Particulate Science and Technology

An International Journal

ISSN: 0272-6351 (Print) 1548-0046 (Online) Journal homepage: https://www.tandfonline.com/loi/upst20

Numerical simulation of particle migration in

suspension flow through heterogeneous porous
media

Niloy De & Anugrah Singh

To cite this article: Niloy De & Anugrah Singh (2019): Numerical simulation of particle migration in
suspension flow through heterogeneous porous media, Particulate Science and Technology, DOI:
10.1080/02726351.2019.1651806

To link to this article: https://doi.org/10.1080/02726351.2019.1651806

Published online: 12 Aug 2019.

Submit your article to this journal

View Crossmark data

Full Terms & Conditions of access and use can be found at

https://www.tandfonline.com/action/journalInformation?journalCode=upst20
PARTICULATE SCIENCE AND TECHNOLOGY
https://doi.org/10.1080/02726351.2019.1651806

Numerical simulation of particle migration in suspension flow through

heterogeneous porous media
Niloy Dea and Anugrah Singhb
a
Wolfson School of Mechanical, Electrical and Manufacturing Engineering, Loughborough University, Loughborough, UK; bDepartment of
Chemical Engineering, Indian Institute of Technology Guwahati, Guwahati, India

ABSTRACT KEYWORDS
Particle migration in porous media is relevant to many applications in chemical and petroleum Suspension transport; fine
engineering. The hydrodynamic interactions between the suspension and grain particles affect the migration; dual porosity
distribution and transport of particles during suspension flow through porous media. Continuum system; Stokes flow
models based on Darcy’s law are widely used to investigate the macroscopic properties of the sys-
tem. However, these models are unable to capture the pore scale dynamics and microstructure.
We have carried out particle migration studies during transport of suspension in heterogeneous
porous media using the Stokesian dynamics simulations. The velocity and concentration distribu-
tions were first studied in a large scale constant porosity system, but with different concentration
of particles in the two adjacent porous layers. It was found that particle migration across the inter-
face is higher when the difference of concentration is large. Simulations were also carried out for
constant areal concentration of particles, where the adjacent layers had different porosity. A high
rate of migration near the interface was observed for the larger difference in porosity. It was
observed that the hydrodynamic interactions are strongly affected by the increase in the number
of grain particles of porous media in comparison to that of suspended particles.

1. Introduction other hand, finite element method (FEM) has also been used
to simulate multiphase flow in porous media for a long time.
Multiphase flow through heterogeneous porous media finds
Multiscale FEM (MsFEM) method has been proposed to
wide-scale application in subsurface processes and packed
study flow in low permeability porous media which requires
bed reactors. Recently, suspension flow in heterogeneous
much shorter computation effort in terms of CPU memory
porous media has been widely explored for the research of
and time (Ye, Xue, and Xie 2004; Yao et al. 2012; Zhang,
shale gas reservoir (Kazmouz, Giusti, and Mastorakos 2016;
Cao et al. 2017; El Amin et al. 2017; Lorinczi et al. 2017; Yao, and Xue 2016). Low permeability regions are more pro-
Lidong et al. 2018), geothermal energy (McDowell, Zarrouk, nounced in fractured oil and gas reservoirs; hence computa-
and Clarke 2016; You et al. 2016; Kaya and Zarrouk 2017), tional models are being built to characterize the flow
enhance oil recovery (Jia, Tsau, and Barati 2018; Wei et al. through fracture formations. More recently, fluid flow
2018), and packed bed reactors (Schulze, Nikrityuk, and through hydraulic fracture has been studied by the Extended
Meyer 2015; Sobti and Wanchoo 2015). Most of these studies Finite Element Method, where structure and geometry-inde-
have considered heterogeneous porous media, where porosity pendent meshes can be used (Mohammadnejad and Khoei
and permeability are highly non-isotropic. Macroscopic 2013; Arzanfudi, Al-Khoury, and Sluys 2014). Discrete
simulation of multiphase flow in such large scale porous Fracture Model (DFM) is a widely known technique for
media does not audit the microscale hydrodynamic interac- simulation of multiphase flow through hydraulic fracture
tions. In continuum-based simulations, the finite volume (Kim and Deo 2000; Li et al. 2016). In this method, unstruc-
method using the fully implicit method, which permits simu- tured meshes are generated explicitly for each fracture, there-
lations with large time steps, was considered in recent simu- fore, computational cost becomes high (Bosma et al. 2017;
lations (Wu and Qin 2009; Hamon, Mallison, and Tchelepi Gl€aser et al. 2017). An improvement of the accuracy was
2018). Due to its ease of implementation, the Implicit achieved by incorporating the structured grid during discret-
Pressure Explicit Saturation (IMPES) method is also com- ization and the lower number of grids reduce the computa-
mon in reservoir simulation (Horgue et al. 2015; Soares et al. tional time (Mi et al. 2017). The method efficiently audits
2016; Zhang et al. 2018). The high accuracy and efficiency of the flow characteristic in large scale fracture reservoirs.
this method come at a high computational cost due to the Subsequently, another work has simulated the multi-layer
non-linear characteristic of governing equations. On the porous shale gas reservoir by Multi-Porosity model, which is

CONTACT Anugrah Singh [email protected] Department of Chemical Engineering, Indian Institute of Technology Guwahati, Guwahati 781039, India.
Color versions of one or more of the figures in the article can be found online at www.tandfonline.com/upst.
ß 2019 Taylor & Francis Group, LLC
2 N. DE AND A. SINGH

Figure 1. Initial configuration of three systems with constant porosity (/ ¼ 0.9) but with different concentration of particles in the top and bottom layers. (a)
ulow ¼ 0.2, uhigh ¼ 0.25 (b), ulow ¼ 0.2, uhigh ¼ 0.30, (c) ulow ¼ 0.2, uhigh ¼ 0.40. The low concentration layer is at the bottom and high concentration layer at the top.

upscaling of earlier mentioned DFM (Yan et al. 2018). works (Bossis and Brady 1984; Brady and Bossis 1988).
Though these models are capable of simulating heteroge- Primitively, SD method was formulated to simulate the
neous porous media at macroscale, pore-scale networking, dynamics of suspended particles in creeping flow regime
and heterogeneity cannot be investigated. with Re
1, where interparticle, bulk, and hydrodynamic
In some recent studies, the pore-scale analyses of hydro- forces were taken into account (Bossis and Brady 1984).
dynamic interactions in porous media have been carried out Thereafter, SD was implemented to study shear of suspen-
by the Lattice Boltzmann Method (Wang et al. 2017; Warda sion of particles bounded between two parallel plates by dis-
et al. 2017). An excellent review of the Lattice Boltzmann cretizing the particles into patches and assuming a uniform
Method (LBM) model for multiphase flow application has distribution of force density in each patch (Durlofsky, Brady,
been presented by Liu et al. (2016). Besides, diffusion at the and Bossis 1987). Hydrodynamic interactions were calculated
interface of oil-water system in fractured porous media was between suspended particles and patches. Nott and Brady
investigated by LBM and the results captured the physical (Nott and Brady 1994) considered the wall to be made up of
process reasonably well (Gunde et al. 2013). Moreover, LBM spheres attached together, creating a bumpy wall. In progres-
simulation effectively captures hydrodynamic interaction at sion, Singh and Nott (2000) suggested using exact sphere
microscale and agreement with experimental result suggests wall resistance for interactions of the suspended particles
that discrete analysis is better than continuum models at with wall particles instead of “bumpy” wall implementation
pore scale (Di Palma et al. 2017; Fakhari et al. 2018). (Nott and Brady 1994). Though SD method was upgraded to
However, significant efforts have not been devoted to study simulate suspension flow in the plane channel, yet its appli-
the migration of particles in heterogeneous porous media. cation is limited to small scale isotropic porous network.
In this work, particle transport in heterogeneous porous In this context, SD method was implemented to audit the
media has been studied by Stokesian dynamics simulation flow dynamics of suspensions in porous media and the
technique, a particle level numerical simulation approach. results were in good agreement with Brinkman approach
Stokesian dynamics (SD) is a molecular dynamics type (Brinkman 1949). Subsequently, a simple version of SD was
simulation method which accurately captures the multi-body deployed to investigate the hydrodynamic interactions of a
hydrodynamic interactions. The main difficulty in dynamic falling sphere through a regular arrangement of fixed par-
simulation of dense suspension is the singular hydrodynamic ticles, but high computational cost deferred its implementa-
interactions due to the lubrication interactions between the tion at large scale (Hassonjee, Ganatos, and Pfeffer 1988).
close particle pairs. SD method uses the local and pairwise However, Vitthal’s (Vitthal and Sharma 1992) work on par-
additive nature of lubrication interactions to avoid this diffi- ticle dynamics in filtration process by SD method presented
culty. It also captures many body hydrodynamic interaction clogging, bridging, and particle deposition. Another study by
by inverting the mobility matrix. However, the inversion of Lee and Koplik (1999) reported microscopic analysis of par-
the mobility matrix for a system with a large number of par- ticle motion through a periodic porous media using the SD
ticles is computationally expensive and this limits the appli- method. They formed a porous media with fixed particles in
cation of the SD method to only a few hundred particles. two dimensions as well as in three dimensions and the sus-
The detailed formulation of SD was laid out in previous pended particle was allowed to pass through the media. A
 PARTICULATE SCIENCE AND TECHNOLOGY 3

Table 1. Configuration of three systems with porosity (/) ¼ 0.9 and suspended particle concentration of different layers.
Particle concentration in the top layer (uhigh) Particle concentration in the bottom layer (ulow)
System 1 0.25 0.20
System 2 0.30 0.20
System 3 0.40 0.20

Figure 2. Steady state particle concentration contour for the systems described in Figure 1. Grain particles are also shown by the circles.

statistical analysis of particle trajectory was given for differ- will be more realistic for wide-scale applications in chemical
ent lattice constant and different initial position of the par- and petroleum engineering.
ticle. Another work reported the dependency of the media Earlier studies with SD simulation dealt with lubrication
permeability on the internal structure of the porous media forces between two moving particles and a short-range
using the SD technique (Otomo and Harada 2011, 2013). repulsive force was introduced between them to overcome
More recently, permeability and diffusivity effect in bi-dis- the numerical overlapping. In this work, the repulsive force
perse porous media has been studied using the SD method was also considered between the suspended particles and
(Wang and Brady 2015). Undoubtedly, these studies are grain particles. The study begins with heterogenous porous
helpful to understand the flow dynamics in porous media at media with constant porosity, however, the distribution of
the microscale, but the study of heterogeneous porous media suspended particles is non-uniform. In the second set of
4 N. DE AND A. SINGH

Figure 3. Steady state particle velocity contours for the configurations as described in Figure 1.

simulations, the areal concentration of suspended particles Brownian suspensions in a Newtonian liquid in the regime
was kept uniform but the porosity of the two adjacent layers of creeping flow, i.e. Rep ¼ q c_ a2 =g ! 0 and Pe ¼
was different. The effects of particle concentration and sys- c_ a2 =2D ! 1; where q and g are the density and viscosity
tem porosity on particle migration is discussed. of the fluid, a is the radius of the sphere, c_ is a nominal
The article is organized in the following structure. The shear rate, and D is the Brownian diffusivity of an isolated
mathematical formulation of Stokesian Dynamics is particle. Rep is the Reynolds number based on particle size,
described in Section 2. Simulation results for particle trans- and Pe is the Peclet number.
port in the dual-porosity system are presented in Section 3, The method computes hydrodynamic interactions among
followed by conclusion in Section 4. particles as a sum of long-range mobility interactions and
short-range lubrication interactions. The long-range interac-
tions are first computed as a multipole moment expansion
2. Mathematical formulation
about the particle’s centers of the force densities acting on
In this section, a brief overview of mathematical formulation their surfaces. The zeroth moment is the net force, the first
along with a modification to deal suspension flow through moment is decomposed into stresslets (symmetric part) and
porous media is presented. We have considered non- torque (anti-symmetric part) and the higher moments of the
 PARTICULATE SCIENCE AND TECHNOLOGY 5

The hydrodynamic forces and torque (F) and stresslets

(S) on the particles are related to their velocities and the
imposed deformation rate by grand resistance tensor:
   
F Uhui
¼ R (2)
S hei
Where,
 
RFU RFE
R¼ (3)
RSU RSE
The 6 N (for N number of particles) vector U in Equation
(2) denotes the translational and rotational velocities of all
particles, hui is the average suspension velocity, and hei is its
rate of strain. Here,RFU is the configuration dependent resist-
ance matrix that gives the hydrodynamic forces on the par-
ticles due to their motion relative to the fluid. Equation (3)
is computed for each time step, then substituted in Equation
(2) to calculate particle velocities. RFE is the configuration
dependent resistance matrix that correlates the hydrodynamic
forces on the particles to the rate of strain tensor. RSU corre-
lates stresslet with the velocity of the particle, and RSE corre-
lates stresslets with the rate of strain tensor.
In this work, we have followed the method of Nott and
Brady (1994) to fix the position of the grain particles in a
porous media. In this procedure, iteration starts with pre-
scribed values of the velocities of wall particles and the
forces on the interior particles. Thereafter, the velocities of
the interior particles and the forces on the wall particles are
calculated. As the flow is driven by the pressure gradient in
the suspension, the bulk shear rate,hei; is set to zero. Also, a
fixed value is assigned to average suspension velocity,hui;
which generates a pressure gradient to satisfy the global con-
servation of momentum. Moreover, grain particles in porous
media are stationary ðU W ¼ 0Þ; so the force required to
hold them stationary will balance the pressure gradient
needed to maintain the non-zero suspension average velocity
(hui). So, the hydrodynamic force on grain particles
(denoted with superscript W) and suspended particles
(denoted with superscript S) in porous media are computed
Figure 4. Instantaneous snapshot showing the particle positions for the cases
described in Figure 1. Grain particles are shown by filled circles, and suspended by the following relations:
particle by open circles.

FW ¼ RWWFU  huiRFU  U hui ;
WS S
(4)
series are discarded. This truncated multipole expansion

FS ¼ RSS
FU  U hui þ RFU  hui:
S SW
(5)
along with Faxen’s laws is used to form the grand mobility
tensor, M: Once computed, the grand mobility tensor is Here, RWWFU represents the interactions between grain par-
inverted to obtain the far-field approximation of grand ticles, RWS
FU for grain-suspended particles and so-on.
resistance tensor, R1 : Inversion of grand mobility tensor Neglecting inertia, the equations of the balance of hydro-
captures many body interactions. Finally, lubrication inter- dynamic force on suspended particles at low Reynolds num-
action is added to resistance tensor in a pair-wise additive ber is simply:
manner to obtain grand resistance tensor, R: F S þ F e ¼ 0: (6)
1
R¼M þ R2B R1
2B ; (1) Here, Fe is the configuration dependent non-hydrodynamic
forces acting on the suspended particles. Upon determining
where, R2B is near field two-body resistance tensor for all par-
U S for a given configuration, the position of the particle is
ticle pairs, and R1
2B ; which is far field part of R2B ; is subtracted updated by integrating the following equation:
to avoid double counting. Ewald summation technique is used
to compute interaction in an infinite extent porous media, dxS
1

¼ hui þ RSS
FU  F e þ RSW
FU  hui : (7)
which is generated by periodic replication of a unit cell with dt
randomly arranged particles throughout all spaces.
6 N. DE AND A. SINGH

Figure 5. Average particle concentration (c) plotted against time for the three configurations as described in Figure 1.

External forces required to hold grain particles stationary very small. The form of repulsive force has already been
are equal and opposite to hydrodynamic force calculated presented in earlier studies (Durlofsky, Brady, and Bossis
from Equation (4): 1987; Nott and Brady 1994),


SS 1

sese
W
Fext ¼ F W ¼ RWWFU  hui þ RFU  RFU
WS
 F e þ RSWFU  hui
r
Fab ¼ F0 ^e ab ; (9)
1  ese
(8)
where Fabr
is the repulsive force between sphere a and sphere
This is an improvement over the concept of using “gluing b; F0 is a constant related to the magnitude of the force, s
force” to balance hydrodynamic and other non-hydro- is related to the range of the force, e is the spacing between
dynamic forces acting on grain particles (Lee and two sphere a and b; and ^e ab is the unit vector connecting
Koplik 1999). the sphere centers. Results in the following sections are rep-
Another difficulty in dynamic simulation with finite-sized resented in dimensionless unit: all lengths are scaled by par-
particles is the frequent overlapping of particles while per- ticle radius, a; which is set to 1 for this study; velocity by
forming the trajectory computations using the finite time hui; time with a=hui and all the forces with 6Pgahui; where
step. To prevent this, we introduced an interparticle repul- g is fluid viscosity,a is particle diameter, hui is average sus-
sive force which also provides a qualitative model of other pension velocity. The model is applicable for particle size
non-hydrodynamic effects when the interparticle distance is smaller than the granular particles but larger than the size
 PARTICULATE SCIENCE AND TECHNOLOGY 7

Figure 6. Initial configuration of three systems with two layers of different porosity. The initial particle concentration (u) in both the layers was equal to 0.2. (a)
/high¼0.9, /low¼0.7, (b) /high¼0.9, /low¼0.75, (c) /high¼0.9, /low¼0.8. In all the three cases, the top layer has low porosity and the bottom layer has
high porosity.

Table 2. Configuration of three systems with suspended particle concentration achieved. The condition when the mean particle position in
(u)¼0.2 and porosity of separate layers. the flow normal direction remained unchanged with time
(/high) (Top layer) (/low) (Bottom layer) was considered a steady-state for this study. The time to
System 1 0.90 0.70 achieve steady-state strongly depends upon the number of
System 2 0.90 0.75
System 3 0.90 0.80
particles. Particle fraction ðcÞ for a system of N particles are
calculated as follows:

considered in macromolecular dynamics simulation. Particle fraction ðcÞ

Nonetheless, the model can simulate the particle size belong- Number of particles at any time
ing to geological flows and filtration systems. ¼
Total number of particle present initially

3. Results and discussion In the following section, the effect of suspended particle
concentration on particle migration has been discussed. This
3.1. Initial configuration of particles in porous media is being followed by the effect of porosity variation on par-
In this section, the flow characteristic of suspended particles ticle migration.
in multi-layer porous media has been discussed. Simulations
were performed to investigate the flow of monodisperse 3.2. Effect of particle concentration on migration
non-Brownian and neutrally buoyant suspensions. Porous
media were generated by Monte-Carlo simulation technique; Initially, a porous media with / ¼ 0.9 was formed and the
the grain particles (shown by filled circles in Figure 1) were medium was filled with suspended particles to get the areal
first arranged in a regular array in the unit cell of size 30 in concentration, u ¼ 0.2. Thereafter, another medium was
both X and Y direction, and then small random displace- formed in a similar process for u ¼ 0.25 and the later one
ments were applied until uniform distribution was achieved. was placed at the top of earlier one to form a double layer
Next, void spaces in porous media were inserted with sus- system (Figure 1(a)). The periodic replication of this unit
pended particles (open circle in Figure 1) to get desired cell along all direction generates an infinite large multilayer
areal concentration, by random number generator program porous media. In the course of our interest, more suspended
in MATLAB. Two such cells were kept adjacent to each particles were introduced into the top cell, while the lower
other to yield a dual-porosity system of size 30 in X and 60 one was kept at constant u (see Table 1). Hence, two more
in Y direction. Simulations were accomplished to audit the double-layer porous media were achieved with u ¼ 0.3, 0.4
alternation of areal particle concentration (u), and porosity at the top layer (uhigh) and u ¼ 0.2 at the bottom layer
(/) with the homologous unit cell, where particle average (ulow) (see Figure 1(b,c)). Simulations were carried out to
velocity hui is set to 1. In all our simulations, a short-range study the effect of areal particle concentration on particle
r migration. The dimensionless time step for performing the
repulsive force of kind Fab (see Equation (9)) with a range
(s) of 100 and the magnitude F0 s ¼ 1 was used. Simulations simulations was chosen to be 0.001. Simulations were per-
were performed until the steady-state condition was formed until a steady-state condition was achieved. In all
8 N. DE AND A. SINGH

Figure 7. Steady state particle concentration contour for the systems described in Figure 6.

the simulations, the steady-state was achieved by the end of interaction, eventually it increased the dispersion of particles.
dimensionless time equal to 5000. The higher diffusion forced the particles of the top layer to
Steady-state microstructure of these systems was analyzed cross the interface. Due to the migration of particles, effective
by contour plots, which were furnished by discretizing the cell porosity at the bottom layer was reduced, therefore, particles
both in X and Y direction. In the figures, the open circles are did not get enough free space to move. As a result, a large
the grain particles of porous media. The dotted line indicates number of particles deposited on the surface of grain particles,
the interface between the two layers. The concentration con- which is pronounced by increasing concentration around
tours of these three systems are presented in Figure 2(a–c). grains (Figure 2(c)). We observed that for a small difference
The high concentration at the interface indicates the migration in the concentration of the two layers, the cross-stream migra-
of particles from higher concentration to lower concentration tion is less and the particles mostly follow the streamlines in
region. Also, the interfacial concentration increased with the the horizontal direction (Figure 2(a)) compared to other two
value of uhigh because cross-stream migration is higher due to systems (Figure 2(b,c)).
the large difference in u between the two layers. At a higher Velocity contours of these systems are shown in Figure
difference of u, particles experienced higher hydrodynamic 3(a–c). The narrower zones of both layers have low average
 PARTICULATE SCIENCE AND TECHNOLOGY 9

Figure 8. Steady state particle velocity contours for the configurations as described in Figure 6.

velocity, while channels are reflected by high average vel- observed in the lower cell for a high value of uhigh
ocity. Furthermore, the maximum average velocity is higher (Figure 4(c)), whereas significant changes are not visible in
at larger uhigh (Figure 3(c)) than the other two systems the top layer in each system (Figure 4(a–c)).
(Figure 3(a,b)). Drift velocity is high at larger uhigh, which An in-depth analysis of particle migration is presented by
increases the average velocity of the particles. The average particle fraction for the total time span of the simulation,
velocity at the interface also increases due to the high value which is shown in Figure 5(a–c). The higher values of c at a
of drift velocity. So, the particle migration is obvious from region of ulow, for all the systems, depict the migration across
the high average velocity at the interface. the interface. Initially, particles experienced a net migration
The steady-state configurations of these systems are flux due to the difference in concentration between the two
shown by the snapshots of particles at different times layers, which drives some particles to cross the interface.
(Figure 4(a–c)). It can be seen that migration increases the These surges in migration are indicated by the initial spike in
number of particles in ulow region and the situation is more the plot of c. However, the concentration difference between
prominent in Figure 4(c). However, both the layers held the two layers is decreasing due to migration and the driving
nearly the same number of particles at the lowest uhigh force also gets diminished. At steady state, very few particles
(Figure 4(a)). Here, migration is less due to the lower cross the interface, which is reflected by a smaller difference
difference in u. Furthermore, many cluster formations are in the concentration of the two regions. The overlapping in
10 N. DE AND A. SINGH

Figure 9. Average particle concentration (c) plotted against time for the three configurations as described in Figure 6.

certain time steps is observed in Figure 5(a), which is happen- concentration (u) of 0.2. In the course of our interest, simu-
ing because the lower value of the driving force is unable to lations were performed for /low¼0.7 and 0.75 (see Table 2
control the crossing of some particles across the interface in for all system configuration). Initial configurations of these
reverse direction. The difference in the steady-state concentra- systems are shown in Figure 6(a,b). Random distribution of
tion of particles in the two porous layers increases with the grain particles in low porosity system created a stratified
increase in uhigh – ulow (Figure 5(b,c)). system and it was impossible to create a random distribution
It is apparent from the above discussion that particle of suspended particles throughout the system. So, a few
migration through porous media strongly depends upon sus- grain particles were clubbed together in hexagonal formation
pended particle concentration. Next, we discuss the migra- and random distribution was applied to those structures to
tion of particles in a dual-porosity media with a uniform create /low region (see Figure 6(a,b)).
initial concentration in the layers of different porosity. Particle concentration contours of these systems at steady
state are presented in Figure 7(a–c), which were furnished
in a similar way as described earlier. At low values of /low,
3.3. Effect of porosity gradient on particle migration particles did not get enough free space to move, also they
experienced more hydrodynamic interactions with grain par-
As described earlier, a porous media with / ¼ 0.8 was ticles (Figure 7(a,b)). In these systems, several trapped zones
formed and it was placed at the top of a media with / ¼ 0.9 (high concentration around the grain) can be seen.
(Figure 6(c)). So, the system became a double porosity sys- Although few particles flow through channels while lubrica-
tem with the low porous region (/low) at the top and high tion forces relaunched them from the trapped zone.
porous region (/high) at the bottom. The void spaces were However, in case of the high value of /low (Figure 7(c)),
flushed with suspended particles to get uniform particle particles are mostly trapped in layers, although free space in
 PARTICULATE SCIENCE AND TECHNOLOGY 11

Figure 10. Instantaneous snapshot showing the particle positions for the cases described in Figure 6. Grain particles are shown by filled circles and suspended par-
ticle by open circles.

the /low region is more than the other two systems. This generated and this force propelled some particles to cross
happened due to the stratified configuration of the media, the interface in reverse direction over time. These few num-
whereas particles at the interface got enough free space to bers of particles create the overlapping of c at a steady state
flow and experienced less interaction with grains. Therefore, (Figure 9(a,c)). Conversely, overlapping is almost absent in
for high difference of / (Figure 7(a,b)), particles were Figure 9(b). This is because of the balance of hydrodynamic
migrating towards the high porous region and those par- force at the interface; the concentration gradient is not
ticles followed the same periodic path with time without any strong enough to create backflow. Also, the magnitude of
further migration. Besides, the difference of / between the the fluctuation of c is very low in Figure 9(c) because of the
two layers is too low to create cross migration, which is sup- low difference of / which allows only a few particles to
ported by the low concentration at the interface cross the interface. However, the migration of particles at
(Figure 7(c)). the interface for these systems is not clear from the contour
Velocity contours of these systems are shown in Figure plots. Therefore, an investigation was carried out with the
8(a–c). Highly trapped zones for all the three systems have snapshots of these systems at steady-state condition, which
almost zero average velocity that is different from the nature is shown in Figure 10(a–c). As discussed earlier, trapped
of concentration contour plots. Besides, channelized zones and chain formations are clearly visible for all systems. It
have very low average velocity, which is more pronounced can be seen that particles are trapped in the layer for the
in Figure 8(c). Here, effective porosity does not reduce sig- largest value of /low (Figure 10(c)), which agrees with the
nificantly, as the migration of particles is very low than the analysis of contour plots. However, particles are mostly free-
other two systems. So, the dispersion of particles is very flowing in the lower zone of this system, which increases
high, which resulted in a slightly higher average velocity. the average velocity at this zone. On another note, particle
Whereas, the high average velocity at the interface depicted migrations are clearly visible in Figure 10(a,b). However, the
the free-flowing particles at the interface (Figure 8(a,b)). migrated particles are mostly flowing at the interface (Figure
However, average velocity is almost zero at the interface 10(a)) and sometimes, the center of these particles cross the
when the difference of / is low (Figure 8(c)), which indi- interface. These occasional crossing of the interface are the
cates particles are either trapped or totally absent at inter- reason for the frequent overlapping of c, whereas, in Figure
face. Furthermore, the maximum average velocity also 10(b), migrated particles are dispersed into the /high region,
increased due to high drift velocity which was thus particles are almost absent at the interface. In this case,
described earlier. there is no overlapping of c as discussed earlier. Although
We have also analyzed the dynamics of particles with the for the largest value of /low, there are no moving particles
plot of particle fraction for the total span of simulation time at the interface, few particles are observed at the lower
(Figure 9(a–c)). Initially, the high driving force pulls a large boundary of the /high region (Figure 10(c)). This lower
number of particles from low porous region to high porous boundary becomes the interface because of periodic repeti-
region, which is evident from the initial spike of c for all tion of a unit cell in all direction. The particles at the lower
the systems. These migrations reduce the effective porosity boundary are actually crossing the interface, as a result,
of /high region but increase the concentration gradient overlapping of c is observed. Therefore, it can be said that
between the two layers. So, a reverse driving force was suspension flow through a dual-porosity system is strongly
12 N. DE AND A. SINGH

affected by the porosity of each layer and the transient con- Cao, C., Q. Zhao, C. Gao, J. Sun, J. Xu, P. Zhang, and L. Zhang. 2017.
centration gradient also plays an important role in particle Discrete fracture model with Multi-field coupling transport for shale
gas reservoirs. Journal of Petroleum Science and Engineering 158:
migration. This study describes the migration of the par-
107–19. doi: 10.1016/j.petrol.2017.08.033.
ticles in heterogeneous porous media from a microscopic Di Palma, P.R., A. Parmigiani, C. Huber, N. Guyennon, and P. Viotti.
point of view. 2017. Pore-scale simulations of concentration tails in heterogeneous
porous media. Journal of Contaminant Hydrology 205:47–56. doi: 10.
1016/j.jconhyd.2017.08.003.
4. Conclusion Durlofsky, L., J. F. Brady, and G. Bossis. 1987. Dynamic simulation of
hydrodynamically interacting particles. Journal of Fluid Mechanics
The current work presented a comprehensive analysis of 180 (1):21–49. Cambridge, UK: Cambridge University Press. doi: 10.
particle transport in multi-layer porous media using the 1017/S002211208700171X.
Stokesian dynamics simulation technique. In a large scale El Amin, M. F., S. Amir, A. Salama, D. Urozayev, and S. Sun. 2017.
Comparative study of Shale-gas production using Single- and Dual-
system, we have fixed the grain particle and modeled the
continuum approaches. Journal of Petroleum Science and
suspended particles as freely moving in the porous space. Engineering 157:894–905. doi: 10.1016/j.petrol.2017.07.011.
Migration of particles across the interface between the two Fakhari, A., Y. Li, D. Bolster, and K T. Christensen. 2018. A Phase-field
layers was observed. The objective of the present work was lattice Boltzmann model for simulating multiphase flows in porous
to study suspension transport in porous media. We believe media: Application and comparison to experiments of CO2 seques-
tration at pore scale. Advances in Water Resources 114:119–34. doi:
that our study provides a microscopic view of particle trans-
10.1016/j.advwatres.2018.02.005.
port that has wide application in the filtration problem and Gl€aser, D., R. Helmig, B. Flemisch, and H. Class. 2017. A discrete frac-
migration of fine. The simulations demonstrated that par- ture model for two-phase flow in fractured porous media. Advances
ticles migrate from higher concentration to lower concentra- in Water Resources 110:335–348. doi: 10.1016/j.advwatres.2017.10.
tion layer as well as low porous to the higher porous layer. 031.
Gunde, A., T. Babadagli, S S. Roy, and S K. Mitra. 2013. Pore-scale
Strong hydrodynamic interactions between the grain and
interfacial dynamics and Oil-water relative permeabilities of capillary
suspended particles are also affecting the migration whose driven counter-current flow in fractured porous media. Journal of
strength depends on the porosity difference between the two Petroleum Science and Engineering 103:106–14. doi: 10.1016/j.petrol.
layers. Furthermore, interesting phenomena like channel for- 2013.02.005.
mation of suspended particles, trapping, and bridging in the Hamon, F.P., B.T. Mallison, and H.A. Tchelepi. 2018. Implicit hybrid
upwinding for Two-phase flow in heterogeneous porous media with
low porosity regions are described with contour plots. In the
buoyancy and capillarity. Computer Methods in Applied Mechanics
future, it would be interesting to calculate the average prop- and Engineering 331:701–27. doi: 10.1016/j.cma.2017.10.008.
erties like permeability reduction in a fractured porous Hassonjee, Q., P. Ganatos, and R. Pfeffer. 1988. A strong-interaction
media, retention coefficient of porous channels of a filter theory for the motion of arbitrary three-dimensional clusters of
medium. These properties can be calculated by using the spherical particles at low Reynolds number. Journal of Fluid
resistant matrix which is formulated during the simulation. Mechanics 197:1–37. Cambridge, UK: Cambridge University Press
doi: 10.1017/S0022112088003155.
Another extension of this work would be to simulate sus- Horgue, P., C. Soulaine, J. Franc, R. Guibert, and G. Debenest. 2015.
pension transport in a three-dimensional porous media An Open-source toolbox for multiphase flow in porous media.
which will be more relevant to a real scale system. This can Computer Physics Communications 187:217–26. doi: 10.1016/j.cpc.
help us in the design of gravel pack for sand control in pro- 2014.10.005.
ducing oil wells. Jia, B., J.-S. Tsau, and R. Barati. 2018. Role of molecular diffusion in
heterogeneous, naturally fractured shale reservoirs during CO2
Huff-n-Puff. Journal of Petroleum Science and Engineering 164:
31–42. doi: 10.1016/j.petrol.2018.01.032.
Acknowledgment Kaya, E., and S. J. Zarrouk. 2017. Reinjection of greenhouse gases into
We are thankful to Indian Institute of Technology, Guwahati for pro- geothermal reservoirs. International Journal of Greenhouse Gas
viding the computing facilities to carry out the simulations. Control 67:111–29. doi: 10.1016/j.ijggc.2017.10.015.
Kazmouz, S. J., A. Giusti, and E. Mastorakos. 2016. Numerical simula-
tion of shale gas flow in Three-dimensional fractured porous media.
References Journal of Unconventional Oil and Gas Resources 16:90–112. doi: 10.
1016/j.juogr.2016.10.002.
Arzanfudi, M.M., R. Al-Khoury, and L.J. Sluys. 2014. A computational Kim, J.-G., and M. D. Deo. 2000. Finite element, discrete-fracture
model for fluid leakage in heterogeneous layered porous media. model for multiphase flow in porous media. American Institute of
Advances in Water Resources 73:214–26. doi: 10.1016/j.advwatres. Chemical Engineers Journal 46 (6):1120–30. doi: 10.1002/aic.
2014.08.016. 690460604.
Bosma, S., H. Hajibeygi, M. Tene, and H.A. Tchelepi. 2017. Multiscale Lee, J., and J. Koplik. 1999. Microscopic motion of particles flowing
finite volume method for discrete fracture modeling on unstructured through a porous medium. Physics of Fluids 11 (1):76–87. doi: 10.
grids (MS-DFM). Journal of Computational Physics 351:145–64. doi: 1063/1.869921.
10.1016/j.jcp.2017.09.032. Li, J.-J., H. Jiang, L. Jiang, L. Mi, T. Xu, and C. Liu. 2016. Varying-
Bossis, G., and J. F. Brady. 1984. Dynamic simulation of sheared sus- scale shale gas flow: Discrete Fracture Networks (DFN) based
pensions. I. general method. The Journal of Chemical Physics 80 numerical simulation. Particulate Science and Technology 34 (5):
(10):5141–54. doi: 10.1063/1.446585. 557–64. doi: 10.1080/02726351.2015.1099065.
Brady, J., and G. Bossis. 1988. Stokesian dynamics. Annual Review of Lidong, M., J. Hanqiao, M. Shanbo, L. Junjian, P. Yanli, and C. Liu.
Fluid Mechanics 20 (1):111–157. doi: 10.1146/annurev.fluid.20.1.111. 2018. Numerical simulation study of shale gas reservoir with stress-
Brinkman, H C. 1949. A calculation of the viscous force exerted by a dependent fracture conductivity using multiscale discrete fracture
flowing fluid on a dense swarm of particles. Flow, Turbulence and network model. Particulate Science and Technology 36 (2):202–11.
Combustion 1 (1):27. doi: 10.1007/BF02120313. doi: 10.1080/02726351.2016.1241844.
 PARTICULATE SCIENCE AND TECHNOLOGY 13

Liu, H., Q. Kang, C. R. Leonardi, S. Schmieschek, A. Narvaez, B. D. Vitthal, S., and M. M. Sharma. 1992. A stokesian dynamics model for
Jones, J. R. Williams, A. J. Valocchi, and J. Harting. 2016. particle deposition and bridging in granular media. Journal of
Multiphase lattice Boltzmann simulations for porous media applica- Colloid and Interface Science 153 (2):314–36. doi: 10.1016/0021-
tions. Computational Geosciences 20 (4):777–805. doi: 10.1007/ 9797(92)90325-G.
s10596-015-9542-3. Wang, J., Q. Kang, Y. Wang, R. Pawar, and S S. Rahman. 2017.
Lorinczi, P., A D. Burns, D. Lesnic, Q J. Fisher, A J. Crook, C. Simulation of gas flow in Micro-porous media with the regularized lat-
Grattoni, and K. Rybalcenko. 2017. Modelling of gas flow in shale tice Boltzmann method. Fuel 205:232–46. doi: 10.1016/j.fuel.2017.05.080.
using a finite volume method. Applied Mathematical Modelling 49: Wang, M., and J. F. Brady. 2015. Short-time transport properties of
394–414. doi: 10.1016/j.apm.2017.05.009. bidisperse suspensions and porous media: A stokesian dynamics
McDowell, A., S J. Zarrouk, and R. Clarke. 2016. Modelling viscous fin- study. The Journal of Chemical Physics 142 (9):094901. doi: 10.1063/
gering during reinjection in geothermal reservoirs. Geothermics 64: 1.4913518.
220–34. doi: 10.1016/j.geothermics.2016.06.003. Warda, H A., S H. Haddara, E M. Wahba, and M. Sedahmed. 2017.
Mi, L., B. Yan, H. Jiang, C. An, Y. Wang, and J. Killough. 2017. An Lattice boltzmann simulations of the capillary pressure bump phe-
enhanced discrete fracture network model to simulate complex frac- nomenon in heterogeneous porous media. Journal of Petroleum
ture distribution. Journal of Petroleum Science and Engineering 156: Science and Engineering 157:558–69. doi: 10.1016/j.petrol.2017.06.058.
484–96. doi: 10.1016/j.petrol.2017.06.035. Wei, P., W. Pu, L. Sun, Y. Pu, S. Wang, and Z. Fang. 2018. Oil recov-
Mohammadnejad, T., and A R. Khoei. 2013. An extended finite elem- ery enhancement in low permeable and severe heterogeneous oil res-
ent method for hydraulic fracture propagation in deformable porous ervoirs via gas and foam flooding. Journal of Petroleum Science and
media with the cohesive crack model. Finite Elements in Analysis Engineering 163:340–8. doi: 10.1016/j.petrol.2018.01.011.
and Design 73:77–95. doi: 10.1016/j.finel.2013.05.005. Wu, Y.-S., and G. Qin. 2009. A generalized numerical approach for
Nott, P. R., and J. F. Brady. 1994. Pressure-driven flow of suspensions: modeling multiphase flow and transport in fractured porous media.
Communications in Computational Physics 6 (1):85–108. https://
simulation and theory. Journal of Fluid Mechanics 275 (6):157–99.
www.scopus.com/inward/record.uri?eid=2-s2.0-67650941657&partner
doi: 10.1017/S0022112094002326.
ID=40&md5=fb1b6246cce3f269cb398d064f2cad96. doi: 10.4208/cicp.
Otomo, R., and S. Harada. 2011. End effect on permeability of loose
2009.v6.p85.
particulate bed having different internal structures. Particulate
Yan, B., L. Mi, Y. Wang, H. Tang, C. An, and J E. Killough. 2018.
Science and Technology 29 (1):2–13. doi: 10.1080/02726351.2010.
Multi-porosity multi-physics compositional simulation for gas stor-
488715.
age and transport in highly heterogeneous shales. Journal of
Otomo, R., and S. Harada. 2013. Fluid permeability in stratified uncon-
Petroleum Science and Engineering 160:498–509. doi: 10.1016/j.pet-
solidated particulate bed. Transport in Porous Media 96 (3):439–56. rol.2017.10.081.
doi: 10.1007/s11242-012-0098-6. Yao, J., N. Zhang, Z. Huang, and Y. Wang. 2012. Multiscale mixed
Schulze, S., P. A. Nikrityuk, and B. Meyer. 2015. Porosity distribution finite element numerical simulation for heterogeneous reservoirs.
in monodisperse and polydisperse fixed beds and its impact on the Shiyou Xuebao/Acta Petrolei Sinica 33 (3):442–7. https://www.sco-
fluid flow. Particulate Science and Technology 33 (1):23–33. doi: 10. pus.com/inward/record.uri?eid=2-s2.0-84863503214&partnerID=40&
1080/02726351.2014.923960. md5=ca021280cbcbe55ef95e6ac15e4acbf2.
Singh, A., and P. R. Nott. 2000. Normal stresses and microstructure in Ye, S., Y. Xue, and C. Xie. 2004. Application of the multiscale finite
bounded sheared suspensions via stokesian dynamics simulations. element method to flow in heterogeneous porous media. Water
Journal of Fluid Mechanics 412:279–301. doi: 10.1017/ Resources Research 40 (9):W0920201–9. doi: 10.1029/2003WR
S0022112000008375. 002914.
Soares, S. R., L. P. R. Maciel, W. R. Brito, and C. D. K. Elisiario. 2016. You, Z., Y. Yang, A. Badalyan, P. Bedrikovetsky, and M. Hand. 2016.
A higher resolution Edge-based finite volume method for the simu- Mathematical modelling of fines migration in geothermal reservoirs.
lation of the oil–water displacement in heterogeneous and aniso- Geothermics 59:123–33. doi: 10.1016/j.geothermics.2015.05.008.
tropic porous media using a modified Impes method. International Zhang, N., J. Yao, and S. Xue. 2016. A modified multiscale finite elem-
Journal for Numerical Methods in Fluids 82 (12):953–78. Hoboken, ent method for nonlinear flow in reservoirs. Journal of Petroleum
NJ: Wiley-Blackwell. doi: 10.1002/fld.4254. Science and Engineering 137:97–106. doi: 10.1016/j.petrol.2015.11.003.
Sobti, A., and R K. Wanchoo. 2015. Creeping flow of viscoelastic fluid Zhang, N., J. Yao, S. Xue, C. Redondo, G. Rubio, E. Valero, J. Zhu,
through a packed bed: Effect of particle shape and porosity. and J. Ma. 2018. On the efficiency of the IMPES method for two
Particulate Science and Technology 33 (5):463–471. doi: 10.1080/ phase flow problems in porous media. Journal of Petroleum Science
02726351.2015.1010759. and Engineering 137 (1):427–36. doi: 10.1016/j.petrol.2015.11.003.