Chemical Engineering Science 175 (2018) 1–11

Contents lists available at ScienceDirect

Chemical Engineering Science

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

Effects of pore size distribution and coordination number on filtration

coefficients for straining-dominant deep bed filtration from percolation
theory with 3D networks
Binbin Ding a, Chaolin Li a,⇑, Yaowu Wang b, Jianhui Xu c
a
Environmental Science and Engineering Research Center, Shenzhen Graduate School, Harbin Institute of Technology, Shenzhen 518055, PR China
b
Shenzhen Key Laboratory of Urban Planning and Decision Making Simulation, Shenzhen Graduate School, Harbin Institute of Technology, Shenzhen 518055, PR China
c
School of Environment and Civil Engineering, Dongguan University of Technology, Dongguan 523000, PR China

h i g h l i g h t s

A straining-dominant deep bed filtration model based 3D network is proposed.

The effects of simulation parameters on the deep bed filtration are illuminated.

The simulated results based on 3D networks agree with the experimental results.

a r t i c l e i n f o a b s t r a c t

Article history: A power law relation between the filtration coefficients of straining and flux through small pores has
Received 26 April 2017 been reported. The effluent concentrations of the colloidal particles and the exponents derived from
Received in revised form 19 August 2017 the experiment are inconsistent with those obtained from 2D network models simulation in their
Accepted 16 September 2017
researches. In this study, a straining-dominant deep bed filtration (DBF) model is proposed based on
Available online 19 September 2017
3D network simulation. The effects of simulation parameters such as lattice type, lattice coordination
number z, pore size distribution (PSD), and particle capture scheme on DBF were investigated.
Keywords:
Consistent with the power law formula and the normalized effluent concentrations (Ce/C0), simulation
Straining
Pore size distribution
results indicate that the exponents increase with increasing coordination number z. The change in the
Percolation PSD parameters alters the flux and weight of path type linked to node, thereby influencing the numerical
Coordination number simulation of staining-dominate DBF on 3D networks. The effects of capture scheme and coordination
number on the simulated normalized effluent concentration could be due to variations in the total cap-
ture probability for different lattices with the same PSD. The results were obtained from simulation on
hexagonal close packing (HCP) and 2D triangular lattices under the optimal conditions. The simulated
Ce/C0 and exponents of the HCP lattice are similar to those of 2D triangular lattice and consistent with
the experimental data.
Ó 2017 Elsevier Ltd. All rights reserved.

1. Introduction 1999). Simulation of filtration is important in various industrial

Colloidal suspension transport in porous media is an essential
feature of various natural and industrial processes, such as petro-
leum production (Civan, 2015; Ding, 2010), ultrafiltration
(Shi et al., 2011; Ruthven, 2009; Madaeni et al., 2006), microfiltra-
tion (Sun et al., 2016; Vicente et al., 2013), packed bed filtration
(Islam et al., 2011; Chi and Payatakes, 1979), microbial transport
in porous media (Bai et al., 2016; Tufenkji, 2007; Harvey et al.,
1993), and formation of riverbeds (Battin and Sengschmitt,
⇑ Corresponding author.
E-mail address: lichaolin@hit.edu.cn (C. Li).
applications to optimize the operation conditions and minimize
costs. Study of colloidal suspension transport in porous media
requires a model that describes the porous matrix. Colloidal parti-
cles can be removed from porous media through gravity segrega-
tion, diffusion, straining, bridging, and electrical forces. The
migration and deposition of colloidal particles can be analyzed
through several models, such as Classic Models (Ives, 1980;
Heertjes and Lerk, 1967; Maroudas and Eisenklam, 1965) and Tra-
jectory analysis model (O’Melia, 1985; Tien and Payatakes, 1979;
Yao et al., 1971). The Classic Models are mainly based on the depo-
sition of suspended particles in the pore space of the filter media
and used to establish a macro empirical formula of effluent

https://doi.org/10.1016/j.ces.2017.09.033
0009-2509/Ó 2017 Elsevier Ltd. All rights reserved.
2 B. Ding et al. / Chemical Engineering Science 175 (2018) 1–11
Nomenclature
Ce effluent concentration
C0 influent concentration
f pore size distribution density
fns flow fraction via the inaccessible small pores
fa + fnl fractional flow via large pores
fl probability of a pore being larger than the injected par-
ticle
f⁄l average probability of particle flow through larger pores
fc conventional percolation threshold
fc⁄ flow-biased percolation threshold
ql flow in the capillary
l length of a capillary (m)
lw distance between two neighboring branches
f⁄inf fraction of flow through infinite cluster
B fraction of flow through the backbones of the infinite
clusters
L0 medium or network length (m)
Nl minimum number of capillaries connecting the inlet
and the outlet
Dp pressure difference between nodes (Pa)
q density of suspension
g gravitational acceleration
rn node radius (m)
rp capillary radius (m)
rs spherical particle radius (m)
rsc particle radius at the threshold
turbidity. The Classic Model is a simple empirical model and does
not provide significant insights into the physics of deposition. The
trajectory analysis model cannot accurately predict permeability
changes due to Deep Bed Filtration (DBF). Some model parameters
(critical interstitial velocity, and porosity of deposits rtrans) must
be approximated based on previous experimental data (Rege and
Fogler, 1988). Both models exhibit limitations in study of DBF.
The pore blockage of filter media in DBF by suspended particles
is similar to cluster growth in percolation model. Percolation
(Broadbent and Hammersley, 1957) is originally derived from fil-
tration, and both processes are similar in many aspects. Therefore,
percolation model can be applied to study filtration and obtain
accurate results (Bell et al., 1996; Allen, 2009; Golden, 1997;
Ewing and Gupta, 1993). Percolation model can also be used to pre-
dict the properties of medium (Perrier et al., 2010; Berkowitz and
Ewing, 1998; Selyakov and Kadet, 1997) and is associated with net-
work modeling of transport in porous media. Network models can
describe the effects of pore scale physics, whereas percolation the-
ory can reflect the effects of randomness on macroscopic proper-
ties (Berkowitz and Balberg, 1992), fluid properties, and their
interplay. Kaiser (Kaiser, 1997) established a microscopic model
of directed percolation to describe the clogging of a porous med-
ium caused by colloidal suspension. Yuan et al. (2012) applied per-
colation theory and random walk in a 2D square network to study
staining-dominate DBF; this study proposed two particle capture
mechanisms and two power-laws that describe the relationship
between the fractions of flow through the small pores and the fil-
tration coefficient. The trend of simulation agrees with that of lab-
oratory test data. However, the normalized effluent concentrations
(Ce/C0) and the exponents calculated from the experimental data
are inconsistent with the simulation results on the 2D network.
Our previous studies (Ding et al., 2015a,b; Ding and Chaolin,
2015) investigated the effects of coordination number, capture
scheme, and pore size distribution (PSD) on simulation of
straining-dominant DBF in 2D networks. The results verify that
the two parameters of PSD significantly influence the simulation
Vn volume of each node
Vb volume of each bond (capillary)
z coordination number
k filtration coefficient, capture probability per unit length
(m1)
kl network filtration coefficient, capture probability per
pore
g suspension dynamic viscosity (Pas)
l location parameter forpore size distribution
rtrans transition from smooth deposition to blocking type
deposition
r scale parameter for pore size distribution
h angle between a capillary and horizontal direction
ha average angle between a pore and flow direction
a exponent of power-law
b exponent of power-law
c exponent of power-law
Abbreviations
HCP Hexagonal Close Packing
BCC Body Center Cubic
PSD Porous Size Distribution
MINCAP Minimum Capture Scheme
MAXCAP Maximum Capture Scheme
PTM Parallel Tube Model
DBF Deep Bed Filtration
of power law and the experimental calculation of exponents. The
simulated data are consistent with the laboratory test results when
proper simulation conditions are used even in 2D networks.
The structure of 2D networks differs from that of actual 3D filter
media. In actual filtration, the influence of gravity should be con-
sidered. Moreover, the coordination number and capillary length
of the actual filter media vary up to a certain range but are constant
in a 2D regular network.
In this study, a numerical simulation model of staining-
dominate DBF was established to verify the power law and inves-
tigate the role and mechanism of microscopic parameters in the
process. The effects of simulation parameters such as lattice type,
lattice coordination number z, PSD, and particle capture scheme
were determined to optimize the simulation conditions. Finally,
the Ce/C0 obtained from the simulation of filtration on the 3D net-
work under the optimal conditions was compared with the labora-
tory test data in the literature.
2. Experimental section
For model validation, a small-scale experimental data set of
staining-dominate DBF was selected. The data set refers to injec-
tion of various monodispersed colloidal particle suspensions into
a porous matrix (laboratory test data of the medium 30/125
(Yuan et al., 2012)) that consists of packing glass bead with moni-
tored inlet and breakthrough particle concentrations. In this work,
we briefly review the experiment, where a plastic column with
packing space of 47 mm in diameter and 50 mm in length was con-
structed to simulate the porous media. Spherical glass granules
(size: 30–125 lm) for packing were acquired from Ballotini Bead,
Potters Industries Pty. Ltd. (Australia). Monodisperse suspensions
of yellow–green fluorescent carboxyl polystyrene latex micro-
spheres of different sizes were used in the experiments. These car-
boxyl groups provided the net negative charge in the alkaline
solution, resulting in mutual particle repulsion, as indicated by
the calculation of the interaction energy using Derjaguin–Landa
 B. Ding et al. / Chemical Engineering Science 175 (2018) 1–11 3

u–Verwey–Overbeek theory (Jia et al., 1998). Suspensions were large pores (fl), can be written in terms of probability density func-
prepared in degassed ultrapure MilliQ water and added with tion of pore throat radius f (rp) as follows:
0.1 M NaOH to achieve a pH of 10.4. Low colloidal concentrations Z 1
(8 ppm) and short colloid injection time (5–10 p.v.i.) were f l ðr s Þ ¼ f ðrp Þdr p ð2Þ
rs
employed to ensure negligible variations in PSD. The effluent con-
centration was monitored until it reaches the steady state. Addi- The probability of colloidal particle entrapment in pore space
tional details can be found in Yuan et al. (2012). Table 1 shows relies on pore conductivity. If the flow resistance in a microtubule
the colloidal particle size and corresponding Ce/C0 at the steady- obeys Poiseuille’s law, then the average probability of colloidal par-
state Ce from the aforementioned laboratory test data of medium ticles penetrating the large pores as the probability of conducting
30/125. particle flow can be calculated as follows:
PSD was estimated from packing glass granules through two R1
methods, namely, Monte Carlo procedure with Latin-Hypercube  r 4p f ðrp Þdr p
f l ðrs Þ ¼ Rr1
s
ð3Þ
Sampling based on Descartes’ theorem (Chalk et al., 2012; Stein, 0
r 4p f ðrp Þdr p
1987) (Monte Carlo) and Parallel Tube Model (PTM) (Chalk et al.,
The f⁄l at the percolation threshold can be written as
2012; You et al., 2013). Monte Carlo method assumes that spheri-
R1
cal granules are compactly packed, and each pore body consists of  r 4p f ðrp Þdr p
four granules. The throat consists of three grains with two mutu- f c ðrs Þ ¼ Rr1
sc
ð4Þ
0
r 4p f ðrp Þdr p
ally tangent. The radius of a particle that comes into contact with

the three granules reaches the minimum if the sphere center is where f c is the flow-biased percolation threshold, and rsc is the par-
located on the plane crossing the center of the granules. PTM is a ticle radius at the threshold.
microscale model for particle straining. The medium is represented The colloidal particles that penetrate the backbones of the infi-
by the model of triangular parallel capillaries alternated with mix- nite cluster can pass through the network media, whereas the par-
ing chambers. The analytical model for low-retention filtration is ticles penetrating the branch connected to the backbone are
derived, and the steady-state solution is as follows: entrapped at the dangling ends (Ding et al., 2015b) (Fig. 1). Yuan

 et al. (2012) reported a derivation process of formulae for percola-
L
C e ðr s Þ ¼ C 0 ðr s Þ½f a ðrs Þ þ f nl ðr s Þ exp f ns ðrs Þ ð1Þ tion, and this process is explained in Appendix A. Here, we briefly
l
review the derivation process of power-law relationship:
  b  
As shown in Eq. (1), rs is the injected particle size, f ns is the flow k / 1  f l f l  f c . The fraction of flow through the backbones
fraction via the inaccessible small pores, fa + fnl is the fractional of the infinite clusters may be estimated by using the following
flow via large pores, L is the length of column, l is the distance equation (Reyes and Jensen, 1985; Larson and Davis, 1982):
between two chambers, and Ce is the lower outlet concentration. qffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
ffi
      
PSD is determined through the PTM model by injecting colloidal B ¼ 2f l  f inf  2 f inf f l  f inf ð5Þ
particle suspensions into engineered porous media with monitored
inlet and breakthrough particle concentrations. The details about where f⁄inf
is the fraction of flow through infinite clusters, and B is the
PSD estimated from packing glass granules through PTM can be fraction of flow through the backbones of the infinite clusters. Far
found in Chalk et al. (2012). The PSD parameters estimated by above the percolation threshold, the branch and dangling ends are
the two methods are listed in Table 2. marginal, and the backbones dominate the main body of the pore
network f⁄inf ? f⁄l . The flux through finite clusters and dangling ends
3. Fundamental statistical parameters and power law can be expressed as:
  
1  B  1  fl fl  fc ð6Þ
In staining-dominate DBF, the injected particles are entrapped
in porous media when the particle radius rs is larger than the pore The distance between two neighboring branches lw may be
throat radius rp. Hence, the first key parameter, namely, fraction of assumed to vary by a power law, similar to the power law in Eq. (6):
  c  
lw / 1  f l fl  fc ð7Þ
Table 1
Particle sizes and the corresponding normalized effluent where c is the exponent. The filtration coefficient k is defined as the
concentrations (from cited reference). capture probability per unit length of porous medium and can be
expressed as:
rs (lm) Medium 30/125
1.57 0.9680 1B   b
k/ ¼ 1  fl ð8Þ
2.18 0.7990 lw
2.84 0.5810
3.17 0.1221 where b is power law exponent, which may be determined based on
4.54 0.0003 PSD shape, pore structure, and coordination number. The power law
exponent of Eq. (8) estimated based on experimental data are listed
in Table 2. Quantitative analysis of the infinite cluster indicates that
the two power laws can describe the filtration coefficients close and
Table 2
Pore size distributions evaluated from grain size distributions and experiment using far away from the percolation threshold (Yuan et al., 2012), as
two methods and estimated power-law exponents. follows:
   a  
Pore size distribution properties Medium 30/125 k / fl  fc fl ! fc
PTM Monte Carlo   b  
ð9Þ
k / 1  fl f l >> f c
Mean (lm) 5.88 5.30
Std (lm) 4.02 0.81 Considering that the effluent concentration close to the thresh-
l 1.58 1.66 old is extremely low and undetectable under the laboratory condi-
r 0.62 0.15
tions, we only discuss the power law relations for filter coefficients
Exponent b 0.87 0.18
far away from the percolation threshold.
4 B. Ding et al. / Chemical Engineering Science 175 (2018) 1–11

Fig. 1. Cluster structure of the viable pathway, filtration backbone, and branch bonds in the diagonal square lattice (all bonds in the figure are larger than the particle size).
Color: green, filtration backbone; purple, branch bonds; black, dangling bonds; red, bonds at the entrance connected to filtration backbone; and blue, bonds at the entrance
connected to the dangling bonds). (For interpretation of the references to colour in this figure legend, the reader is referred to the web version of this article.)

4. Network model setup
A model of straining-dominant DBF based on 3D network is
similar to that in 2D lattice (Ding et al., 2015a; Yuan et al., 2012).
The 2D and 3D models differ in terms of network structure and
influence of gravity. In this work, we briefly review the procedure
of establishing numerical DBF model on 3D network. Several 3D
lattices were used to represent porous media, consisting of inter-
connected capillaries that represent the pores. The radii rp of the
cylindrical capillaries are distributed homogeneously in space,
and the PSD is the same as that used in the experiment. The fluxes
in pores are determined by solving the pressure equations accord-
ing to Poiseuille and Kirchhoff rules. The particles are injected into
the network one by one, and only one particle randomly walks on
the lattice each time. These configurations focus on calculating the
capture probability of staining for dilute suspensions. The indepen-
dent walks of different particles also make parallel computing pos-
sible. The flow-biased random walk and the two capture schemes
were applied to move and capture the particles through the net-
work. The effluence probability (Ce/C0) and the filtration coefficient
can be obtained by injecting large number of particles into the
network.
The influence of gravity on the flow field is considered in 3D
networks. Three 3D lattices with interconnected microtubules are
used to describe random porous media, such as 3D Voronoi, Hexag-
onal Close Packing (HCP), and body-centered cubic (BCC). The 3D
Voronoi lattice contains 3868 nodes and 7686 bonds and has a
coordination number of 4. The 3D Voronoi lattice was obtained
by employing Matlab function mpt_voronoi in The Multi-
Parametric Toolbox (MPT 2.6.3, http://people.ee.ethz.ch/~mpt/
2/support.php). The seed points for Voronoi cell generation are
the ball’s center from a packed cylinder of spherical granules.
The coordinate of these seed points can be obtained by applying
discrete element method simulation with open-source software
Yade (http://yade-dem.org/doc/) from a given granule size. When
bounding polytope (cylindrical boundary) is provided, the Voronoi
cells generated by the function will be bounded by this polytope.
The coordination numbers of HCP and BCC are six and eight,
respectively. The size of HCP and BCC lattice is 40 40 60 nodes.
Periodic boundary conditions are applied to avoid surface effects.
Fig. 2 shows the schematic of the 3D network structures. The node
structures of BCC and 3D Voronoi lattice are shown in Figs. 3 and 4,
respectively.
In this study, the bonds (cylindrical capillaries) of BCC and HCP
are considered to have a constant length. The bond lengths of 3D
Voronoi lattices can be calculated using the 3D coordinates of the
nodes at both ends of the bond. The volumes of each node and
bond are:
Vn ¼ 4=3  p  r3n ð10Þ
Vb ¼ p  r2p  Lb ð11Þ
In these equations, Vn and Vb are the volumes of each node and
bond respectively; rn and rp are the radii of the node and bond
respectively; and Lb is the length of the bond. The combined vol-
umes of the bonds and nodes make up the total void space of the
porous media and can be used to calculate porosity (voidage).
The conductivity of each microtubule is determined by its
radius. When ignoring the effects of microtubule intersections,
the flux in each microtubule is achieved from the Poiseuille law
as follows:
pr4p
ql ¼  ðDp  qgl sin hÞ; ð12Þ
8gl
 B. Ding et al. / Chemical Engineering Science 175 (2018) 1–11 5

2.5

1.5

0.5

0
0

2 3
2
3 1
0

Fig. 2. Schematic of three 3D networks.

where ql is the flow in the capillary, g is the suspension dynamic the suspension and the particles is almost the same. The resultant
viscosity, rp is the capillary radius, l is the length of a capillary, Dp force of the buoyancy and gravity of the particles is almost zero.
is the applied pressure difference, q is the density of suspension, Therefore, the effect of buoyancy force acting on the particles can
g is the gravitational acceleration, and h is the angle between a cap- be neglected.
illary and horizontal direction. In this work, the densities of 0.1 M Similar to the filtration coefficient for experiment
NaOH qL (density of suspension) and polystyrene latex micro- k ¼  lnðC e =C 0 Þ=L0 , the network filtration coefficient can be calcu-
spheres qp are 1043 and 1047 kg/m3, respectively. The density of lated as follows:
6 B. Ding et al. / Chemical Engineering Science 175 (2018) 1–11

Fig. 3. Scheme of three element structures of 3D Voronoi network.

Fig. 4. Schemes of the path types of BCC lattice.

ln ðC e =C 0 Þ In this work, two capture schemes, namely, the Minimum Cap-

kl ¼ kl cos ha ¼  ð13Þ
Nl ture Scheme (MINCAP) and the Maximum Capture
Scheme (MAXCAP), are applied in the simulation. The MAXCAP
where Nl = L0/(lcosha) is the minimum number of pores connecting has same suppositions as traditional straining theory (Rege and
the inlet and the outlet, kl is the network filtration coefficient, and Fogler, 1987) theory. A particle moves following flow-biased rules
ha is the average angle between a pore and flow direction.
 B. Ding et al. / Chemical Engineering Science 175 (2018) 1–11 7

and cannot jump against the flow direction [Fig. 5(A)]. In the MIN-
CAP, a particle cannot be captured if the capillary has a large radius
and exiting flow [Fig. 5(B)]. In contrast to that in MAXCAP, in MIN-
CAP, the particle can jump against the flow direction and selects it
to penetrate when a viable microtubule is connected to the node.
Additional details on the establishment of numerical network
model can be found in Yuan et al. (2012), Ding et al. (2015a).
5. Modeling results and discussion
The effects of simulation parameters such as PSD, coordination
number of network, and capture scheme were discussed to identify
differences between the simulation results and laboratory test
data.
5.1. Validation of power law
The filtration coefficient with the given particle radius and
porous media from the straining-dominant filtration simulation
in the 3D network and (1-fl⁄) are plotted in logarithmic scales
to validate the power-law relation between the fraction of flow
through the small pores (1-fl⁄) (the average capture probability)
and the filtration coefficients (Fig. 6). The straight lines prove
the presence of power-law relations between (1-fl⁄) and the filtra-
tion coefficients. The power-law exponents estimated using Eq.
(8) (estimated slopes) based on the simulation data are listed in
Table 3. The power-law exponent estimated from simulation data
with MINCAP is larger than that of MAXCAP for the same PSD
parameters.

Fig. 5. Schematic of two capture schemes. (A) Maximum capture scheme: [The probability of choosing a jump direction from viable capillaries is distributed in accordance
with the flows in them. If the particle selects a capillary with a smaller radius, then it is captured]. (B) Minimum capture scheme: [A neighboring capillary is not viable if its
size is smaller than the particle. The probability of choosing a jump direction from viable capillaries is distributed in accordance with the flows in them. If a particle does not
have a viable capillary to jump through, then it is captured] (captured particles are marked with ‘X’).

Fig. 6. (A) Normalized effluent concentration and (B) filtration coefficients in logarithm scales from different 3D lattices under two capture schemes (PSD estimated by Monte
Carlo method, medium 30/125).
8 B. Ding et al. / Chemical Engineering Science 175 (2018) 1–11

Table 3
Exponents (slopes) evaluated from simulation of filtration coefficients on 3D lattices
in logarithm scales with PSD parameters estimated by two methods.
Lattice Maximum capture Minimum capture
scheme scheme
Monte Carlo PTM Monte Carlo PTM
Voronoi (z = 4) 0.85 0.48 1.08 0.47
HCP (z = 6) 1.00 0.80 1.19 1.00
BCC (z = 8) 1.03 0.82 3.08 1.39
5.2. Effect of lattice coordinate number
When PSD parameters are estimated by Monte Carlo method,
the variation regularity of the exponents and normalization efflu-
ent concentration are the same in 2D and 3D networks. The expo-
nents of the MINCAP increase with increasing coordination
number z. By contrast, the exponents of the MAXCAP do
not change significantly with increasing coordination number z.
Meanwhile, in PSD parameters estimated by PTM method, the
exponents of MINCAP and MAXCAP significantly increase with
increasing coordination number z. We also found that the expo-
nents and effluent concentrations under MINCAP and MAXCAP
are almost the same in the Voronoi network but vary greatly in
the other networks.
The results are illustrated in Fig. 6 and Table 4. Adopting a sim-
ilar approach with the 2D network, we analyzed the influence of
coordination number on the 3D network simulation results. As
shown in Fig. 6, large pores allow the particles to pass through.
The other are regarded as small pores. After calculation by sam-
pling methods, the liquid flux through the large pore is twice the
flux of the small pore when PSD parameters are estimated by
Monte Carlo method. However, the pores exhibit a broad distribu-
tion when PSD parameters are estimated by PTM method. With the
same sampling methods, the liquid flow of the large pore is sixfold
of the flux through the small pore. The calculation is simplified
with the assumption that the fractions of various path types in
the network are equal.
As shown in Table 4, the capture probability of particles for the
three 3D lattices in the two capture schemes can be calculated
using PSD parameters estimated by Monte Carlo method. For the
HCP lattice in the MAXCAP, the capture probability values of the
flow-biased concept for four path types (A–D) are 1, 1/2, 1/5, and
0, respectively. Moreover, the total capture probability is 0.43
(1.7/4), similar to that in triangular lattice (Ding et al., 2015a).
The path types of BCC lattice are shown in Fig. 4. For the BCC lattice
in the MAXCAP, the capture probabilities of five path types (A–E)
are 1, 3/5, 1/3, 1/7, and 0, and the total capture probability is
0.42. The node types of 3D Voronoi lattice are shown in Fig. 3.
The node type II is the same as square lattice. The total capture
probability values for the MAXCAP and MINCAP are 0.44 and 1/3,
respectively. The results are consistent with the triangular because
the node type I of 3D Voronoi possesses three outflow channels.
However, the inflow channel of this node is only one. If blocked,
then no difference is observed between the maximum and mini-
mum capture mechanisms. Therefore, the total capture probability
for the MAXCAP of this node is 0.71 (0.5 + 0.5 0.425 = 0.71),
whereas that for the MINCAP of this node is 0.63 (0.5
+ 0.5 0.25 = 0.63). The total capture probability is 0.5 because
node type III of 3D Voronoi has only one outflow channel, regard-
less of the capture mechanism used. After classifying the three
node structure types of 3D Voronoi lattice, the proportions of three
types of nodes are approximately 10%, 80%, and 10%. Finally, we
calculated the total capture probability with weighted-average
method and obtained the values of 0.47 and 0.38 for MAXCAP
and MINCAP, respectively. The total capture probability slightly
differs among the three 3D lattices in the MAXCAP. This result
agrees with the findings presented in Fig. 6, wherein the Ce/C0
and exponents do not significantly change with increasing coordi-
nation number z in the MAXCAP. However, the difference in total
capture probability is relatively large in the MINCAP. Moreover,
with the same method, the total capture probabilities achieved
for HCP, 3D Voronoi, and BCC lattices in the MINCAP are 1/4,
0.38, and 1/5, respectively. Fig. 6 shows the normalized effluent
concentrations; the exponents significantly increase with increas-
ing coordination number z in the MINCAP.
With the same method, the total capture probabilities for these
3D networks can be calculated when PSD parameters are estimated
by PTM method. Table 5 shows the probability of particles being
captured in pore space for the three 3D lattices in the two capture
schemes when PSD parameters are estimated by PTM method. As
shown in Table 5, the differences in the total capture probability
of 3D Voronoi between the MAXCAP (0.42) and MINCAP (0.38)
are not significant. This result agrees with the findings presented
in Fig. 7, wherein the Ce/C0 and exponents do not significantly
change between the MAXCAP and MINCAP in 3D Voronoi. Fig. 7
shows that the normalized effluent concentrations of HCP (z = 6)
are close to that of BCC (z = 8) in the MAXCAP. The power-law
exponent of HCP (0.80) is also very similar to that of BCC (0.82).
This result may be due to the similar total capture probability of
these two lattices in the MAXCAP (0.34 and 0.32, respectively).
At the same time, the normalized effluent concentrations of 3D
Voronoi considerably differ those of BCC (or HCP) in the MAXCAP
because of the large differences in their capture probabilities.

Table 4
Probability of particles captured in pore space for three 3D lattices in two capture schemes (PSD estimated by Monte Carlo method).

Lattice type Path type weight Maximum capture scheme Minimum capture scheme
Capture probability Total Capture probability Total
HCP (z = 6) A 1 1 0.43 1 1/4
B 1 2/(2 + 1+1) 0
C 1 1/(2 + 2+1) 0
D 1 0 0
Voronoi (z = 4) (1) 10% 0.71 0.47 0.63 0.38
(2) 80% 0.44 1/3
(3) 10% 0.50 0.50
BCC (z = 8) A 1 1 0.42 1 1/5
B 1 3/(2 + 1 + 1 + 1) 0
C 1 2/(2 + 2+1 + 1) 0
D 1 1/(2 + 2 + 2 + 1) 0
E 1 0 0
 B. Ding et al. / Chemical Engineering Science 175 (2018) 1–11 9

Table 5
Probability of particles captured in pore space for three 3D lattices in two capture schemes (PSD estimated by PTM method).

Lattice type Path type weight Maximum capture scheme Minimum capture scheme
Capture probability Total Capture probability Total
HCP (z = 6) A 1 1 0.33 1 1/4
B 1 2/(6 + 1 + 1) 0
C 1 1/(6 + 6 + 1) 0
D 1 0 0
Voronoi (z = 4) (1) 10% 0.66 0.42 0.63 0.38
(2) 80% 0.38 1/3
(3) 10% 0.50 0.50
BCC (z = 8) A 1 1 0.31 1 1/5
B 1 3/(6 + 1 + 1 + 1) 0
C 1 2/(6 + 6 + 1 + 1) 0
D 1 1/(6 + 6 + 6 + 1) 0
E 1 0 0

Fig. 7. (A) Normalized effluent concentration and (B) filtration coefficients in logarithm scales from different 3D lattices under two capture schemes (PSD estimated by PTM
method, medium 30/125).

5.3. Comparison with experimental results
The experimental data sets used for model validation refer to
the short-term injection of monosized particles performed in pack-
ing glass granules (laboratory test data of the medium 30/125
(Yuan et al., 2012)).
For 2D network, the simulated Ce/C0 and exponents agree with
the laboratory test data when the simulation was performed in
square lattices (z = 4) by using MINCAP and PSD estimated by
PTM method (Ding et al., 2015a). Meanwhile, for 3D network, the
simulated Ce/C0 and exponents are consistent with the laboratory
test when the simulation was performed in HCP lattices (z = 6) or
BCC lattice (z = 8) by using MAXCAP and PSD estimated by PTM
method. The power-law exponents in the simulation are
0.91 ± 0.02 (square), 0.80 ± 0.01 (HCP), and 0.88 ± 0.03 (BCC),
which are consistent with the experimental value of 0.87 ± 0.28.
The total capture probability of the HCP lattice (0.33) or the BCC
lattice (0.31) in the MAXCAP is very close to that of the square lat-
tice (1/3) in the MINCAP. In the 3D Voronoi lattice, the coordina-
tion number is 4, and the total capture probability values of
MAXCAP (0.42) and MINCAP (0.38) are largely different from the
former lattices, such as HCP, BCC, and square.
These results imply that if the total capture probability of differ-
ent lattices with the same PSD is not quite different, then the sim-
ulated normalized effluent concentration is basically the same. To
verify this inference, we performed the simulation on HCP lattice
and 2D triangular lattice, which have similar node structure and
channel type and thus possess the same total capture probability.
Many studies demonstrate that an increase in the column length
causes the particle concentration of effluent to decrease (Yuan
et al., 2012; Ding et al., 2015a; Rege and Fogler, 1987). Therefore,
the simulation was performed in the MAXCAP by using HCP lattice
(40 40 60) and triangular lattice (200 60) and PSD estimated
by PTM. As shown in Fig. 8, the simulated Ce/C0 and exponents of
the HCP lattice are consistent with those of the 2D triangular lat-
tice. The power-law exponent of HCP (0.80 ± 0.01) is also similar
10 B. Ding et al. / Chemical Engineering Science 175 (2018) 1–11
Fig. 8. (A) Normalized effluent concentration and (B) filtration coefficients in logarithm scales from HCP and triangular lattices with maximum capture scheme (PSD
estimated by PTM method, medium 30/125).
to that of the 2D triangular lattice (0.90 ± 0.01) and consistent with
the experimental exponent (0.87 ± 0.28).
6. Conclusion
The simulated Ce/C0 for various rs agrees with the power-law
formula. The exponent and Ce/C0 of the MINCAP should be larger
than that of the MAXCAP on the same lattice with the same PSD.
With the same PSD parameters, Ce/C0 and exponents increase with
increasing coordination number z. This trend is particularly signif-
icant in the MINCAP. The change in PSD parameters alters the flux
and weight of path type linked to node, thereby affecting the
numerical simulation of staining-dominate DBF on networks.
Particle total capture probability was estimated for different 2D
and 3D networks with the same PSD. If different networks have the
same total capture probability, then the simulated Ce/C0 is basi-
cally the same. By contrast, the different total capture probabilities
lead to significant variation in Ce/C0 on different lattices.
The simulation results are consistent with the laboratory test
data if proper simulation conditions are used. For laboratory test
data of the medium 30/125, various 3D networks were simulated.
The simulated Ce/C0 and exponents agree with the laboratory test
when the simulation was performed in the MAXCAP by using BCC
lattice (z = 8) and PSD estimated by PTM method. The power-law
exponent of the simulation (0.88 ± 0.03) is consistent with the
experimental value (0.87 ± 0.28) in the range of permitted errors.
Acknowledgments
The authors would like to thank Zhang Jian for valuable discus-
sion and suggestions.
References
Allen, H., 2009. Relevance of percolation theory to power-law behavior of dynamic
processes including transport in disordered media. Complexity 15 (2), 13–27.
Bai, H., Cochet, N., Drelich, A., et al., 2016. Comparison of transport between two
bacteria in saturated porous media with distinct pore size distribution. RSC Adv.
6 (18), 14602–14614.
Battin, T.J., Sengschmitt, D., 1999. Linking sediment biofilms, hydrodynamics, and
river bed clogging: evidence from a large river. Microb. Ecol. 37 (3), 185–196.
Bell, J.D., Deckmyn, P., Davies, A.G., 1996. Percolating clusters on Voronoi lattices
and the relationship to particle fouling on filters. Zeitschrift für Angewandte
Mathematik und Mechanik, 599.
Berkowitz, B., Balberg, I., 1992. Percolation approach to the problem of hydraulic
conductivity in porous media. Transp. Porous Media 9 (3), 275–286.
Berkowitz, B., Ewing, R.P., 1998. Percolation theory and network modeling
applications in soil physics. Surv. Geophys. 19 (1), 23–72.
Broadbent, S.R., Hammersley, J.M., 1957. Percolation processes.
Chalk, P., Gooding, N., Hutten, S., et al., 2012. Pore size distribution from challenge
coreflood testing by colloidal flow. Chem. Eng. Res. Des. 90 (1), 63–77.
Chi, T., Payatakes, A.C., 1979. Advances in deep bed filtration. AIChE J. 25 (5), 737–
759.
Civan, F., 2015. Reservoir formation damage: fundamentals, modeling, assessment,
and mitigation. Herd Health Environ. Res. Design J. 7 (2), 60–77.
Ding, Y., 2010. Modeling formation damage for flow simulations at reservoir scale.
Spe J. 15 (3), 737–750.
Ding, B., Chaolin, L., 2015. Estimation of pore size distribution of porous media
based on percolation theory. CIESC J.
Ding, B., Li, C., Zhang, M., et al., 2015a. Effects of pore size distribution and
coordination number on the prediction of filtration coefficients for straining
from percolation theory. Chem. Eng. Sci. 127, 40–51.
Ding, B., Li, C., Dong, X., 2015b. Percolation-based model for straining-dominant
deep bed filtration. Sep. Purif. Technol. 147, 82–89.
Ewing, R.P., Gupta, S.C., 1993. Percolation and permeability in partially structured
networks. Water Resour. Res. 29 (9), 3179–3188.
Golden, K.M., 1997. Percolation Models for Porous Media. Springer, New York.
Harvey, R.W., Kinner, N.E., Dan, M.D., et al., 1993. Role of physical heterogeneity in
the interpretation of small-scale laboratory and field observations of bacteria,
microbial-sized microsphere, and bromide transport through aquifer
sediments. Water Resour. Res. 29 (8), 2713–2721.
Heertjes, P.M., Lerk, C.F., 1967. The functioning of deep bed filters. Trans. Inst. Chem.
Eng. 45, 129–145.
Islam, M., Jia, X., Fairweather, M., et al., 2011. Prediction of the permeability and
filtration performance of packed beds. Comput. Aided Chem. Eng. 29 (2), 1045–
1049.
Ives, K.J., 1980. Deep bed filtration: theory and practice. Filtr. Sep. 17 (2), 157–166.
Jia, X., Gregory, J., Williams, R., 1998. Particle Deposition & Aggregation:
Measurement, Modelling and Simulation. Elsevier Science.
Kaiser, C., 1997. A directed percolation model for clogging in a porous medium with
small inhomogeneities. Transp. Porous Media 26 (2), 133–146.
Larson, R.G., Davis, H.T., 1982. Conducting backbone in percolating Bethe lattices. J.
Phys. C: Solid State Phys. 15 (11), 2327.
Madaeni, S.S., Sedeh, S.N., Nobili, M.D., 2006. Ultrafiltration of humic substances in
the presence of protein and metal ions. Transp. Porous Media 65 (3), 469–484.
 B. Ding et al. / Chemical Engineering Science 175 (2018) 1–11
Maroudas, A., Eisenklam, P., 1965. Clarification of suspensions: a study of particle
deposition in granular media: Part I—Some observations on particle deposition.
Chem. Eng. Sci. 20 (10), 867–873.
O’Melia, C., 1985. Particles, pretreatment, and performance in water filtration. J.
Environ. Eng. 111 (6), 874–890.
Perrier, E.M.A., Bird, N.R.A., Rieutord, T.B., 2010. Percolation properties of 3-D
multiscale pore networks: how connectivity controls soil filtration processes.
Biogeosci. Discuss 7 (2), 2997–3018.
Rege, S.D., Fogler, H.S., 1987. Network model for straining dominated particle
entrapment in porous media. Chem. Eng. Sci. 42 (7), 1553–1564.
Rege, S.D., Fogler, H.S., 1988. A network model for deep bed filtration of solid
particles and emulsion drops. AIChE J. 34 (11), 1761–1772.
Reyes, S., Jensen, K.F., 1985. Estimation of effective transport coefficients in porous
solids based on percolation concepts. Chem. Eng. Sci. 40 (9), 1723–1734.
Ruthven, D.M., 2009. Diffusion through porous media: ultrafiltration, membrane
permeation and molecular sieving. Diffusion Fundam.
Selyakov, V.I., Kadet, V., 1997. Percolation Models for Transport in Porous Media:
with Applications to Reservoir Engineering. Springer.
Shi, Y., Lee, Y.T., Kim, A.S., 2011. Permeability calculation of sphere-packed porous
media using dissipative particle dynamics. Desalination Water Treat. 34 (34),
277–283.
11
Stein, M., 1987. Large sample properties of simulations using latin hypercube
sampling. Technometrics 29 (2), 143–151.
Sun, D., Meng, M., Yin, Y., et al., 2016. Highly selective, regenerated ion-sieve
microfiltration porous membrane for targeted separation of Li+. J. Porous
Mater., 1–9
Tien, C., Payatakes, A.C., 1979. Advances in deep bed filtration. AIChE J. 25 (5), 737–
759.
Tufenkji, N., 2007. Modeling microbial transport in porous media: Traditional
approaches and recent developments. Adv. Water Resour. 30 (6–7), 1455–1469.
Vicente, J., Wyart, Y., Moulin, P., 2013. Characterization (two-dimensionalthree-
dimensional) of ceramic microfiltration membrane by synchrotron radiation:
new and abraded membranes. J. Porous Media 16 (6), 537–545.
Yao, K., Habibian, M.T., O’Melia, C.R., 1971. Water and waste water filtration.
Concepts and applications. Environ. Sci. Technol. 5 (11), 1105–1112.
You, Z., Badalyan, A., Bedrikovetsky, P., 2013. Size-exclusion colloidal transport in
porous media-stochastic modeling and experimental study. SPE J. 08, 620–633.
Yuan, H., Shapiro, A., You, Z., et al., 2012. Estimating filtration coefficients for
straining from percolation and random walk theories. Chem. Eng. J. 210 (6), 63–
73.