of particle attachment in DPM process We also developed the UDFs for calculations of adhesive and hy-
drodynamic torques, which were activated once a tracked particle was
We developed UDFs for calculating the interaction energies between determined to be deposited in either secondary or primary minimum.
a particle and filter surface and torques acting on the attached particles The comparison between both torques was performed to decide if the
to determine the particle detachment from the filter media. Fig. S2 in particle would stay at or be detached from the filter surface. It should be
SM 2 shows the flow chart of the executed UDFs during the DPM pro- mentioned that particle removal mechanisms by the hydrodynamic
cess. The details of calculations can be found in the following sections. effect from a collector surface are lifting, sliding and rolling, and the
dominant mechanism under laminar flow conditions has been known to
3.3.1. Maxwell model be rolling, which occurs when the hydrodynamic torque, Thyd, over-
Small colloids in a liquid at thermal equilibrium are considered to comes the adhesive torque, Tadh [45]. We employed the most widely
have a velocity distribution close to the Maxwell distribution and, used torque calculations for attached colloids developed by Torkzaban
therefrom, many researchers utilized Maxwell approach in estimating et al. [45]. The torque analysis was also used in our previous study on
particle deposition behaviors [22,26,38,39]. We employed the Maxwell particle deposition onto Nuclepore filter and was able to predict colloid
approach by assuming particle deposition in the primary and secondary detachment in the primary minimum as well as secondary minimum
minima. In this model, the velocity distribution, f(vp), of colloidal accurately [22]. The details in hydrodynamic and adhesive torque
particles is described as: calculations can be found in SM 3.
2
mp ⎞3/2 2 ⎛ mp vp ⎞
f (vp) = 4π ⎛ ⎜ vp exp ⎜−

⎟. 3.4. Calculations of filtration efficiency in DPM process
⎝ 2πkB T ⎠ ⎝ 2kB T ⎠ (8)
Based on the Maxwell model, a particle can have one of three major 3.4.1. Single sphere filtration efficiency
consequences once it reaches the secondary minimum. i) If the kinetic The schematic of the statistical Lagrangian particle tracking (SLPT)
energy of the particle is smaller than the interaction energy of the approach for calculating the filtration efficiency of single spherical
secondary minimum, Φsec, it remains in the secondary minimum. ii) If collector is shown in Fig. 3a. The SLPT approach with the use of CFD
the particle has the kinetic energy larger than the energy barrier, ΔΦ simulations has been first developed by Yook et al. [67] and verified as
(sum of Φmax and Φsec), it transports to the primary minimum and is an effective method for calculating the rate of deposited particles in the
deposited at the primary minimum well with the corresponding at- 2-D axisymmetric calculation domain [68,69]. In this study, injected
tractive interaction energy. iii) If the kinetic energy of the particle is particles (Nin = 10,000) with the same diameter were positioned on an
larger than the secondary minimum energy, Φsec, and smaller than the injection line at a constant spacing of Δr [ = bH/(Nin-1)] in radial di-
energy barrier, ΔΦ, the particle escapes to bulk suspension. These as- rection and injected with a velocity same as a fluid velocity, U, as
sumptions can be expressed as: shown in Fig. 3a. The radius of the injection plane was set to be the
fluid envelope radius, bH, which is the same as the domain size. The
Φsec
4 2 radial position of the ith particle, ri, can be obtained as:
αsec = ∫ π 1/2
x exp (−x 2) dx
0 (9) ri = (i − 1) × Δr (12)
and In the SLPT approach, the number of particles injected through the
∞ annular cross-sectional area of 2πri × Δr was considered to be in pro-
4 2
αpri = ∫ π 1/2
x exp (−x 2) dx portion to the flow rate of the fluid passing through the area based on
ΔΦ (10) the assumption of the homogeneous particle concentration at the in-
where x is the dimensionless kinetic energy of the particle, which is jection plane. By adding the deposition probability, fi, of the ith particle
defined as: to this relation, which is 0 or 1 if the ith particle is escaped or deposited,
respectively, the number of particles deposited on the spherical col-
mc vp2 lector in 3-D, Nde, is proportional to sum of the flow rates corresponding
x2 = .
2kB T (11) to the deposited particles, Qde. This can be expressed as:

These estimated αsec and αpri were used as probabilities for a particle Nin

behavior in the DPM process to determine which minimum, i.e., sec-


Nde ∝ Qde = ∑ 2πri ΔrUfi .
i=1 (13)
ondary or primary minimum, the particle would be deposited when it
reached the vicinity of the filter surface from bulk suspension. This It should be clearly noted that Nde is not the number of captured
process was done by comparing αsec and αpri with a generated random particles in the 2-D DPM process, but Nde represents all particles placed
number ranging from 0 to 1 given to each particle at the beginning of in the annular cross-sectional area. Similarly, the proportional relation
the DPM process. For example, αsec and αpri of a 20 nm particle can be can be applied between the total number of injected particles in 3-D
obtained by using the Maxwell approach when all parameters of DLVO through the particle injection plane, Ntot, and the total flow rate of the
interaction energies are given with the ionic strength of suspension of fluid through the plane, Qtot, which is expressed as:
0.05 M and both particle and filter zeta potentials of −20 mV. Under
Ntot ∝ Qtot = πbH 2U. (14)
these unfavorable conditions, αsec and αpri are 0.10 and 0.83. This in-
dicates that the particle has a 10% and 83% probability of deposition in Therefore, the particle filtration efficiency on the spherical

326
H. Lee, et al. Journal of Membrane Science 582 (2019) 322–334

where A1 is a constant, which is a function of only the depth of the


planar calculation domain. The total number of particles deposited on
the fibrous collector in 3-D, Nde, is also considered to be proportional to
the number of captured particles injected at the particle injection line in
2-D, Ncap, and the constant A1. Therefore, this relation can be expressed
as:
Nde = A1 × Ncap. (17)
Consequently, the filtration efficiency in the 2-D planar calculation
domain can be calculated as:
Nde A1 × Ncap Ncap
Efiber = = = .
Ntot A1 × Nin Nin (18)

4. Results and discussion

4.1. Validation of CFD simulations using single spherical and fibrous


collectors

Since we developed and employed the UDFs for interception and


other interaction-related effects in the DPM process, the results ob-
tained from the CFD simulations must be validated. We compared
single collector filtration and collision efficiencies with the theoretical
models and experiments.

4.1.1. Spherical collector efficiency under favorable conditions


Tufenkji and Elimelech [25] developed a correlation equation for
calculating spherical collector efficiency under favorable conditions
(Econtact), i.e., single collector contact efficiency, by considering sum of
contributions from diffusion, interception and gravity:

Econtact = 2.4AS1/3 NR−0.081 NPe


−0.715 0.052
NVDW + 0.55AS NR1.675 N A0.125
+ 0.22NR−0.24 NG1.11 NVDW
0.053
(19)
where the dimensionless parameters in Eq. (19) are defined in Table S1
in SM 4. AS is the porosity-dependent parameter, which is defined for
the spherical collector used in Happel's model [48]:
2(1 − p5 )
AS =
2 − 3p + 3p5 − 2p6 (20)
1/3
where p is defined as (1-ε) . The single sphere efficiencies under fa-
vorable conditions, i.e., single spherical collector contact efficiencies,
by Tufenkji and Elimelech [25] correlation shown in Eq. (19) and by
our CFD simulations with the use of the DPM and UDFs shown in Eq.
(15), are presented in Fig. 4a. The contact efficiency of the particle size
Fig. 3. Statistical Lagrangian particle tracking approach for (a) single spherical smaller than 1 μm increased as the particle size decreased due to the
and (b) single fibrous collector. diffusion effect. The larger sized particles showed the increasing effi-
ciency due to the interception effect. As a result, the present CFD si-
collector, Esphere, can be calculated using Eqs. (13) and (14) as mulation using the DPM and the UDFs was proven to correctly predict
N
the efficiency of the spherical collector with consideration of dominant
Nde Q ∑ in 2πri ΔrUfi effects from diffusion and interception under favorable conditions, as-
Esphere = = de = i = 1 2 .
Ntot Qtot πbH U (15) suming no particle detachment after colliding with the collector.

4.1.2. Fibrous collector efficiency under favorable conditions


3.4.2. Single fiber filtration efficiency Iwasaki [70] first introduced filter coefficient, λ, as a filtration rate
The filtration efficiency of the single fiber using the DPM process in on a single fiber and the coefficient has been used to predict the contact
the 2-D planar calculation domain can be calculated with a simpler efficiency under favorable conditions in the single collector. With the
process compared to the axisymmetric case for the spherical collector. assumption that deposition is due to the combined influences of inter-
Again, with the assumption of a constant particle number concentration ception, gravity and London-van der Waals force, Choo and Tien [71]
in the free stream, particles were positioned with the same spacing and performed trajectory calculations by employing Kuwabara's model to
injected at the particle injection plane as shown in Fig. 3b. The total predict the initial filter coefficient, λ0, under favorable conditions:
number of particles injected from the injection plane toward the fibrous
collector, Ntot, is assumed to be proportional to the number of injected 6 1−ε
λ0 = ⎛ ⎞ AS (0.216 × 10−0.41εNR1.55 N LO
0.1542
+ 2.99 × 10−4
particles at the particle injection line (Nin = 10,000), which can be ⎝ π ⎠ Dc
expressed as: × 103εNG1.1 NR−0.3) (21)
Ntot = A1 × Nin (16) where AS used in Eq. (21) is based on Kuwabara's model and differently

327
H. Lee, et al. Journal of Membrane Science 582 (2019) 322–334

Here, εs is the solidity, which is the same as SVF. The correlation Eq.
(21) does not include the diffusion effect of the colloids. Choo and Tien
[71] suggested the filter coefficient for Brownian diffusion, λBM, ob-
tained from the solution of the convection-diffusion equation:
−2/3
λBM = (9.2/ π )(c1 + c3)1/3 [(1 − ε )/ Dc ] NPe (28)
The relationship between the filter coefficient and the single fiber
contact efficiency is found to be:
πDc
Econtact = × (λ 0 + λBM )
4(1 − ε ) (29)
Fig. 4b shows the single fiber efficiencies under favorable condi-
tions, i.e., single fibrous collector contact efficiency, obtained by Choo
and Tien [71] and the CFD simulations shown in Eq. (29) and Eq. (18),
respectively, for three different fiber diameters ranging from 0.5 to
20 μm. It should be noted that since Eq. (21) is valid only for
10−3 < NR < 10−1, the upper limits of the particle size for these fiber
diameters of 0.5 μm, 2 μm and 20 μm are 50 nm, 200 nm and 2 μm,
respectively. Under the same conditions of porosity and fluid velocity
except for the fiber diameter, i.e., ε = 0.8 and U = 5 × 10−4 m/s, the
single fiber efficiency increased as the fiber diameter decreased. The
present simulation results using the DPM modified by the UDF agreed
very well with the results of Choo and Tien correlation.

4.2. Collision efficiency of single sphere under unfavorable conditions

In the previous sections, the CFD simulation using the UDF for the
interception effect was validated by comparing the single collector ef-
ficiencies under favorable conditions. Fig. 5 shows the comparison of
the collision efficiencies for the single spherical collector as a function
of ionic strength. The collision efficiency was obtained by the present
CFD simulations as:

Fig. 4. Comparison of the single collector contact efficiencies to validate the


present CFD simulations. The results of our numerical results (symbols) are
compared with the results of Tufenkji and Elimelech [25] and Choo and Tien
[71] for (a) single spherical and (b) single fibrous collector, respectively.

defined from the one in Happel's model due to the different config-
uration of the collector that results in the different fluid flow char-
acteristic.
(2/3)(4c1 + c4 )
AS =
c1 [(1/εs ) − 2 + εs ] + (c4 /2)(εs − 1 − ln εs ) (22)

where
c1 = −εs c4/4 (23)
c2 = −c1 − c3 (24)

c3 = c1 + (c4/2) (25)
Fig. 5. Comparison of collision efficiencies of the single spherical collector as a
c4 = −4/(2 ln εs + 3 − 4εs + εs2) (26) function of ionic strength. The present CFD simulations were compared with the
results from the theoretical models, empirical correlation and experimental
εs = 1 − ε (27) data.

328
H. Lee, et al. Journal of Membrane Science 582 (2019) 322–334

Efiltration (unfavorable )
Ecollision =
Econtact (favorable ) (30)

where Efiltration and Econtact are the filtration efficiency and contact effi-
ciency obtained under unfavorable and favorable condition using Eq.
(15) with and without consideration of interaction energies, respec-
tively. It should be mentioned that since there is no comparable study
on the collision efficiency for the single fibrous collector, we only
compared the results for the single spherical collector. Our DPM results
were compared with theoretical and experimental studies to validate
the accuracy of our simulation method in the presence of interaction
energies. The experimental data were collected from sand column ex-
periments using 30 and 1156 nm polystyrene latex (PSL) colloidal
particles conducted by Shen et al. [26].
Bai and Tien [44] suggested the correlation equation obtained from
partial regression analysis based on their own experiments and other
data from Vaidyanathan and Tien [72] and Elimelech [73]. However,
this correlation often fails to reflect the effect of parameters, e.g., flow
velocity, on the collision efficiency due to its empirical nature of lim-
itations. In this case, therefore, the Bai and Tien correlation over-
estimates the collision efficiency for both small and large particles. The
collision efficiency obtained by the IFBL approximation is greatly de-
pendent on particle size. For larger particles with much larger energy
barrier, i.e., 1156 nm PSL particles, the IFBL model reveals the sig-
nificant underestimation of collision efficiency as the ionic strength
decreases, i.e., increasing -Log(I). This is likely due to the fact that,
based on the IFBL model, the successful attachment under unfavorable
conditions occurs only when the colloidal particles overcome the en-
ergy barrier to attach in the primary minimum. Finally, Shen et al. [26]
calculated the collision efficiency using the Maxwell approach by
considering both the secondary and primary minimum depositions,
which can be estimated by sum of Eqs. (9) and (10). Their theoretical
model with the consideration of the secondary minimum deposition
overestimated the collision efficiency of 1156 nm particles due to the
relatively deep secondary minimum well. Then, they further developed
the analytical approach by adding the hydrodynamic effects on the
detachment of colloidal particles [39]. The collision efficiencies ob-
tained by the abovementioned theoretical approaches are represented
all together in Fig. 5.
Compared to the experimental data, the correlation by Bai and Tien
[44] and the prediction by Shen et al. [26] without the hydrodynamic
effects overestimated the collision efficiency, in particular, at low ionic
strength, i.e., high -Log(I). A relatively good agreement is shown for the Fig. 6. Effects of ionic strength on (a) filtration and (b) collision efficiency of
prediction by Shen et al. [39] with the hydrodynamic effects, which disordered fibrous filters.
indicated that the detachment due to the flow must be considered for an
accurate prediction of collision efficiency. Our CFD simulation results of 4.3. Filtration efficiency of disordered fibrous filters
collision efficiency obtained by the DPM and UDFs agree well with the
experimental collision efficiencies. However, complete breakthroughs, The present simulation results under favorable and unfavorable
i.e., zero attached particles, of both sized colloidal particles were ob- conditions showed good agreements when compared to theoretical
tained by the DPM when ionic strength was less than 0.005 M, i.e., -Log studies and experimental data, validating the accuracy of the DPM and
(I) > 2.3, due to the detachment of the particles by the higher hy- UDFs on interception, interaction energy and hydrodynamic effects.
drodynamic torque than adhesive torque. The work done by Bradford Therefore, in this session, we obtained the filtration efficiency of dis-
et al. [74] revealed that, in the granular packed-bed filtration experi- ordered fibrous filters by employing the DPM with the developed UDFs.
ments, straining of large micron-sized particles was observed under The following DPM results for the disordered fibrous filters were ob-
unfavorable conditions. Therefore, we expected that there might have tained by tracking 50,000 particles, i.e., Nin = 50,000, and the filtration
been mechanically sieved or intercepted particles between collectors efficiencies were calculated using Eq. (18). Filtration efficiencies under
that caused the low collision efficiencies even at very low ionic favorable conditions, i.e., contact efficiency, are shown in Fig. S3 in SM
strength, e.g., I < 0.005 M. We also assumed that this sieving effect, 5. It was found that the smaller fiber size, higher SVF and lower fluid
i.e., straining, of 1156 nm particles in the experiments is the main velocity enhanced the filtration efficiency, which was expected because
reason for the higher collision efficiency compared to the 30 nm par- the trends are identical to ones for aerosol filtration [75,76].
ticles. Compared to the theoretical models or empirical correlation, the
predictions of particle behavior under unfavorable conditions using our
CFD simulations sufficiently match well with experimental data 4.3.1. Ionic strength
without any empirical correction factors. Therefrom, we employed our Due to the complexity of the possible interactions between a col-
developed numerical simulations to estimate the disordered fibrous loidal particle and filter surface, which might cause the failure of suc-
filters under different chemical and physical conditions. cessful attachment, it is very difficult to predict the filtration efficiency.

329
H. Lee, et al. Journal of Membrane Science 582 (2019) 322–334

I = 0.1, 0.2 M and favorable conditions, we found that larger particles,


e.g., 200 nm, were more affected by the ionic strength than smaller
ones. This result has important implications because many studies have
excluded the hydrodynamic effects on deposited particles, which re-
sulted in even larger particles with successful deposition at relatively
high ionic strengths, e.g., I = 0.1 and 0.2 M. Interestingly, all test par-
ticles at the 0.1 and 0.2 M ionic strengths tend to be deposited at the
primary minimum, i.e., αpri = 1.0 calculated by Eq. (10), which is
usually considered as irreversible attachment; however, the detachment
was observed, i.e., Ecollision < 1. Particle detachment in the primary
minimum was observed frequently in other experiments and considered
as a prominent issue [40,41,77–79]. Fig. 7 shows the comparison be-
tween hydrodynamic and adhesive torques acting on the 40, 100 and
200 nm colloidal particles attached to the fibers with a diameter of 1 μm
in the primary minimum at the 0.1 M ionic strength for the further
evaluation of the results in Fig. 6. The experienced hydrodynamic and
adhesive torques by attached particles were saved during the DPM
process. The hydrodynamic torques (black symbols) varied with the
local position of the attached colloidal particles onto the fibers with the
Fig. 7. Comparison between hydrodynamic (black open symbols) and adhesive maximum value at the midpoint regions and the minimum value at the
(red lines) torques acting on 40, 100 and 200 nm colloidal particles attached to front and rear stagnation regions of each fiber. However, the adhesive
the fibers in the primary minimum at I = 0.1 M. (For interpretation of the re- torque (red lines) is independent of the location and associated only
ferences to color in this figure legend, the reader is referred to the Web version
with the ionic strength and particle size, which were fixed for each DMP
of this article.)
case. Larger particles have the deeper primary minimum, meaning
higher adhesive torque applies on the particles. Even though the ad-
Moreover, the filtration performance can be affected by solution che- hesive torque on 200 nm particles is highest, i.e., 5.1 × 10−20 Nm
mical conditions significantly. (Fig. 7), most of the tracked particles experienced the higher hydro-
Fig. 6a and b represent the filtration and collision efficiency as a dynamic torques than the adhesive torque and the high chance of
function of particle size obtained by Eqs. (18) and (30) via CFD simu- particle detachment from the fiber surfaces. On the other hand, most of
lations, respectively. Hamaker constant was set to be 1 × 10−20 J and the hydrodynamic torques acting on 40 nm particles were lower than
zeta potentials of both particles and filter media were assumed to be the adhesive torque indicating the higher chance of successful attach-
−20 mV, which indicated unfavorable conditions. The diameter of ment on the fiber surfaces.
filter fibers is 1 μm with the SVF of 15% and the fluid velocity is Fig. 8 depicts the exemplary particle trajectories for 200 nm parti-
1 × 10−3 m/s. It should be noted that in many studies on liquid fil- cles under the different ionic strengths. Since the complete penetrations
tration, Hamaker constants for PSL-water-glass beads or fibrous mem- occurred at the ionic strengths of 0.02 and 0.05 M, the particle trajec-
branes were assumed to be between 1 × 10−20 and 2 × 10−20 J, which tory of the 0.05 M ionic strength case is shown as a representative in
can be calculated based on Lifshitz theory [26,27,38,66]. It can be seen Fig. 8d. The number of particles escaping through outlet was clearly
that the overall filtration and collision efficiency increased as the ionic reducing with increasing ionic strength, indicating the enhanced suc-
strength increased. From the comparison between three curves of cessful adhesion to fiber surfaces.

Fig. 8. Particle trajectories of 200 nm particles through the disordered fibrous filter with 15% SVF and 1 μm fiber diameter under different conditions: (a) favorable,
(b) I = 0.2 M, (c) I = 0.1 M and (d) I = 0.05 M.

330
H. Lee, et al. Journal of Membrane Science 582 (2019) 322–334

particles larger than 100 nm, the increasing particle diameter results in
increasing overall hydrodynamic torques partially overcoming the ad-
hesive torque as explained in Fig. 7. Moreover, as the particle size de-
creased, even though the depth of the attractive primary minimum well
decreased, αpri increased due to the higher kinetic energy of smaller
particles and the lower energy barrier, which result in more fractions of
the smaller particles that overcome the energy barrier easily compared
to the larger particles. In addition, the weaker hydrodynamic torque
due to the smaller particle diameter, the substantial increase in filtra-
tion and collision efficiency was observed, except for the case of the
Hamaker constant of 1 × 10−20 J and particle zeta potential of −30
mV, which indicates very week particle attraction (gray triangle).

4.3.3. Fluid velocity, fiber diameter and SVF


It is important to note again that it is very difficult and complex to
predict the collision efficiency of a whole fibrous filter via existing
theoretical or mathematical approaches because the flow characteristic
around each fiber is arbitrary due to randomly distributed fibers. Our
CFD simulations using the DPM and UDFs, however, enables to obtain
the collision efficiency of the disordered fibrous filters by accessing the
flow field information around each individual fiber.
Fig. 10a, c and 10e depict the collision efficiencies of 100 nm col-
loidal particles for disordered fibrous filters obtained by the CFD si-
mulations to investigate the effects of fluid velocity, fiber diameter and
SVF. We also denoted hydrodynamic (open symbols) and adhesive
torques (red lines) acting on the 100 nm particles deposited on every
fiber under the tested conditions in Fig. 10b, d and 10f to explain the
results of the obtained collision efficiencies. The zeta potentials of both
particle and filter fiber and Hamaker constant were set to be −20 mV
and 1 × 10−20 J, respectively, and the other fixed conditions used in
each comparison result are shown in Fig. 10a, c and 10e. In Fig. 10a, the
collision efficiency decreased with increasing fluid velocity due to the
enhanced hydrodynamic torques overcoming the adhesive torque. With
the ionic strength of 0.05 M, the secondary minimum deposition be-
comes dominant due to the hydration layer that prevents the primary
minimum deposition. Therefore, Fig. 10b represents the hydrodynamic
and adhesive torques acting on the particles deposited in the secondary
minimum. The adhesive torque was estimated as 5.4 × 10−22 Nm and
most of the estimated hydrodynamic torques (open circles) are lower
than the adhesive torque when the fluid velocity is 1 × 10−5 m/s, in-
dicating if a particle has a kinetic energy enough for the particle to be
deposited in the secondary minimum, i.e., R < αsec (please refer to Fig.
S2 of SM 2), the particle will be successfully deposited in the secondary
minimum without an interference from the insignificant fluid flow.
Therefore, the collision efficiency in this case, i.e., 0.54, become very
similar to αsec, which is estimated as 0.58 by Eq. (9). When the fluid
velocity increased to 1 × 10−4 m/s, hydrodynamic torques (open tri-
angles) increased and large fractions of them exceeded the adhesive
torque, meaning particle detachment. Finally, at 1 × 10−3 m/s, all es-
Fig. 9. Effects of Hamaker constant and zeta potential on filtration efficiency of
disordered fibrous filters with the SVF of 15%. The zeta potential of the fiber, ζc,
timated hydrodynamic torques acting on the particles deposited in the
was set to the constant value of −20 mV and that of the particles, ζp, varied. secondary minimum became higher than the adhesive torque, resulting
in zero collision efficiency representing complete penetration.
The relation between the collision efficiency and torque analysis
4.3.2. Hamaker constant and zeta potential
data in Fig. 10c and d and Fig. 10e and f can be examined in a similar
The effects of Hamaker constant and zeta potential of colloidal
way. Interestingly, the lower hydrodynamic torques were estimated for
particles on filtration and collision efficiencies are shown in Fig. 9. The
particles attached to larger fibers, i.e., Dc = 5 μm, in Fig. 10d, leading to
SVF and fluid velocity were set to be 15% and 1 × 10−3 m/s, respec-
the higher collision efficiency in Fig. 10c. The similar trend has been
tively. Based on Eqs. (4) and (5), the higher Hamaker constant enhances
observed in the granular filtration system in other studies, which show
the VDW attraction, and the lower absolute value of zeta potential
that the increasing collector size enhanced the collision efficiency
decreases the EDL repulsion. Under these conditions the filtration ef-
[45,74]. Our torque analysis provides an important insight by ex-
ficiency increases. Interestingly, with the Hamaker constant of
plaining this trend clearly. Fig. 11 represents velocity profiles from the
2 × 10−20 J under the unfavorable conditions (red, green and blue
midpoint of the single fibrous collector to the upper boundary of the
symbols), the filtration efficiencies for the particles smaller than or
domain under the conditions represented in Fig. 10c and e. In Fig. 11a,
equal to 100 nm were estimated as same as the filtration efficiencies
the higher flow velocity for the 1 μm collector was estimated at the
under favorable conditions indicating successful particle attachment
same distance from the fiber surface at midpoint. It means that the
once contacting filter surfaces, i.e., Ecollision = 1. However, for the
velocity near the collector surface for the same SVF is more significantly

331
H. Lee, et al. Journal of Membrane Science 582 (2019) 322–334

Fig. 10. Effects of (a) fluid velocity, (b) fiber diameter and (c) SVF on collision efficiency of disordered fibrous filters.

changed for the smaller collector due to the decrease in collector-to- surface for the collector with the 15% SVF than the smaller SVFs. Even
collector distance. Therefore, the higher hydrodynamic torque is in- though the velocity profiles depicted in Fig. 11 are for single fibrous
duced and, thereby, enhances the particle detachment. collectors, the general trend of deposition behavior in the disordered
The collision efficiency decreased with increasing SVF, i.e., de- fibrous filters is same. We found that collision efficiency, torque com-
creasing porosity as shown in Fig. 10e. We assumed that this happened parison and velocity profile were associated very well one another.
because the narrower flow paths between neighboring fibers, formed by It should be noted that complex deposition behaviors of colloidal
the denser disordered fibers due to the higher SVF, accelerated the local particles on fibrous filters was described through the current study
fluid velocity through the filtration regions. It is clearly shown in depending on chemical and physical factors, showing different filtra-
Fig. 11b that the velocity magnitude increases much faster from the tion and collision efficiencies. It should be noted that, therefore, exact

332
H. Lee, et al. Journal of Membrane Science 582 (2019) 322–334

empirical correction factors. Therefore, the good agreement with ex-


perimental observations shown in the current work indicates the strong
validation for the accuracy of our numerical approach. Then we em-
ployed our numerical method to model the disordered fibrous filters
and understand the chemical and physical effects on filtration perfor-
mance.
Disordered fibrous filters with randomly placed fibers were gener-
ated and simulated to analyze the filtration and collision efficiencies.
We found that chemical and physical factors, i.e., ionic strength,
Hamaker constant, zeta potential, fluid velocity, fiber size and solid
volume fraction (SVF), influenced the filtration performance sig-
nificantly by altering interaction energies and executing forces.
Hydrodynamic and adhesive torques were obtained by injecting the
sufficient number of particles through domains under different condi-
tions. The results revealed that high fluid velocity, small fiber and high
SVF (or low porosity) enhanced overall particle detachment by in-
creasing the hydrodynamic effects. Finally, our CFD simulations with
the developed UDFs will be very useful for further studies on different
structures of filter media and other issues on liquid-borne particles and
their filtrations in the presence of complex interactions.

Acknowledgments

The authors thank the support of members of the Center for


Filtration Research: 3M Corporation, A.O. Smith Company, Applied
Materials., BASF Corporation, Boeing Company, Corning Co., China
Yancheng Environmental Protection Science and Technology City,
Cummings Foundation, Donaldson Company, Inc., Entegris, Inc., Ford
Motor Company, Guangxi Wat Yuan Filtration System Co., Ltd, LG
Electronics Inc., Mott Corporation, MSP Corporation, Parker Hannifin,
Samsung Electronics, Xinxiang Shengda Filtration Technology Co., Ltd.,
Shigematsu Works Co., Ltd., TSI Inc., W. L. Gore & Associates, Inc., and
the affiliate member National Institute for Occupational Safety and
Health (NIOSH).

Appendix A. Supplementary data

Supplementary data to this article can be found online at https://


doi.org/10.1016/j.memsci.2019.03.092.

Fig. 11. Velocity profiles from the midpoint of the single collector surface to
References
the upper boundary. The velocity profiles are compared between single fibrous
collectors with different (a) fiber diameters and (b) SVFs.
[1] P. Biswas, C.-Y. Wu, Nanoparticles and the environment, J. Air Waste Manag. Assoc.
55 (2005) 708–746.
values of those efficiencies might vary according to operating condi- [2] C.J. Huang, B.M. Yang, K.S. Chen, C.C. Chang, C.M. Kao, Application of membrane
technology on semiconductor wastewater reclamation: a pilot-scale study,
tions. However, the newly developed DPM process with an aid of UDFs Desalination 278 (2011) 203–210.
will provide the better understanding of the filtration behaviors under [3] H. Bai, X. Zan, L. Zhang, D.D. Sun, Multi-functional CNT/ZnO/TiO2 nanocomposite
membrane for concurrent filtration and photocatalytic degradation, Separ. Purif.
different conditions and the guidance for the most effective way of Technol. 156 (2015) 922–930.
using and designing fibrous filters. [4] O. Ganzenko, N. Oturan, D. Huguenot, E.D. van Hullebusch, G. Esposito,
M.A. Oturan, Removal of psychoactive pharmaceutical caffeine from water by
electro-Fenton process using BDD anode: effects of operating parameters on re-
5. Conclusions moval efficiency, Separ. Purif. Technol. 156 (2015) 987–995.
[5] T. Masciangioli, W.-X. Zhang, Peer reviewed: environmental technologies at the
nanoscale, Environ. Sci. Technol. 37 (2003) 102A–108A.
We evaluated the filtration performance of fibrous filters using [6] G. Oberdörster, E. Oberdörster, J. Oberdörster, Nanotoxicology: an emerging dis-
computational fluid dynamics (CFD) simulations by calculating filtra- cipline evolving from studies of ultrafine particles, Environ. Health Perspect. 113
(2005) 823–839.
tion and collision efficiencies. We developed user-defined functions [7] K.A. Dunphy Guzmán, M.R. Taylor, J.F. Banfield, Environmental risks of nano-
(UDFs) to modify a standard discrete phase model (DPM) process in technology: national nanotechnology initiative funding, 2000–2004, Environ. Sci.
ANSYS Fluent by incorporating interaction energies on particle de- Technol. 40 (2006) 1401–1407.
[8] N.C. Mueller, B. Nowack, Exposure modelling of engineered nanoparticles in the
position behaviors under unfavorable conditions with the presence of environment, Environ. Sci. Technol. 42 (2008) 4447–4453.
repulsion. Prior to the modeling of disordered fibrous filters with ran- [9] L. Reijnders, Cleaner nanotechnology and hazard reduction of manufactured na-
noparticles, J. Clean. Prod. 14 (2006) 124–133.
domly distributed fibers, the DPM with the developed UDFs used for [10] T.M. Benn, P. Westerhoff, Nanoparticle silver released into water from commer-
particle tracking methods were validated by directly comparing the cially available sock fabrics, Environ. Sci. Technol. 42 (2008) 4133–4139.
filtration and collision efficiencies of single spherical and fibrous col- [11] S.A. Blaser, M. Scheringer, M. MacLeod, K. Hungerbühler, Estimation of cumulative
aquatic exposure and risk due to silver: contribution of nano-functionalized plastics
lectors with existing theoretical and experimental results, showing that and textiles, Sci. Total Environ. 390 (2008) 396–409.
our CFD simulations provided the best fit with the experimental data [12] Y. Zhang, Y. Chen, P. Westerhoff, K. Hristovski, J.C. Crittenden, Stability of com-
mercial metal oxide nanoparticles in water, Water Res. 42 (2008) 2204–2212.
under unfavorable conditions. It is worth mentioning that our numer-
[13] H. Hyung, J.-H. Kim, Dispersion of C60 in natural water and removal by
ical simulation using the modified DPM process does not require any

333
H. Lee, et al. Journal of Membrane Science 582 (2019) 322–334

conventional drinking water treatment processes, Water Res. 43 (2009) 2463–2470. hydrodynamics and DLVO forces on colloid attachment in porous media, Langmuir
[14] T.E. Abbott Chalew, G.S. Ajmani, H. Huang, K.J. Schwab, Evaluating nanoparticle 23 (2007) 9652–9660.
breakthrough during drinking water treatment, Environ. Health Perspect. 121 [46] V. Rastegar, G. Ahmadi, S.V. Babu, Filtration of aqueous colloidal ceria slurries
(2013) 1161–1166. using fibrous filters – an experimental and simulation study, Separ. Purif. Technol.
[15] R. Srinivasan, G.A. Sorial, Treatment of perchlorate in drinking water: a critical 176 (2017) 231–242.
review, Separ. Purif. Technol. 69 (2009) 7–21. [47] S.A. Hosseini, H.V. Tafreshi, 3-D simulation of particle filtration in electrospun
[16] J. Altmann, A.S. Ruhl, D. Sauter, J. Pohl, M. Jekel, How to dose powdered activated nanofibrous filters, Powder Technol. 201 (2010) 153–160.
carbon in deep bed filtration for efficient micropollutant removal, Water Res. 78 [48] J. Happel, Viscous flow in multiparticle systems: slow motion of fluids relative to
(2015) 9–17. beds of spherical particles, AIChE J. 4 (1958) 197–201.
[17] L. Huang, S. Zhao, Z. Wang, J. Wu, J. Wang, S. Wang, In situ immobilization of [49] S. Kuwabara, The forces experienced by randomly distributed parallel circular cy-
silver nanoparticles for improving permeability, antifouling and anti-bacterial linders or spheres in a viscous flow at small Reynolds numbers, J. Phys. Soc. Jpn. 14
properties of ultrafiltration membrane, J. Membr. Sci. 499 (2015) 269–281. (1959) 527–532.
[18] Y. Bessiere, D.F. Fletcher, P. Bacchin, Numerical simulation of colloid dead-end [50] C. Tien, Principles of Filtration, first ed., Elsevier, Oxford, UK, 2012.
filtration: effect of membrane characteristics and operating conditions on matter [51] J.N. Ryan, M. Elimelech, J.L. Baeseman, R.D. Magelky, Silica-coated titania and
accumulation, J. Membr. Sci. 313 (2008) 52–59. zirconia colloids for subsurface transport field experiments, Environ. Sci. Technol.
[19] R. Wang, S. Guan, A. Sato, X. Wang, Z. Wang, R. Yang, B.S. Hsiao, B. Chu, 34 (2000) 2000–2005.
Nanofibrous microfiltration membranes capable of removing bacteria, viruses and [52] J.L. Cleasby, F.W. Pontius (Ed.), Water Quality and Treatment, fourth ed., McGraw-
heavy metal ions, J. Membr. Sci. 446 (2013) 376–382. Hill, New York, 1990.
[20] H. Etemadi, R. Yegani, M. Seyfollahi, The effect of amino functionalized and [53] C. Regula, E. Carretier, Y. Wyart, G. Gésan-Guiziou, A. Vincent, D. Boudot,
polyethylene glycol grafted nanodiamond on anti-biofouling properties of cellulose P. Moulin, Chemical cleaning/disinfection and ageing of organic UF membranes: a
acetate membrane in membrane bioreactor systems, Separ. Purif. Technol. 177 review, Water Res. 56 (2014) 325–365.
(2017) 350–362. [54] B.E. Logan, T.A. Hilbert, R.G. Arnold, Removal of bacteria in laboratory filters–-
[21] J. Walter, T. Thajudeen, S. Süss, D. Segets, W. Peukert, New possibilities of accurate models and experiments, Water Res. 27 (1993) 955–962.
particle characterisation by applying direct boundary models to analytical cen- [55] H. Lee, S.J. Yook, Deposition velocity of particles in charge equilibrium onto a flat
trifugation, Nanoscale 7 (2015) 6574–6587. plate in parallel airflow under the influence of simultaneous electrophoresis and
[22] H. Lee, D. Segets, S. Süß, W. Peukert, S. Chen, D.Y.H. Pui, Liquid filtration of na- thermophoresis, J. Aerosol Sci. 67 (2014) 166–176.
noparticles through track-etched membrane filters under unfavorable and different [56] A. Li, G. Ahmadi, Dispersion and deposition of spherical particles from point
ionic strength conditions: experiments and modeling, J. Membr. Sci. 524 (2017) sources in a turbulent channel flow, Aerosol Sci. Technol. 16 (1992) 209–226.
682–690. [57] S. Lin, M.R. Wiesner, Theoretical investigation on the interaction between a soft
[23] K.M. Yao, M.T. Habibian, C.R. O'Melia, Water and waste water filtration: concepts particle and a rigid surface, Chem. Eng. J. 191 (2012) 297–305.
and applications, Environ. Sci. Technol. 5 (1971) 1105–1112. [58] J. Gregory, Approximate expressions for retarded van der Waals interaction, J.
[24] R. Rajagopalan, C. Tien, Trajectory analysis of deep‐bed filtration with the spher- Colloid Interface Sci. 83 (1981) 138–145.
e‐in‐a‐cell porous media model, AIChE J. 2 (1976) 523–533. [59] J. Gregory, Interaction of unequal double layers at constant charge, J. Colloid
[25] N. Tufenkji, M. Elimelech, Correlation equation for predicting single‐collector ef- Interface Sci. 51 (1975) 44–51.
ficiency in physicochemical filtration in saturated porous media, Environ. Sci. [60] S. Lin, M.R. Wiesner, Exact analytical expressions for the potential of electrical
Technol. 38 (2004) 529–536. double layer interactions for a sphere-plate system, Langmuir 26 (2010)
[26] C. Shen, B. Li, Y. Huang, Y. Jin, Kinetics of coupled primary and secondary 16638–16641.
minimum deposition of colloids under unfavorable chemical conditions, Environ. [61] A.R. Petosa, D.P. Jaisi, I.R. Quevedo, M. Elimelech, N. Tufenkji, Aggregation and
Sci. Technol. 41 (2007) 6976–6982. deposition of engineered nanomaterials in aquatic environments: role of physico-
[27] M.W. Hahn, D. Abadzic, C.R. O'Melia, Aquasols: on the role of secondary minima, chemical interactions, Environ. Sci. Technol. 44 (2010) 6532–6549.
Environ. Sci. Technol. 38 (2004) 5915–5924. [62] J.N. Ryan, P.M. Gschwend, Effects of ionic strength and flow rate on colloid release:
[28] J.N. Ryan, M. Elimelech, Colloid mobilization and transport in groundwater, relating kinetics to intersurface potential energy, J. Colloid Interface Sci. 164
Colloids Surf., A 107 (1996) 1–56. (1994) 21–34.
[29] J.C. Baygents, J.R. Glynn, O. Albinger, B.K. Biesemeyer, K.L. Ogden, R.G. Arnold, [63] S.A. Bradford, Y. Wang, S. Torkzaban, J. Šimůnek, Modeling the release of E. coli
Variation of surface charge density in monoclonal bacterial populations: implica- D21g with transients in water content, Water Resour. Res. 51 (2005) 3303–3316.
tions for transport through porous media, Environ. Sci. Technol. 32 (1998) [64] S. Usui, T. Yamasaki, Adhesion of mercury and glass in aqueous solutions, J. Colloid
1596–1603. Interface Sci. 29 (1969) 629–638.
[30] T.A. Camesano, B.E. Logan, Influence of fluid velocity and cell concentration on the [65] G. Frens, J.T.H.G. Overbeek, Repeptization and the theory of electrocratic colloids,
transport of motile and nomotile bacterias in porous media, Environ. Sci. Technol. J. Colloid Interface Sci. 38 (1972) 376–387.
32 (1998) 1699–1708. [66] J.N. Israelachvili, Intermolecular and Surface Forces, third ed., Academic Press, San
[31] S.F. Simoni, H. Harms, T.N.P. Bosma, A.J.B. Zehnder, Population heterogeneity Diego, 2011.
affects transport of bacteria through sand columns at low flow rates, Environ. Sci. [67] S. Yook, H. Fissan, C. Asbach, J. Hyeun, J. Wang, P. Yan, D.Y.H. Pui, Evaluation of
Technol. 32 (1998) 2100–2105. protection schemes for extreme ultraviolet lithography (EUVL) masks against
[32] C.H. Bolster, A.L. Mills, G.M. Hornberger, J.S. Herman, Spatial distribution of top–down aerosol flow, 38 (2007) 211–227.
bacteria experiments in intact cores, Water Resour. Res. 35 (1999) 1797–1807. [68] H. Lee, S.-J. Yook, S.Y. Han, The effects of simultaneous electrophoresis and ther-
[33] J.A. Redman, M.K. Estes, S.B. Grant, Resolving macroscale and microscale hetero- mophoresis on particulate contamination of an inverted EUVL photomask surface in
geneity in pathogen filtration, Colloids Surf., A 191 (2001) 57–70. parallel airflow, Eur. Phys. J. Plus. 127 (2012) 122–133.
[34] M. Elimelech, C.R. O'Melia, Effect of particle size on collision efficiency in the de- [69] H. Lee, S. Yook, K. Lee, Deposition of charged particles on a flat plate in parallel
position of Brownian particles with electrostatic energy barriers, Langmuir 6 (1990) flow in the presence of an electric field, IEEE Trans. Semicond. Manuf. 27 (2014)
1153–1163. 287–293.
[35] G.M. Litton, T.M. Olson, Particle size effects on colloid deposition kinetics: evidence [70] T. Iwasaki, Some notes on sand filtration, J. Am. Water Work. Assoc. 29 (1937)
of secondary minimum deposition, Colloids Surf., A 107 (1996) 273–283. 1591–1602.
[36] N. Tufenkji, M. Elimelech, Deviation from the classical colloid filtration theory in [71] C. Choo, C. Tien, Hydrosol deposition in fibrous beds, Sep. Technol. 1 (1991)
the presence of repulsive DLVO interactions, Langmuir 20 (2004) 10818–10828. 122–131.
[37] N. Tufenkji, M. Elimelech, Breakdown of colloid filtration theory: role of the sec- [72] R. Vaidyanathan, C. Tien, Hydrosol deposition in granular beds-An experimental
ondary energy minimum and surface charge heterogeneities, Langmuir 21 (2005) study, Chem. Eng. Commun. 81 (1989) 123–144.
841–852. [73] M. Elimelech, Predicting collision efficiencies of colloidal particles in porous media,
[38] M.W. Hahn, C.R. O'Melia, Deposition and reentrainment of Brownian particles in Water Res. 26 (1992) 1–8.
porous media under unfavorable chemical conditions: some concepts and applica- [74] S.A. Bradford, S.R. Yates, M. Bettahar, J. Simunek, Physical factors affecting the
tions, Environ. Sci. Technol. 38 (2004) 210–220. transport and fate of colloids in saturated porous media, Water Resour. Res. 38
[39] C.Y. Shen, Y.F. Huang, B.G. Li, Y. Jin, Predicting attachment efficiency of colloid (2002) 1–12.
deposition under unfavorable attachment conditions, Water Resour. Res. 46 (2010) [75] K.W. Lee, B.Y.H. Liu, K.W. Lee, B.Y.H. Liu, Experimental study of aerosol filtration
1–12. by fibrous filters experimental study of aerosol filtration by fibrous filters, Aerosol
[40] X. Li, P. Zhang, C.L. Lin, W.P. Johnson, Role of hydrodynamic drag on microsphere Sci. Technol. 1 (1982) 35–46.
deposition and re‐entrainment in porous media under unfavorable conditions, [76] S. Huang, C. Chen, Y. Kuo, C. Lai, R. Mckay, C. Chen, Factors affecting filter pe-
Environ. Sci. Technol. 39 (2005) 4012–4020. netration and quality factor of particulate respirators, Aerosol Air Qual. Res. 13
[41] W.P. Johnson, M. Tong, Observed and simulated fluid drag effects on colloid de- (2013) 162–171.
position in the presence of an energy barrier in an impinging jet system, Environ. [77] S.A. Bradford, S. Torkzaban, S.L. Walker, Coupling of physical and chemical me-
Sci. Technol. 40 (2006) 5015–5021. chanisms of colloid straining in saturated media, Water Res. 41 (2007) 3012–3024.
[42] M. Tong, W.P. Johnson, Excess colloid retention in porous media as a function of [78] S.A. Bradford, H.N. Kim, B.Z. Haznedaroglu, S. Torkzaban, S.L. Walker, Coupled
colloid size, fluid velocity, and grain angularity, Environ. Sci. Technol. 40 (2006) factors influencing concentration-dependent colloid transport and retention in sa-
7725–7731. turated porous media, Environ. Sci. Technol. 43 (2009) 6996–7002.
[43] M. Elimelech, Predicting collision efficiencies of colloidal particles in porous media, [79] T. Phenrat, J.E. Song, C.M. Cisneros, D.P. Schoenfelder, R.D. Tilton, G.V. Lowry,
Water Res. 26 (1992) 1–8. Estimating attachment of nano- and submicrometer-particles coated with organic
[44] R. Bai, C. Tien, Particle deposition under unfavorable surface interactions, J. Colloid macromolecules in porous media: development of an empirical model, Environ. Sci.
Interface Sci. 218 (1999) 488–499. Technol. 44 (2010) 4531–4538.
[45] S. Torkzaban, S. a. Bradford, S.L. Walker, Resolving the coupled effects of

334

You might also like