infrastructures

Article
A Coupled CFD–DEM Simulation of Slurry
Infiltration and Filter Cake Formation during Slurry
Shield Tunneling
Zixin Zhang 1,2, * and Tong Yin 1,2
1 Department of Geotechnical Engineering, College of Civil Engineering, Tongji University,
Shanghai 200092, China; yintong@tongji.edu.cn
2 Key Laboratory of Geotechnical Engineering, Ministry of Education, Tongji University,
Shanghai 200092, China
* Correspondence: zxzhang@tongji.edu.cn




Received: 22 May 2018; Accepted: 13 June 2018; Published: 15 June 2018


Abstract: Tunneling in highly permeable ground using a slurry shield machine can be challenging
because it is difficult to form the so-called filter cake on the tunnel face to transport the support
pressure. Consequently, destructive accidents might happen, such as face instability and water inrush.
How to form an efficient filter cake in time is crucial during engineering practice, especially in ground
with high permeability. Various theoretical and experimental analyses regarding the formation of
filter cakes have been conducted. However, due to the complexity of this problem, which has to
incorporate the mechanical and hydraulic behaviors of the fluid–solid mixture system, few numerical
simulations are found in the literature. In this paper, with the aid of a newly developed numerical
tool, a coupled CFD (computational fluid dynamics)–DEM (discrete element method) simulation is
established to study the slurry infiltration and filter cake formation during slurry shield tunneling.
The slurry infiltration process is simulated by modelling the scheme of the infiltration column test,
in which sedimentation behaviors of slurry particles are captured and compared with experimental
results. The results show that the sedimentation behaviors of the slurry particles and filter cake
formation phenomenon are well captured by simulations and in accordance with the experiments,
which indicates the robustness of the coupled CFD–DEM simulation used in present work.

Keywords: slurry shield tunneling; filter cake; coupled CFD–DEM simulation

1. Introduction
Slurry shield machines are widely used in mechanized tunneling projects, especially in undersea
and cross-river tunnels [1]. During construction, the tunnel face is stabilized by pressurized slurry. In
order to transport the support pressure efficiently, a so-called filter cake must be formed on the tunnel
face under slurry pressure. However, when the soil on the tunnel face is highly permeable, such as
sand and gravel, the slurry would infiltrate into the ground instead of being retained on the tunnel
face. Consequently, a loss of the support pressure might occur, leading to geological hazards such
as tunnel face collapse and underground water inrush. Thus, a comprehensive study of the slurry
infiltration behavior is crucial to the safety and economic efficiency of the tunneling project.
Regarding the theoretical studies, one could refer to the conventional cake infiltration theory, in
which the slurry infiltrates across a membrane medium and is retained on the medium face to form a
filter cake as shown in Figure 1 [2]. According to conventional cake infiltration theory, both the fluid
phase and solid phase are treated as a continuum. Slurry flow equations are established based on
the continuity condition and momentum balance. Incorporating constitutive relations between the
solid pressure and permeability and the solid fraction of the cake, the partial differential slurry flow

Infrastructures 2018, 3, 15; doi:10.3390/infrastructures3020015 www.mdpi.com/journal/infrastructures

Infrastructures
Infrastructures2018,
2018,3,
3,x15FOR PEER REVIEW 22of
of11
11

solid pressure and permeability and the solid fraction of the cake, the partial differential slurry flow
equationscould
equations couldbe be analytically
analyticallysolved
solvedunder
undercertain
certaininitial
initialboundary
boundaryconditions
conditionsandandassumptions.
assumptions.
The analytical solutions could be used to acquire the pressure distribution within
The analytical solutions could be used to acquire the pressure distribution within the system the system and
and
time-dependentgrowth
time-dependent growth ofofthe
thefilter
filter cake.
cake.However,
However, the
the assumptions
assumptions such
such as
as ignoring
ignoringthe
thevelocity
velocityof
of
the solid phase and constant fluid velocity during the whole process impose further
the solid phase and constant fluid velocity during the whole process impose further limitations to its limitations to
its application
application [3].[3]. Besides
Besides this,
this, thethe continuous
continuous treatment
treatment ofof fluidand
fluid andsolid
solidphase
phaseforforsimplifying
simplifyingthe
the
problem is
problem is not
not in
in accordance
accordance with
with the
the highly
highly polydisperse
polydisperse nature
nature of
of slurry.
slurry.

Figure
Figure1.1.Schematic
Schematicdiagram
diagramof
ofthe
theconventional
conventionalslurry
slurryinfiltration
infiltrationprocess
process(adapted
(adapted from
from [2]).
[2]).

Various experimental studies have also been carried out, in most of which the slurry infiltration
Various experimental studies have also been carried out, in most of which the slurry infiltration
column test was adopted.
column test was adopted.
Krause [4] proposed an empirical equation for calculating the infiltration length of slurry from
Krause [4] proposed an empirical equation for calculating the infiltration length of slurry from
various laboratory tests. The equation showed that the infiltration length was related to the applied
various laboratory tests. The equation showed that the infiltration length was related to the applied
pressure, slurry viscosity, and particle size of the ground. However, this equation could lead to large
pressure, slurry viscosity, and particle size of the ground. However, this equation could lead to large
error due to the difficulty in determining the empirical parameters. Talmon et al. [5] conducted a
error due to the difficulty in determining the empirical parameters. Talmon et al. [5] conducted a
plastering test using a Perspex cylindrical cell. According to the results, he concluded that the slurry
plastering test using a Perspex cylindrical cell. According to the results, he concluded that the slurry
invasion process could be separated into two stages: mud spurt and consolidation. A semi-empirical
invasion process could be separated into two stages: mud spurt and consolidation. A semi-empirical
equation was also proposed to evaluate the time-dependent cake growth. Min et al. [6] investigated
equation was also proposed to evaluate the time-dependent cake growth. Min et al. [6] investigated
the effects of clay content on the filter cake properties such as thickness and permeability. The results
the effects of clay content on the filter cake properties such as thickness and permeability. The results
showed that clay content was a controlling parameter during slurry infiltration. Min et al. [7] used
showed that clay content was a controlling parameter during slurry infiltration. Min et al. [7] used an
an infiltration column test apparatus and conducted 45 slurry infiltration tests as shown in Figure 2,
infiltration column test apparatus and conducted 45 slurry infiltration tests as shown in Figure 2, and
and different infiltration behaviors were observed. Their results also provided instruction and
different infiltration behaviors were observed. Their results also provided instruction and comparison
comparison of the numerical simulation, which will be further discussed in Section 3. Though
of the numerical simulation, which will be further discussed in Section 3. Though experimental
experimental investigations could obtain the slurry infiltration behavior at the end of the experiments,
investigations could obtain the slurry infiltration behavior at the end of the experiments, it is nearly
it is nearly impossible to capture the whole infiltration process. In addition, some critical pieces of
impossible to capture the whole infiltration process. In addition, some critical pieces of information
information such as the interaction between slurry particle and fluid phase and porosity distribution
such as the interaction between slurry particle and fluid phase and porosity distribution evolution
evolution with time could not be provided by the experiments either.
with time could not be provided by the experiments either.
Infrastructures 2018, 3, 15 3 of 11
Infrastructures 2018, 3, x FOR PEER REVIEW 3 of 11

Figure 2. Schematic diagram of a slurry infiltration column test (adapted from [7]).
Figure 2. Schematic diagram of a slurry infiltration column test (adapted from [7]).

In order to study the mechanism of slurry infiltration, a suitable numerical approach is

In order to
indispensable. study the mechanism
Conventional of slurrysuch
numerical methods infiltration, a suitable
as FEM (finite elementnumerical
method),approach
etc., which is
indispensable. Conventional numerical methods such as FEM (finite element method),
use a continuous mesh to simulate the solid, cannot possibly capture the interactions between slurry etc., which
use a continuous
particles and sand mesh to simulate
particles. the solid,
However, usingcannot
a purepossibly capture the
DEM simulation weinteractions between slurry
could not incorporate the
particles and sand particles. However, using a pure DEM simulation we could not
rheology properties of the slurry. To simulate a highly polydisperse system with fluid such as slurry, incorporate the
rheology
a coupledproperties
CFD andofDEM the slurry.
approachTo simulate
might be a highly polydisperse
the optimal system with
choice available fluid such
at present. as slurry,
This paper
aproposes
coupled aCFD and DEM approach might be the optimal choice available at present.
coupled CFD and DEM approach to investigate the slurry infiltration and filter cake This paper
proposes
formationabehavior
coupled during
CFD and DEM
slurry approach
shield to investigate
tunneling. the slurryapproach,
In this CFD–DEM infiltration
theand filter
solid phasecakeis
formation behavior during slurry shield tunneling. In this CFD–DEM approach,
treated as discrete slurry particles while the fluid phase is treated as a continuous Newtonian fluidthe solid phase is
treated as discrete
since these treatmentsslurry
areparticles whileregarding
more realistic the fluid the
phase is treated
highly as a continuous
polydisperse nature of Newtonian fluid
slurry. The paper
since these treatments are more realistic regarding the highly polydisperse nature
is organized as follows. Firstly, the theoretical background and methodology of the CFD and DEM of slurry. The paper
is organized
are explainedasinfollows.
SectionFirstly, the theoretical
2. Secondly, background
the numerical and methodology
implementation of theinCFD
is introduced and3.
Section DEM
The
are explained in Section 2. Secondly, the numerical implementation is introduced
slurry infiltration process is simulated by modelling the scheme of the infiltration column test. Thenin Section 3. The
slurry infiltration
the results processand
are discussed is simulated
comparedby modelling
with the scheme
the published of the infiltration
experimental results incolumn
Sectiontest. Then
4. Finally,
the results
Section are discussed
5 summarizes theand
majorcompared with of
conclusions thethepublished experimental results in Section 4. Finally,
present work.
Section 5 summarizes the major conclusions of the present work.
2. Theory and Methodology
2. Theory and Methodology
The coupled CFD–DEM simulation is based on two open source codes: LAMMPS (Large-scale
The coupled CFD–DEM simulation is based on two open source codes: LAMMPS (Large-scale
Atomic/Molecular Massively Parallel Simulator)-based DEM code named LIGGGHTS (LAMMPS
Atomic/Molecular Massively Parallel Simulator)-based DEM code named LIGGGHTS (LAMMPS
improved for general granular and granular heat transfer simulations) [8] and CFD code OpenFOAM
improved for general granular and granular heat transfer simulations) [8] and CFD code OpenFOAM
(Open Source Field Operation and Manipulation) (www.openfoam.com). The framework of the
(Open Source Field Operation and Manipulation) (www.openfoam.com). The framework of the
coupling computation of DEM and CFD, named CFDEM, written by Goniva et al. [9] is employed in
coupling computation of DEM and CFD, named CFDEM, written by Goniva et al. [9] is employed in
present paper.
present paper.
The fluid phase is described by the Navier–Stokes equations in Eulerian form, which are solved
The fluid phase is described by the Navier–Stokes equations in Eulerian form, which are solved
by OpenFOAM codes through the PISO (Pressure-Implicit with Splitting of Operators) loop, and the
by OpenFOAM codes through the PISO (Pressure-Implicit with Splitting of Operators) loop, and the
motion of particles is governed by Newton’s Laws of Motion in Lagrangian form that are calculated
motion of particles is governed by Newton’s Laws of Motion in Lagrangian form that are calculated by
by the LIGGGHTS packages. The CFDEM codes exchange the particle–fluid interaction information
the LIGGGHTS packages. The CFDEM codes exchange the particle–fluid interaction information at
at certain time step intervals and update the data accordingly (see Figure 3 [10]). The governing
certain time step intervals and update the data accordingly (see Figure 3 [10]). The governing equations
equations of each method are discussed in the following subsections.
of each method are discussed in the following subsections.
Infrastructures 2018, 3, 15 4 of 11
Infrastructures 2018, 3, x FOR PEER REVIEW 4 of 11

Figure 3. Coupling process of computational fluid dynamics (CFD)–discrete element method (DEM)
Figure 3. Coupling process of computational fluid dynamics (CFD)–discrete element method (DEM)
(adapted from [10]).
(adapted from [10]).
2.1. Governing Equations of DEM
2.1. Governing Equations of DEM
DEM, proposed by Cundall and Strack [11], is a well-validated and widely adopted method to
DEM, proposed
investigate by Cundall
the behavior and material
of granular Strack [11],
suchisasa well-validated and widelytoadopted
sand and rock. Compared method to
the conventional
investigate
continuum the behavior
method, of granular
it treats material
granular materialsuch
as a as sand and rock.
composition Compared
of various to the
individual conventional
particles, and
continuum method, it are
particle interactions treats granular
governed bymaterial
differentascontact
a composition of various
models. As long as individual
interparticleparticles,
forces areand
particle interactions
determined, are governed
the particle by different
motion could contact
be calculated models. second
by Newton’s As longlawasofinterparticle
motion: forces are
 n
determined, the particle motion could be calculated by Newton’s second law of motion:
d vi   
c

mi →
ddt
vi
  nc →
Fijc  F→i f  F→i
g
(1)
=∑ j 1 f g
mi Fijc + Fi + Fi (1)
dt  
j =1 nc 
dw
Ii → i   M ij (2)
dwdti nj c1 →
Ii = ∑ Mij   (2)
dt
where mi and I i are the mass and inertia moment of particle i ; vi and wi represent the
j =1
   → →
where mi and Ii and
translational are the mass and
angular
→
velocities; Fijc , Fi fof, particle
inertia moment g
and Fii; viare wi represent
andthe the translational
contact force, and
fluid–particle
c
→ f
→
g 
angular
interaction force,Fand
velocities; ij , Figravitational
, and Fi are the
force; and M
contact force,isfluid–particle
ij
interaction
the torque acting force,i and
on particle . gravitational
→
force; and Mthat
Note ij is the
amongtorque
theacting
forceson particle
acting i. particles, a proper determination of fluid–particle
on the
Note that among the forces acting on
interaction force is crucial to a rigorous coupled the particles,
CFD–DEMa proper determination
simulation, of fluid–particle
which will be further
discussedforce
interaction in Section 2.2. to a rigorous coupled CFD–DEM simulation, which will be further discussed
is crucial
in Section 2.2.
2.2. Governing Equations of CFD and Fluid–Particle Intercation
2.2. Governing Equations of CFD and Fluid–Particle Intercation
The fluid phase is solved in the CFD program by the locally averaged incompressible Navier–
Stokes
The equations
fluid phase [12,13]:
is solved in the CFD program by the locally averaged incompressible
Navier–Stokes equations [12,13]:  
(s U
→s   f U
→f )  0 (3)
∇·(φs Us + φ f U f ) = 0 (3)

 f U
→
f
 1 
∂φ f U f  ( f U
→ → U )  1 (p       g  F→ ) (4)
 t + ∇·(φ f U ffU f )f =  f(−∇ p + φ ff ∇·τ + φf f ρff g + Fff ) (4)
∂t ρf
→

where
where s and
φs and  f the
φ f are arevolume
the volume fraction
fraction of solid
of solid andand fluid
fluid in in
a fluid cell;U Uisf the
a fluidcell; is the fluid
fluid velocity
velocity and
→  f

is theUsolid
Us and s is velocity,
the solidwhich
velocity, which is obtained
is obtained by averaging
by averaging the velocities
the particle particle velocities
inside ainside a fluid
fluid cell; ρ f is
thecell;
fis the fluid density; p denotes the pressure gradient and  is the divergence ofthe
fluiddensity; ∇ p denotes the pressure gradient and ∇· τ is the divergence of stress tensor; g is
→ → →
gravity; and Ff is the fluid–particle interaction force. The Eulerian fields φ, Us , and Ff are all obtained
by averaging the information of Lagrangian particles.
Infrastructures 2018, 3, 15 5 of 11

→
The particle–fluid interaction force Ff consists of pressure gradient force, drag force, lift force,
and virtual mass force. The pressure gradient force is induced by the pressure gradient around a
particle, which includes the buoyancy force due to gravity and the acceleration pressure gradient force
in fluid [14]. The drag force is caused by the viscous shearing effect of fluid on the particle; the virtual
mass force is caused by relative acceleration between particle and fluid [15,16]. The lift force, including
Saffman lift forces and Magnus lift forces, accounts for the lift or upward force due to particle rotation.
In the process of particle sedimentation, particles may fall in the vertical direction with rather little
acceleration, in which the effect of lift force and virtual mass force could be neglected [17]. Hence, in
the present simulation, only the drag force and pressure gradient force are considered.
In this paper, the drag force is calculated using the empirical expression proposed by Di Felice [18]:

1 πd2 → → → →
Fd = Cd ρ f |U f − Us |(U f − Us )ε−χ+1 (5)
2 4
where d is the particle diameter, and Cd is the drag force coefficient given by

4.8 2
Cd = (0.63 + p ) (6)
Re p

→ →
where Re p = ρ f dε|U f − Us |/µ and µ is the fluid viscosity.
ε−χ+1 in Equation (5) denotes the porosity correction function to consider the presence of other
particles in the flow system, as proposed by Di Felice [17]:

(1.5−log10 Re p )2
χ = 3.7 − 0.65e− 2 . (7)

The validity of the above drag force expression has been investigated by many researchers [19–21],
who have indicated that this expression works well in flow systems with relatively low
Reynolds numbers.
The pressure gradient force is considered as the buoyancy force only since the acceleration
pressure gradient is neglected in present work. This scheme is also adopted in [22] as the acceleration
of particles is rather small in sedimentation. The buoyancy is given by

1
Fb = πρd3 g. (8)
6

3. Model Implementation
Model implementation includes three major procedures—(1) CFD meshing, (2) DEM initialization,
and (3) CFD–DEM coupling settings—which will be illustrated in the following subsections after a
brief discussion of validation.

3.1. Valiadation
Validation for particle sedimentation has already been conducted by several researchers. Most
of them validate their simulations by a benchmarking case of a single particle settling in the water.
The final settling velocity could be measured and compared with an available analytical solution. The
results show good comparability between the CFD–DEM simulation and analytical solution [20,22].
Consequently, validation will not be further discussed in this paper.

3.2. CFD Meshing

During the CFD–DEM coupling simulation, the CFD mesh works independently of the particle
motions. However, the size of the CFD mesh is important to the selection of particle size. The cell size
should be larger than two to four times the particle diameter [23] due to the nature of the algorithm of
Infrastructures 2018, 3, x FOR PEER REVIEW 6 of 11

3.2. CFD Meshing

During2018,
Infrastructures the CFD–DEM
3, 15 coupling simulation, the CFD mesh works independently of the particle 6 of 11
motions. However, the size of the CFD mesh is important to the selection of particle size. The cell size
should be larger than two to four times the particle diameter [23] due to the nature of the algorithm
thethe
of calculations of of
calculations porosity and
porosity drag
and drag force. Large
force. particles
Large would
particles wouldlead to inaccurate
lead predictions
to inaccurate of the
predictions of
porosity and drag force in terms of unresolved simulation. However, an overly
the porosity and drag force in terms of unresolved simulation. However, an overly coarse mesh coarse mesh would
decrease
would the accuracy
decrease of the CFD
the accuracy calculations.
of the Hence, aHence,
CFD calculations. properasize of thesize
proper CFDof mesh is significant
the CFD mesh is
significant to the simulation. After some trial simulations, the mesh in Figure 4 was determinedthe
to the simulation. After some trial simulations, the mesh in Figure 4 was determined to meet to
requirement
meet above and
the requirement balance
above andthe accuracy
balance and efficiency.
the accuracy and efficiency.

(a) (b)

Figure 4. The CFD mesh: (a) top view, (b) side view.
Figure 4. The CFD mesh: (a) top view, (b) side view.

3.3. DEM Initialization

3.3. DEM Initialization
The major purpose of DEM initialization is to generate a sand column, which occupies half the
The major purpose of DEM initialization is to generate a sand column, which occupies half
volume of the infiltration column. The sand particles are set to settle purely under the gravity.
the volume of the infiltration column. The sand particles are set to settle purely under the gravity.
Consequently, in this process, the interaction force is not included. As the length of the center square
Consequently, in this process, the interaction force is not included. As the length of the center square
fluid cell of the CFD mesh is 2 cm, the particle diameter of the sand column could be selected as 0.6
fluid cell of the CFD mesh is 2 cm, the particle diameter of the sand column could be selected as 0.6 cm.
cm. Forty thousand monodisperse sand particles are generated randomly inside the infiltration
Forty thousand monodisperse sand particles are generated randomly inside the infiltration column
column and then settle at the bottom with an average porosity of 0.4 as shown in Figure 5a.
and then settle at the bottom with an average porosity of 0.4 as shown in Figure 5a.
After generating the sand column, slurry particles are inserted above it. The slurry particle
After generating the sand column, slurry particles are inserted above it. The slurry particle
diameter is determined based on the experimental study by Min et al. [7]. According to their
diameter is determined based on the experimental study by Min et al. [7]. According to their
experiments, three types of filter cake are observed and termed as Type I (all grains of the slurry are
experiments, three types of filter cake are observed and termed as Type I (all grains of the slurry
clogged on the surface of soil, and a thick layer of filter cake is formed), Type II (some fine grains of
are clogged on the surface of soil, and a thick layer of filter cake is formed), Type II (some fine grains of
the slurry are penetrated into the soil, and an infiltrated zone of fine grains can be identified), and
the slurry are penetrated into the soil, and an infiltrated zone of fine grains can be identified), and Type
Type III (all grains of the slurry are penetrated into the soil, and no filter cake is formed on the surface
III (all grains of the slurry are penetrated into the soil, and no filter cake is formed on the surface of the
of the soil). The particle size ratio between sand and slurry, as the authors indicate, serves as the
soil). The particle size ratio between sand and slurry, as the authors indicate, serves as the controlling
controlling parameter of the type of filter cake. When D15 / d85  5.26 , a Type I filter cake is formed
parameter of the type of filter cake. When D15 /d85 ≤ 5.26, a Type I filter cake is formed on the soil
on
face.theWhen
soil 5.26
face.≤When 5.26
D15 /d85  D15 a/ dType
≤ 10.53, 85  10.53
II filter, cake
a Type II filterand
is formed, cake
whenis D
formed, and
15 /d85 ≥ when
10.53, no
D15 / d85  10.53 , no filter cake (Type III)
filter cake (Type III) is observed, where D 15 is observed, where D
denotes the sand particle size larger than 15% of the total
15 denotes the sand particle size
particles, and d85 is the slurry particle size larger than 85% of the total particles.
larger than 15% of the total particles, and d 85 is the slurry particle size larger than 85% of the total
particles
 Infrastructures 2018, 3, 15 7 of 11
Infrastructures 2018, 3, x FOR PEER REVIEW 7 of 11

(a) (b)

Figure 5. Generation of sand column and slurry particles: (a) sand column, (b) slurry particles.
Figure 5. Generation of sand column and slurry particles: (a) sand column, (b) slurry particles.

In accordance with the above conclusions, three sets of slurry particle size are selected to form
In accordance with the above conclusions, three sets of slurry particle size are selected to form the
the aforementioned types of filter cake. In each case, D15 / d85  3, 6,12 , respectively. Figure 5b
aforementioned types of filter cake. In each case, D15 /d85 = 3, 6, 12, respectively. Figure 5b shows the
shows the generation
generation of slurry particles.
of slurry particles.

3.4.
3.4.CFD–DEM
CFD–DEMCoupling
CouplingSettings
Settings
The
TheCFD–DEM
CFD–DEMcoupling couplingsettings
settingscould
couldbe be rather
rather complex.
complex. We Wecarried
carriedout outaaserials
serialsofof trial
trial
simulations
simulationstotodetermine
determinethe thecrucial
crucialsettings
settingssuch
suchas asDEM
DEMtime timestep,
step,coupling
couplinginterval,
interval,andandfluid
fluid
model.
model.For Forbrevity,
brevity, thethedetails
detailswill notnot
will be discussed
be discussedin this
in paper. SomeSome
this paper. key settings are listed
key settings are in Table
listed in
1.Table 1.
AsAsforforboundary
boundaryconditions
conditionsand andinitial
initialconditions,
conditions,the theinitial
initialpressure
pressureof ofthe
theinlet
inletand
andoutlet
outletare are
given a fixed value and the side is set as zero-gradient (see Figure 4b for definitions
given a fixed value and the side is set as zero-gradient (see Figure 4b for definitions of the boundaries). of the boundaries).
The
The initial velocitiesofofinlet
initial velocities inlet
and and outlet
outlet are both
are both givengiven a zero-gradient
a zero-gradient conditioncondition and aboundary
and a nonslip nonslip
boundary condition is set for
condition is set for the side boundary. the side boundary.
Note
Notethatthatififwe
weset the
set thepressure
pressure of of
thethe
inlet to be
inlet to the realreal
be the pressure
pressureof Min at el.’s
of Min experiments,
at el.’s experiments, the
velocity of the
the velocity ofparticles would
the particles be extremely
would large.
be extremely Consequently,
large. Consequently, the DEM
the DEMtimetime
step step
needs to betoset
needs be
to be at least 10 times smaller than the present setting to keep the simulation
set to be at least 10 times smaller than the present setting to keep the simulation numerically stable.numerically stable. For
computational
For computational efficiency, the pressure
efficiency, at theat
the pressure inlet
theboundary is set to
inlet boundary iszero,
set towhich
zero, means that thethat
which means slurrythe
particles settle merely
slurry particles settle under
merelythe gravity
under theand particle–fluid
gravity interaction
and particle–fluid force. Thisforce.
interaction is slightly different
This is slightly
from the experiments,
different but we focus
from the experiments, but weon focus
the sedimentation behaviorbehavior
on the sedimentation on the surface
on the of the soil
surface in this
of the soil
simulation, rather than
in this simulation, rather thethan
exact
thepressure valuesvalues
exact pressure along along
the slurry infiltration
the slurry column.
infiltration Hence,
column. this
Hence,
simplification is reasonable
this simplification within
is reasonable the scope
within of theofpresent
the scope study.
the present study.

Table
Table1.1.Some
Somekey
keyCFD–DEM
CFD–DEMsettings.
settings.

Settings Value
Settings Value
DEM time step 1 × 10−6
−6
100 1 × 10
DEM time step
Coupling interval
Coupling interval 100
Model Type
Model Type B B
VoidFractionModel
VoidFractionModel divideddivided
AverageingModel
AverageingModel dense dense
ForceModels DeFeliceDrag−6Archimedes
DeFeliceDrag
ForceModels
Fluid Viscosity 1 × 10 m2 /s
Fluid Type
Archimedes
Newtonian
Fluid Viscosity 1 × 10−6 m2/s
Fluid Type Newtonian

4. Discussion

4.1. Filter Cake Formation

Infrastructures 2018, 3, 15 8 of 11

4. Discussion
Infrastructures 2018, 3, x FOR PEER REVIEW 8 of 11
Infrastructures 2018, 3, x FOR PEER REVIEW 8 of 11
4.1. Filter Cake Formation
As
As shown
shown in in Figure
Figure 6, 6, three
three typestypes ofof filter
filter cake
cake are
are observed
observed from from numerical
numerical simulation
simulation results.
results.
of D D1515 // dd8585  33 (Figure
In As shown in Figure 6, three types of filter cake are observed from numerical simulation results.
In the
the case
case of (Figure 6a), 6a), nearly
nearly allall of
of the
the slurry
slurry particles
particles accumulate
accumulate on on the
the surface
surface
In the case of D15 /d85 = 3 (Figure 6a), nearly all of the slurry particles accumulate on the surface of
of
of the
the sand
sand column
column and form aa rather thick and dense filter cake, which corresponds to the
the Type I
the sand column andand form forma rather rather
thickthick
andand dense dense
filterfilter
cake,cake,
which which corresponds
corresponds to thetoType Type
I filterI
filter
filter cake. In the case of D D1515=//6dd85  6 (Figure 6b), most of of the slurry
slurry particles remain on
on the sand
cake. cake.
In theIncasetheof case
D15of /d85 85  6 (Figure
(Figure 6b), most 6b),
of most
the slurry theparticles particles
remain on remain
the sand the sand
column
column
surface. surface.
column However, However,
surface. However, a
a small fraction small fraction
a smalloffraction
the slurry of the slurry
of particles particles
the slurryinfiltrate
particles infiltrate
deeper deeper
into deeper
infiltrate into
the sandinto the
column sand
the sandand
column
create
column and create
a so-called a so-called
and create“infiltration
a so-called zone”. “infiltration zone”.
This behavior
“infiltration This
zone”. This behavior
is in accordance
behavior is in accordance
is inwith the Type
accordance with the
II filter
with Type II
cake.IIAs
the Type filter
for
filter
cake.
the
cake. As
Asfor
case forDthe
of case
15 /d
the 85 =
case of12D
of 15 // d
D(Figure
15 d 85 
85  12
6c),12 (Figure
nearly
(Figureall 6c),
of nearly
the
6c), slurry
nearly all of the
particles
all of the slurry particles
penetrate
slurry intopenetrate
particles the sand into
penetrate column
into the
the
and no
sand obvious filter cake is observed on observed
the surface of the sand column, which shows the Type III
sand column and no obvious filter cake
column and no obvious filter cake is
is observed on on the
the surface
surface of of the
the sand
sand column,
column, which which shows
shows
filter
the cake. For better illustration of the different types of filtertypes
cake, the slurry particles at the end of
the Type
Type IIIIII filter
filter cake.
cake. For For better
better illustration
illustration of of the
the different
different types of of filter
filter cake,
cake, the
the slurry
slurry particles
particles
thethe
at simulations are extractedare and visualized in Figure 7. in Figure 7.
at the end
end ofof the
the simulations
simulations are extracted extracted and and visualized
visualized in Figure 7.

(a)
(a) (b)
(b) (c)
(c)
Figure
Figure6.
Figure 6.Three
6. Threetypes
Three typesof
types offilter
of filtercake
filter cakeon
cake onthe
thesand
sandcolumn
columnsurface:
surface:(a)
(a)Type
(a) TypeI;I;(b)
Type (b)Type
TypeII;
II; (c)
(c) Type
TypeIII.
Type III.
III.

(a)
(a) (b)
(b) (c)
(c)
Figure
Figure 7. Extracted slurry particles inside the sand column: (a) Type I; (b) Type II;
II;(c)
(c)Type
TypeIII.
Figure 7.
7. Extracted
Extracted slurry
slurry particles
particles inside
inside the
the sand
sand column:
column: (a)
(a) Type
Type I;
I; (b)
(b) Type
Type II; (c) Type III.
III.
4.2.
4.2. Porosity
Porosity Distribution
Distribution Inside
Inside the
the Filtration
Filtration Column
Column
4.2. Porosity Distribution Inside the Filtration Column
The
The change
change of of porosity
porosity distribution
distribution is is another
another indication
indication ofof the
the filter
filter cake
cake formation.
formation. The
The slurry
slurry
The
particles change
might of
clog porosity
the poresdistribution
inside the is another
sand indication
column and of
change the filter
the cake
initial formation.
porosity The slurry
distribution
particles might clog the pores inside the sand column and change the initial porosity distribution
particles
along might column.
clog the pores inside the sand column and change the initial porosity distribution along
along thethe sand
sand column.
the sand
Figure column.
Figure 88 shows
shows that that three
three types
types ofof slurry
slurry filtration
filtration induce
induce different
different changes
changes in in the
the porosity
porosity
distribution from the initial state to the final state. The porosity of the initial state along
distribution from the initial state to the final state. The porosity of the initial state along the filtration the filtration
column
column can can be
be divided
divided into
into two
two parts
parts due
due to to the
the existence
existence ofof the
the sand
sand column,
column, slurry
slurry particles,
particles, and
and
their interface. There is a sharp increase in porosity from around 0.4 to 0.98 at the
their interface. There is a sharp increase in porosity from around 0.4 to 0.98 at the interface. However, interface. However,
after
after slurry
slurry infiltration,
infiltration, aa slight
slight difference
difference could
could bebe observed
observed in in terms
terms of of porosity
porosity distribution.
distribution. For
For
Type
Type I filter cake as Figure 8a shows, the porosity distribution curve exhibits a plateauwith
I filter cake as Figure 8a shows, the porosity distribution curve exhibits a plateau withaaporosity
porosity
Infrastructures 2018, 3, 15 9 of 11

Figure 8 shows that three types of slurry filtration induce different changes in the porosity
distribution from the initial state to the final state. The porosity of the initial state along the filtration
column can be divided into two parts due to the existence of the sand column, slurry particles, and
their interface. There is a sharp increase in porosity from around 0.4 to 0.98 at the interface. However,
after slurry infiltration, a slight difference could be observed in terms of porosity distribution. For
Infrastructures 2018, 3, x FOR PEER REVIEW 9 of 11
Type I filter cake as Figure 8a shows, the porosity distribution curve exhibits a plateau with a porosity
around 0.63
around 0.63 at
at the
the interface, indicating the
interface, indicating the formation
formation of of the
the filter
filter cake. Above the
cake. Above the filter
filter cake,
cake, asas the
the
sedimentation of
sedimentation of slurry
slurry particles,
particles, the
the porosity
porosity tends
tends to to be
be 1.
1. Since
Since nono more
more particles
particles penetrate
penetrate deeper
deeper
into the sand column, the porosity below the filter cake nearly remains the same
into the sand column, the porosity below the filter cake nearly remains the same as the initial state. as the initial state.
However, for a Type II filter cake, due to the formation of the “infiltration zone” (see Figure 7b), aa
However, for a Type II filter cake, due to the formation of the “infiltration zone” (see Figure 7b),
slight decrease
slight decrease of of porosity
porositycouldcouldbebeobserved
observedbelowbelowthe theinterface
interfaceasasFigure
Figure8b8b shows.
shows. TheThe porosity
porosity at
the plateau (0.66) is larger than that of the Type I filter cake curve, which indicates a looser filter filter
at the plateau (0.66) is larger than that of the Type I filter cake curve, which indicates a looser cake
cake forming
forming at theatsand
the sand column
column surface.
surface. As As
for for
thethe
TypeType IIIIII porositydistribution
porosity distributioncurve,
curve,except
except for
for the
the
porosity distribution
porosity distribution above
above thethe sand
sand column,
column, no no obvious
obvious difference
difference occurs
occurs between
between thethe initial
initial state
state
and final state since the slurry particles are too small to clog the pores inside
and final state since the slurry particles are too small to clog the pores inside the sand column. the sand column.

0.5 Initial state 0.5 Initial state

After infiltration After infiltration

0.4 0.4
Pure water Pure water
Distance to bottom (m)

Distance to bottom (m)

0.3 0.3
Filter cake Filter cake

0.2 0.2
Infiltration zone
Sand column
0.1 0.1 Sand column

0.0 0.0

0.3 0.4 0.5 0.6 0.7 0.8 0.9 1.0 1.1 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1.0 1.1
Porosity Porosity

(a) (b)
0.5 Initial state
After infiltration

0.4
Pure water
Distance to bottom (m)

0.3

0.2 Sand column

0.1

0.0

0.3 0.4 0.5 0.6 0.7 0.8 0.9 1.0

Porosity
(c)

Figure 8. Porosity distribution along the infiltration column of initial state and after infiltration: (a)
Figure 8. Porosity distribution along the infiltration column of initial state and after infiltration: (a) Type
Type I; (b) Type II; (c) Type III.
I; (b) Type II; (c) Type III.

5. Conclusions
5. Conclusions
In this paper, a coupled CFD–DEM simulation of slurry filtration and filter cake formation is
In this paper, a coupled CFD–DEM simulation of slurry filtration and filter cake formation is
presented to analyze the infiltration behavior between slurry particles and a sand column in a
presented to analyze the infiltration behavior between slurry particles and a sand column in a filtration
filtration column test. Three sets of sand and slurry particle sizes were selected according to Min et
column test. Three sets of sand and slurry particle sizes were selected according to Min et al.’s
al.’s experiments to model the formations of three types of filter cake. Based on the simulation results,
experiments to model the formations of three types of filter cake. Based on the simulation results, the
the following conclusions could be drawn:
following conclusions could be drawn:
(1) The interaction between sand and slurry particles could be classified into three types by the
sedimentation behavior of slurry particles on the sand particles: Type I, Type II, and Type III.
For Type I, a dense and thick filter cake forms on the sand column surface. For Type II, a
relatively loose filter cake and an “infiltration zone” are observed. For Type III, no obvious filter
cake forms since the slurry particles are too small to clog the pores inside the sand column. These
results are in accordance with Min et al.’s experimental results.
Infrastructures 2018, 3, 15 10 of 11

(1) The interaction between sand and slurry particles could be classified into three types by the
sedimentation behavior of slurry particles on the sand particles: Type I, Type II, and Type III.
For Type I, a dense and thick filter cake forms on the sand column surface. For Type II, a relatively
loose filter cake and an “infiltration zone” are observed. For Type III, no obvious filter cake forms
since the slurry particles are too small to clog the pores inside the sand column. These results are
in accordance with Min et al.’s experimental results.
(2) Three types of distribution of porosity along the column are obtained. Based on the porosity
distribution, the ranges of zones of pure water, filter cake, infiltration, and sand column could
be identified.
(3) The coupled CFD–DEM method is capable of simulating slurry filtration and filter cake formation
during a filtration column test. With the development of computing technology, this method has
the potential to simulate more complex problems such as full-scale modelling of an in situ slurry
shield project.

Author Contributions: Conceptualization, Z.Z. and T.Y.; Methodology, Z.Z and T.Y.; Software, T.Y; Formal
Analysis, T.Y; Investigation, T.Y.; Data Curation, T.Y.; Writing-Original Draft Preparation, T.Y; Writing-Review &
Editing, Z.Z and T.Y.; Visualization, T.Y.; Supervision, Z.Z.; Project Administration, Z.Z.; Funding Acquisition, Z.Z.
Funding: The research was financed by the Natural Science Foundation of China (No. 41372276 and No. 51509186).
Conflicts of Interest: The authors declare no conflict of interest.

References
1. Eisensteib, Z.D. Large undersea tunnels and progress of tunneling technology. Tunn. Undergr. Sp. Tech. 1994,
9, 283–292. [CrossRef]
2. Lee, D.J.; Wang, H.C. Theories of cake filtration and consolidation and implications to sludge dewatering.
Wat. Res. 2000, 34, 1–20. [CrossRef]
3. Tien, C.; Bai, R. An assessment of the conventional cake filtration theory. Chem. Eng. Sci. 2003, 58, 1323–1336.
[CrossRef]
4. Krause, T. Schildvortrieb Mit Flüssigkeits- und Erdgestützter Ortsbrust. Ph.D. Thesis, Technischen
Universität Carolo-Wilhelmina, Braunschweig, Germany, 1987.
5. Talmon, A.M.; Mastbergen, D.R.; Huisman, M. Invasion of pressurized clay suspensions into granular soil.
J. Porous Media 2013, 16, 351–365. [CrossRef]
6. Min, F.L.; Zhu, W.; Han, X.R.; Zhong, X.C. The effect of clay content on filter cake formation in highly
permeable gravel. Geotech. Spec. Publ. 2010, 204, 210–215.
7. Min, F.L.; Zhu, W.; Han, X.R. Filter cake formation for slurry shield tunneling in highly permeable sand.
Tunn. Undergr. Sp. Tech. 2013, 38, 423–430. [CrossRef]
8. Kloss, C.; Goniva, C. LIGGGHTS: A new open source discrete element software simulation. In Proceedings
of the 5th International Conference on Discrete Element Methods, London, UK, 25–26 August 2010.
9. Goniva, C.; Kloss, C.; Hager, A.; Pirker, S. A new open source CFD-DEM Perspective. In Proceedings of the
5th OpenFOAM Workshop, Gothenburg, Sweden, 21–24 June 2010.
10. Tao, H.; Tao, J.L. Quantitative analysis of piping erosion micro-mechanisms with coupled CFD and DEM
method. Acta Geotech. 2017, 12, 573–592. [CrossRef]
11. Cundall, C.T.; Strack, O.D.L. A discrete numerical model for granular assemblies. Geotechnique 1979, 29,
47–65. [CrossRef]
12. Anderson, T.; Jackson, R. A fluid mechanical description of fluidized beds: Equations of motion. Ind. Chem.
Eng. Fundam. 1967, 6, 527–534. [CrossRef]
13. Kaifui, K.; Thornton, C.; Adams, M. Discrete particle-continuum fluid modelling of gas-solid fluidised beds.
Chem. Eng. Sci. 2002, 57, 2395–2410. [CrossRef]
14. Zhu, H.P.; Zhou, Z.Y.; Yang, R.Y.; Yu, A.B. Discrete particle simulation of particulate system: Theoretical
developments. Chem. Eng. Sci. 2007, 62, 3378–3396. [CrossRef]
15. Drew, D.A.; Lahey, R.T. Some supplemental analysis concerning the virtual mass and lift force on a sphere in
a rotating and straining flow. Int. J. Multiph. Flow 1990, 16, 1127–1130. [CrossRef]
Infrastructures 2018, 3, 15 11 of 11

16. DEM Solutions. EDEM-CFD Coupling for FLUENT: User Guide; DEM Solutions: Edinburgh, UK, 2010.
17. Xu, S.L.; Sun, R.; Cai, Y.Q.; Sun, H.L. Study of sedimentation of non-cohesive particles via CFD-DEM
simulations. Granul. Matter 2018, 20, 4. [CrossRef]
18. Di Felice, R. The voidage function for fluid-particle interaction systems. Int. J. Multiph. Flow 1994, 20, 153–159.
[CrossRef]
19. Duan, C.L.; Sheng, C.; Wu, L.L.; Wu, Y.M.; He, J.F. CFD-DEM simulation of fluid-solid flow of a tapered
column separation bed. Int. J. Min. Sci. Technol. 2015, 25, 855–859. [CrossRef]
20. Zhao, T.; Houlsby, G.T.; Utili, S. Investigation of granular batch sedimentation via DEM-CFD coupling.
Granul. Matter 2014, 16, 921–932. [CrossRef]
21. Chen, J.K.; Wang, Y.S.; Li, X.F.; He, R.Y.; Han, S.; Chen, Y.L. Erosion prediction of liquid-particle two-phase
flow in pipeline elbows via CFD-DEM coupling method. Powder Technol. 2015, 275, 182–187. [CrossRef]
22. Zhao, J.D.; Shan, T. Coupled CFD-DEM simulation of fluid-particle interaction in geomechanics.
Powder Technol. 2013, 239, 248–258. [CrossRef]
23. Peng, Z.B.; Doroodchi, E. Influence of Void Fraction Calculation on Fidelity of CFD-DEM Simulation of
Gas-Solid Bubbling Fluidized Beds. Part. Technol. Fluid. 2014, 60, 2000–2018. [CrossRef]

© 2018 by the authors. Licensee MDPI, Basel, Switzerland. This article is an open access
article distributed under the terms and conditions of the Creative Commons Attribution
(CC BY) license (http://creativecommons.org/licenses/by/4.0/).