You are on page 1of 17

European Journal of Environmental and Civil Engineering

ISSN: 1964-8189 (Print) 2116-7214 (Online) Journal homepage: https://www.tandfonline.com/loi/tece20

Three-dimensional numerical model of internal


erosion

Ahmed Chetti, Ahmed Benamar & Khaled Korichi

To cite this article: Ahmed Chetti, Ahmed Benamar & Khaled Korichi (2019): Three-dimensional
numerical model of internal erosion, European Journal of Environmental and Civil Engineering,
DOI: 10.1080/19648189.2019.1585296

To link to this article: https://doi.org/10.1080/19648189.2019.1585296

Published online: 23 Mar 2019.

Submit your article to this journal

Article views: 61

View related articles

View Crossmark data

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


https://www.tandfonline.com/action/journalInformation?journalCode=tece20
EUROPEAN JOURNAL OF ENVIRONMENTAL AND CIVIL ENGINEERING
https://doi.org/10.1080/19648189.2019.1585296

Three-dimensional numerical model of internal erosion


Ahmed Chettia,b, Ahmed Benamarc and Khaled Korichid
a
University Center of El-Bayadh, El-Bayadh, Algeria;bCivil Engineering and Environmental Laboratory "LGCE",
University of Djillali Liabes, Sidi Bel Abbes, Algeria; cNormandie University, UNIHAVRE, CNRS, LOMC, Le
Havre, France;dDjillali Liabes University, Sidi Bel Abbes, Algeria

ABSTRACT ARTICLE HISTORY


The purpose of this study concerns the establishment of three-dimensional Received 3 May 2018
numerical simulation of the coupling between fluid flow and internal ero- Accepted 15 February 2019
sion within a porous medium. In order to properly investigate suffusion
KEYWORDS
process in porous media, we present a finite volumes method using expli-
Finite volume; fine particles;
cit scheme for solving advection–dispersion–deposition equation on three- suffusion; simulation; three-
dimensional grids. The accuracy and the robustness of the proposed finite dimensional
volume method are assessed by means of various numerical test cases
and validated by the adjustment of experimental results.

1. Introduction
The 1D and 2D numerical models presented in the literature and devoted to simulate soil internal
erosion cannot adequately model the phenomenon of internal erosion because of the three-dimen-
sional mechanism involved. Soil suffusion is defined as hydromechanical process through which finer
particles migrate inside soil skeleton, under seepage flow. To characterise suffusion phenomenon,
Sail, Marot, Sibille, and Alexis (2011) have developed a large oedo-permeameter device in which the
soil sample was submitted to an axial load with a downward flow under a constant hydraulic gradi-
ent. The authors realised that suffusion induces a settlement and a localised increase of interstitial
pressure. Benamar and Seghir (2017) presented an analytical model based on the integration of the
transport equation of particles in a homogeneous porous medium, where they showed that detach-
ability parameter deceases when fines content increases. Zhen-Ming, Hong-Chao, Song-Bo, Ming, and
Tao (2018) presented a coupled computational fluid dynamics and discrete element method to inves-
tigate the seepage characteristics of fine and coarse materials relevant to landslide dam. Through a
microcosm approach, the authors showed that the failure mode of fine-grained soils and coarse-
grained soils was flowing soil and piping, respectively. Bonelli and Marot (2012) considered that suf-
fusion is an interfacial erosion at microscopic scale. Golay and Bonelli (2012) presented numerical
model for simulating surface erosion occurring at a fluid/soil interface subject to a flow process at
pore scale. However, most of works based on 3D approach have focused on the study of the initi-
ation and evolution of piping phenomena. Following the collapse of the Martin Plant Dam (Florida
Power and Light Company), Townsend, Schmertmann, Logan, Pietrus, and Wong (1981) from the
University of Florida conducted a study devoted to the conditions and causes of this rupture. The
authors have developed a numerical finite element 3D infiltration model to evaluate the gradient
concentration coefficients parallel and perpendicular to an internal erosion pipe under various
boundary conditions. They have concluded that the horizontal infiltration velocities of water are not

CONTACT Ahmed Chetti chetti_ahmed@yahoo.fr University Center of El-Bayadh, El-Bayadh, Algeria.


ß 2019 Informa UK Limited, trading as Taylor & Francis Group
2 A. CHETTI ET AL.

enough to cause internal erosion by piping. The authors concluded that there is a significant vertical
gradient that acts on soil particles to become more mobile. According to the authors, particle trans-
port depends mainly on vertical infiltration gradients. Based on the schematic of the infiltration flow
net, Schmertmann (2000) has quantified the vertical gradients acting at the end of a pipe and linked
these vertical gradients to the average gradient by a concentration coefficient, iv ¼ ciav, where
iv ¼ vertical gradient at the end of the pipe, c ¼ concentration coefficient, and iav ¼ average hydraulic
gradient (total hydraulic head/total linear flow length). The development of these correction coeffi-
cients to account for the effects of three-dimensional flow, required a finer flow network design.
Unfortunately, the capacity of computers at that time prevented to further refine the seepage flow
net. The 3D numerical model developed by Vandenboer, Van Beek, and Bezuijen (2014) shows the
three-dimensional nature of the piping phenomenon. The authors evoke the reliability of 3D numer-
ical model calculations against 2D approaches, because 3D numerical results enable a more suitable
understanding of the complex mechanism of piping erosion. Ferguson (2012) concluded through the
results of 2D infiltration modelling that the 2D model was a simplification and recognised the three-
dimensional nature of piping evolution or internal erosion. Schmertmann (2000) describes in his 3D
model, the sinuous behaviour of piping phenomenon by noting that the pipe does not progress
from downstream to upstream straight line. On the contrary, it advances in a meandering network
of channels and often the progression of the conduit stops in a path obstructed to restart in another
path. This is compatible with the conclusion of the Delft research which notes that the meandering
of the pipes is due to the search for weak points of the granular structure of the sand. Wong (1981)
cited by Townsend et al. (1981) presented a model on 3D gradients at pipehead, based on approxi-
mations of the 2D network infiltration. The results of the model conclude that the 3D influence of
the pipe sustains the fact that gradients are enough to erode soil particles during the piping process
since the moment of initiation to the formation of the breach. The gradient vector field at the pipe-
head is multidirectional, ranging from the horizontal to the top of the pipe adjacent to the upper
confining layer to the vertical along the axis of the pipe.
From this literature review, we note the complexity of internal erosion phenomenon, but the
reported models display mainly the mechanism of piping. Statistics showed that about 50% of
internal erosion incidents are due to suffusion phenomena (Engemoen & Redlinger, 2009; Engemoen,
2011; Fry, Degoutte, & Goubet, 1997). However, the internal erosion by suffusion in soils differs from
the piping mechanism by its geometric and hydraulic boundary conditions. Thus, the breakage risk is
consequently different. In this study, the aim of our approach is to establish a 3D suffusion model
which take into account the three-dimensional mechanism of this phenomenon. However, the chal-
lenge in this study is to model suffusion process through solving 3D advection–dispersion–deposition
equation, by taking into account the instantaneous variation of porosity.

2. Mathematical model of suffusion


Based on the 1D model presented by Chetti, Benamar, and Hazzab (2016), which assumed that
the flow is in one-dimensional, equations governing soil erosion are described by:
8
>  
> @/ ¼ k  Cme  /0 /ðtÞ qðx; tÞ
>
>
>
> @t
>
>   
< @ ðCðx; tÞ1Þ/ðtÞ @ ðCðx; tÞÞ @ 2 /ðtÞ  Cðx; tÞ  
þ qðx; tÞ ¼ DL Kd /ðtÞ  Cðx; tÞ (1)
>
> @t @x @x 2
>
>
>
> k0 ð1/0 Þ2 /3
>
> qðx; tÞ ¼  1 þ 2; 5  C x; t gradH
: ð ð ÞÞ /0 ð1/ðtÞÞ2
3

We recall that this model was established on macroscopic approach, where the REV is
assumed to be composed by three phases (solid, water and suspension). The velocity field is
EUROPEAN JOURNAL OF ENVIRONMENTAL AND CIVIL ENGINEERING 3

assumed to be Darcian and calculated by using Carman type permeability and a linear kinetics
deposition was used.
As this is not always the case, it is appropriate, therefore, to generalise this model to 3D form as:
8
> @/  
>
> ¼ k  Cme  /0 /ðtÞ jjqðt; x; y; zÞjj
>
> @t
>
> @ ðCðt; x; y; zÞ1Þ/ðtÞ
>
> !
>
>
< þ! q ðt; x; y; zÞ  grad Cðt; x; y; zÞ ¼
 @t
> !     (2)
>
> div D  grad / ð t Þ  C ð t; x; y; z Þ K / ðt Þ  C ðt; x; y; z Þ
>
>
d
>
>
>
> ! k0 ð1/0 Þ2 /ðtÞ
3
!
>
>
: q ðt; x; y; zÞ ¼  ð1 þ 2; 5  Cðt; x; y; zÞÞ grad Hðx; y; zÞ
/0 ð1/ðtÞÞ2
3

where C is the volume concentration of eroded particles (volume fraction), D is the dispersion
coefficient (m2 =s), k0 is the initial hydraulic permeability (m/s) and Kd is the initial deposition kin-
etic coefficient (s1 ).
Scheidegger (1961) defines the dispersivity tensor for an isotropic porous media in terms of
two constants. He showed that kinematic dispersion is proportional to the Darcy velocity:

DL ¼ aL juj and DT ¼ aT juj


8
>
> u2x u2y u2
>
> D ¼ a þ a þ aT z
>
> xx L
juj
T
juj juj
>
>
>
>
>
> u 2 u 2
u 2
>
> D ¼ a x
þ a
y
þ a z
>
> yy T
juj
L
juj
T
juj
>
>
>
>
>
>
2 u2 2
< Dzz ¼ aT ux þ aT y þ aL uz
juj juj juj (3)
>
> ða a Þu u
>D ¼ D ¼
> L T x y
>
> xy yx
>
> juj
>
>
> a a
> Dxz ¼ Dzx ¼ ð L T Þ x z
> uu
>
>
>
> juj
>
>
>
> ðaL aT Þuy uz
>
: Dyz ¼ Dzy ¼
juj
where aL, longitudinal dispersivity; aT, transverse dispersivity.
The hydraulic conductivity in an anisotropic porous media can be written as:
8
>
> @H @H @H
>
> qx ¼ kxx þ kxy þ kxz
>
> @x @y @z
>
< @H @H @H
qy ¼ kyx þ kyy þ kyz (4)
>
> @x @y @z
>
>
>
> @H @H @H
>
: qz ¼ kzx þ kzy þ kzz
@x @y @z
In matrix form, the above equations system can be written as:
2 3
@H
2 3 2 36 7
qx kxx kxy kxz 6 @x 7
6 7 6 7 6 @H 7
4 qy 5 ¼ 4 kyx kyy kyz 56 7
6 @y 7 (5)
qz kzx kzy kzz 6 7
4 @H 5
@z
4 A. CHETTI ET AL.

Figure 1. 3D discretization.

where each component kij relates the flow rate in one direction (i) to pressure differences in
orthogonal directions (j).
By matrix diagonalisation procedure, it is possible to find a coordinate system (x0 , y0 , z0 ) which
the three coordinate axes coincide with the principal axes of the dispersion and permeability
tensor (the direction of maximum, minimum, and intermediate hydraulic permeability).
The tensor of dispersion D takes the diagonal form:
2 3
aL 0 0
6 7
D ¼ 4 0 aT 0 5juj (6)
0 0 aT

and the tensor of k takes the diagonal form:


2 3
kx 0 0
6 7
k¼40 ky 05 (7)
0 0 kz

which kx, ky and kz are the eigenvalues of k. In our case, it is an isotropic porous medium,
so: kx ¼ ky ¼ kz :
In the following, we study the case of constant hydraulic gradient which is the most represen-
tative case of the real conditions of an earth dam.

2.1. Numerical resolution


A 3D spatial discretization (Figure 1) using finite volume method is proposed in the present art-
icle, where the computational domain is discretised into NPK meshes of centre (xi, yj, zk) (i, j
and k varying, respectively, from 1 to N, from 1 to P and from 1 to K). It will be assumed that
the space steps in each direction Dx ¼ xiþ1=2 xi1=2 ; Dy ¼ yjþ1=2 yj1=2 and Dz ¼ zkþ1=2 zk1=2
are constants.
We recall that the explicit conservative scheme (Godunov finite volume scheme) is written
Dt
as: Unþ1
i ¼ Uni  Dx ðFiþ1=2 Fi1=2 Þ:
Where U is the vector of conservative variable and Fiþ1=2 ; Fi1=2 are the fluxes at the interfaces
i þ 1=2; i1=2; respectively.
The second relation from system of Eq. (2) can be written in conservative form; Ut þ FðUÞx ¼
SðUÞ (note that subscripts are used to denote partial derivatives with respect to t and x) as:
EUROPEAN JOURNAL OF ENVIRONMENTAL AND CIVIL ENGINEERING 5

     
  @ ð/C Þ @ ð/C Þ @ ð/CÞ
@ ðC1Þ/ @ qC  D L @x @ qC  D T @y @ qC  D T @z
þ ¼  Kd ð/CÞ (8)
@t @x @y @z
where U ¼ ðC1Þ/
FðUÞ ¼ qCDL @ð/CÞ
@x : the flux in x direction
@ð/CÞ @ð/CÞ
@ðqCDT Þ @ðqCD Þ
SðUÞ ¼  @y
@y
 @z
T @z
Kd ð/CÞ: the source term.
To take into account the source term, we use the fractional-step method (Leveque, 2004), by
resolving the above equation without second member Ut þ FðUÞx ¼ 0; that is give the solution
U ; then from the obtained solution, we resolve the ODE including the source term:
)
Ut þ F ðUÞx ¼ 0
) U (9)
Uðx; tn Þ ¼ Un
The ODE is then:
9
dU
¼ SðUÞ =
dt ) Unþ1 (10)
;
Uðx; t Þ ¼ U
n

The initial conditions for the homogeneous problem are the initial conditions of the problem
with source term and the solution after a time step Dt is U : The system of Eq. (11) can then be
solved for a time Dt and with the initial condition given by the resolution of the system of Eq.
(10), i.e. U : The solution so obtained can be seen as the approximate solution of the problem
with source term. The advection–dispersion operator AD(U) and the source term operator S(U)
are defined as follows:
8 h i
< ADx : U ¼ Un  Dt F n
>
iþ1=2;j;k Fi1=2;j;k
n
i;j;k i;j;k
Dx (11)
: S : Unþ1 ¼ U þ Dt:SU 
>
x i;j;k i;j;k i;j;k

By following this method, we obtain:


 
  @ ð/CÞ
@ ðC1Þ/ @ qC  D L @x
þ ¼0 (12)
@t @x
After integration in space of Eq. 13 on the mesh Ki for 1  i  N; then in time on the interval
n (coming from a uniform discretization in time, with Dt the step of subdivision), we obtain:
   n  n
nþ1
Ci;j;k 1 /nþ1
i;j;k  Ci;j;k 1 /i;j;k
n
Fiþ1=2;j;k Fi1=2;j;k
n
þ ¼0 (13)
Dt Dx
The numerical fluxes at cell interfaces, Fiþ1=2;j;k n
¼ ðqCDL @ð/CÞ n
@x Þiþ1=2;j;k ; are obtained using an
upwind scheme.
/n Cn /ni;j;k Ci;j;k
n
n
So, we getFiþ1=2;j;k ¼ qni;j;k Ci;j;k
n
DL iþ1;j;k iþ1;j;k
Dx
In this case, we obtain:
/niþ1;j;k Ciþ1;j;k
n
/ni;j;k Ci;j;k
n
/ni;j;k Ci;j;k
n
/ni1;j;k Ci1;j;k
n
qni;j;k Ci;j;k
n
DL Dx qni1;j;k Ci1;j;k
n
þ DL Dx
Dx
   n  n (14)
nþ1
Ci;j;k 1 /nþ1
i;j;k  Ci;j;k 1 /i;j;k
þ ¼0
Dt
The porosity is assumed to be uniformly distributed and remains a constant function over the
space. But in our case, it changes with time due to internal erosion. In fact, we assume that
6 A. CHETTI ET AL.

porosity takes the same value over the space at a given time, and the porosity amount decreases
uniformly with time owing to erosion process. So, @/
@x ¼ 0; and the above equation becomes:

   n  n
nþ1
Ci;j;k 1 /nþ1 i;j;k  Ci;j;k 1 /i;j;k
n n
(15)
qi;j;k Ci;j;k qni1;j;k Ci1;j;k
n n
Ciþ1;j;k 2Ci;j;k
n
þ Ci1;j;k
n
þDt  DL /i;j;k
n
¼0
Dx Dx2

Thus, we obtain:

   n  n
nþ1
Ci;j;k 1 /nþ1 i;j;k ¼ Ci;j;k 1 /i;j;k
n n
(16)
qi;j;k Ci;j;k qni1;j;k Ci1;j;k
n n
Ciþ1;j;k 2Ci;j;k
n
þ Ci1;j;k
n
þDt  þ DL /ni;j;k
Dx Dx2

The term ðCi;j;k


nþ1
1Þ/nþ1
i;j;k of the above equation represents the solution of homogeneous equa-
tion (without source term)

n n

   qi;j;k Ci;j;k qni1;j;k Ci1;j;k


n n
Ciþ1;j;k 2Ci;j;k
n
þ Ci1;j;k
n
U ¼ n
Ci;j;k 1 /ni;j;k þ Dt  þ DL /i;j;k
n
(17)
Dx Dx 2

So, the solution of the global problem (with source term) is:

  qni;j;k Ci;j;k
n
qni1;j;k Ci1;j;k
n n
Ciþ1;j;k 2Ci;j;k
n
þ Ci1;j;k
n
 
nþ1
U ¼ n
Ci;j;k 1 /ni;j;k þ Dt  þ DL /ni;j;k þ S Ui;j;k
Dx Dx2
(18)

 2    3
@ ð/C Þ @ ð/CÞ
4 @ qC  D T @y @ qC  D T @z 5
Unþ1 ¼ U þ Dt   Kd ð/CÞ (19)
@y @z

  n n
 n  n qi;j;k Ci;j;k qni1;j;k Ci1;j;k
n
nþ1
Ci;j;k 1 /nþ1 ¼ C 1 / þ Dt 
i;j;k i;j;k i;j;k
Dx

n
Ciþ1;j;k 2Ci;j;k
n
þ Ci1;j;k
n
þ DL /ni;j;k
Dx2
(20)
qni;j;k Ci;j;k
n
qni;j1;k Ci;j1;k
n n
Ci;jþ1;k 2Ci;j;k
n
þ Ci;j1;k
n
þ Dt  þ DT /ni;j;k
Dy Dy 2

qni;j;k Ci;j;k
n
qni;j;k1 Ci;j;k1
n n
Ci;j;kþ1 2Ci;j;k
n
þ Ci;j;k1
n
 þ DT /ni;j;k Kd ð/CÞ
Dz Dz2

Knowing that: divðq  CÞ ¼ C  divðqÞ þ q  gradðCÞ and the fluid incompressibility as defined
requires that div(q) ¼ 0.
EUROPEAN JOURNAL OF ENVIRONMENTAL AND CIVIL ENGINEERING 7

Figure 2. Boundary conditions.

So, Eq. (20) becomes:


 
 n  n n
Ci;j;k Ci1;j;k
n
Ci;j;k 1 /i;j;k ¼ Ci;j;k 1 /i;j;k þ Dt qni;j;k
nþ1 nþ1
Dx
n
Ciþ1;j;k 2Ci;j;k
n
þ Ci1;j;k
n
þ DL /ni;j;k
Dx2
(21)
n
Ci;j;k Ci;j1;k
n n
Ci;jþ1;k 2Ci;j;k
n
þ Ci;j1;k
n
qni;j;k þ DT /ni;j;k
Dy Dy2

n
Ci;j;k Ci;j;k1
n n
Ci;j;kþ1 2Ci;j;k
n
þ Ci;j;k1
n
qni;j;k þ DT /ni;j;k Kd ð/CÞ
Dz Dz2
Thus, we obtain the final solution:
n n  !
/ C i;j;k 1 Dt C n
i;j;k C n
i1;j;k C n
i;j;k C n
i;j1;k C n
i;j;k C n
i;j;k1
nþ1
Ci;j;k ¼1þ þ nþ1 qni;j;k þ þ
/nþ1 / Dx Dy Dz
n
Ciþ1;j;k 2Ci;j;k
n
þ Ci1;j;k
n
þ Dni;j;k  /n
Dx2 (22)
n
Ci;jþ1;k 2Ci;j;k
n
þ Ci;j1;k
n
þ DT /ni;j;k
Dy2

n
Ci;j;kþ1 2Ci;j;k
n
þ Ci;j;k1
n
þ DT /ni;j;k  Kd  /  n n
Ci;j;k
Dz2
From the mass conservation and behaviour laws, we obtained a system of partial differential
equations with three unknowns (porosity /; volume concentration of particles in the fluid C and
Darcy velocity q).
8  
>
> /nþ1 ¼ /n þ k  Dt  qnijk Cem  /0 /n
>
>  n 
>
> Cðni1Þjk
>
> /n Cijk 1 Dt
n
Cijk
>
> C nþ1
¼ 1 þ þ q n
>
> ijk
/nþ1 /nþ1
ijk
Dx
>
>
>
> n
2C n
þ n
Cinðj1Þk
>
> C
n ðiþ1Þjk
Cði1Þjk
n
Cijk
>
> þD n
 /
ijk
 q n
>
> ijk
Dx2 ijk
Dy
<
C n
i ðjþ1Þk 2C n
ijk þ C n
i ðj1 Þk
C n
ijk C n
ijðk1Þ (23)
>
> þD n
 / n
q n
>
> ijk
Dy 2 ijk
Dz
>
>

>
> Cn 2Cijk
n
þ Cijnðk1Þ
>
> n ijðkþ1Þ
>
> þD n
 /  K  / n
 C n
>
>
ijk
Dz2
d ijk
>
>  
>
> 2
>
> k0 1/0 /n 3
>
> q n
¼    gradH
: ijk
1 þ 2:5:Cn /0 3 ð1/n Þ2
ijk
8 A. CHETTI ET AL.

Figure 3. Relative influence of (Cme1) on the model (at the outlet).


EUROPEAN JOURNAL OF ENVIRONMENTAL AND CIVIL ENGINEERING 9

Figure 4. Relative influence of k on the model (at the outlet).


10 A. CHETTI ET AL.

Figure 5. Relative influence of Kd on the model (at the outlet).


EUROPEAN JOURNAL OF ENVIRONMENTAL AND CIVIL ENGINEERING 11

Table 1. Sample characteristics and test conditions.


Sample dimensions Numerical parameters
Fine mass
4
content (%) Length (mm) Section (mm )2
k0 (10 m=s) /0 gradH (m/m) k (m1 ) Cme 1 Kd (s1 )
08 35 803.75 11.54 0.32 11.7 03.7 0.0545 0.0100
06 43 803.75 07.70 0.32 10.7 06.7 0.0450 0.0100
02 22 803.75 04.10 0.36 06.2 18.0 0.0250 0.0100

where the term Gnijk is the discrete value of the function Gðt; x; y; zÞ at the node (xi ; yj ; zk ) at the
time nDt:
Two initial conditions are used, /0 ¼ /initial ; Cijk 0
¼ 0; while the boundary conditions are the
following: in upstream no fluidised solid is contained in the injected fluid
n
(C0jk ¼ 0; Ci0k
n
¼ 0; Cij0
n
¼ 0), and in downstream (outlet) the solid concentration in the fluid is
@Cðt;x;L ;zÞ
equal to the last layer concentration (@Cðt;L@xx ;y;zÞ ¼ 0; @y y ¼ 0; @Cðt;x;y;L
@z

¼ 0), where Lx, Ly and Lz
are the sample length component in x, y and z directions.
So, two types of boundary conditions are imposed: Diritchlet inflow at the upstream and
transmissive outflow at the downstream (cf. Figure 2).
The choice of space and time increments Dt; Dx; Dy and Dz must verify the CFL condition:
minðDx; Dy; DzÞ
Dt    (24)
max qnijk =/nijk

For /0ijk  /nijk  /max and 0  Cijk


n
 1 and from qnijk expression, we deduce: qnijk  qmax where
qmax is given by the following relation:
2  2 3
0
6 1/ijk 7 ð/ Þ3
qmax ¼ k0 6
4 max  3 7
max
5 ð1/ Þ2 gradH (25)
/ijk
0 max

ð1/0 Þ2 ð1minj /0ijk Þ2


In addition, we have: /nijk  /0ijk  /min ; and max ð/0 ijkÞ3 ¼ ðmin/0ijk Þ3
ijk
We know that:
 
min /0ijk ¼ /0 ¼ /initial (26)

So,
ð1/initial Þ2 ð/max Þ3
qmax ¼ k0 gradH (27)
ð/initial Þ3 ð1/max Þ2
/n /n
Using the relation: qnijk  /qinitial
max
; one can deduce that: Dx  minð qnijk Þ  Dx /qinitial
max
: In this case, taking
/initial
ijk ijk
Dt ¼ minðDx; Dy; DzÞ qmax ; the CFL expressed by Eq. (25) is always verified.

3. Results and comments


3.1. Sensitivity analysis
The model involves many parameters which can affect differentially the computed results. So,
the influence of the variation of each parameter is addressed in this section. The aim of this ana-
lysis is to assess the sensitivity of the mathematical model toward each parameter, mainly the
erosion coefficient, the maximum erodibility coefficient and the deposition kinetics coefficient,
which also depend on the hydraulic gradient, the initial fine fraction and the sample dimensions.
The sensitivity analysis ( Figures 3–5) shows the concentration–time evolution of different values
of the adjusted parameters and their influence. The parameter variations are adapted to its relative
influence on the provided erosion (material released at outlet) and their initial values are determined
12 A. CHETTI ET AL.

Table 2. Relative influence of model parameters.


Sample dimensions Parameter
Fine mass Max.
content (%) Length (mm) Section (mm )2
k Dk (%) concentration (g/l) DCð%Þ
5 123 1256.64 8.5 100 0.182 75.82
17 0.320
Kd DKd (%)
0.007 100 0.193 43.11
0.014 0.110
Cme 1 DCme (%)
0.001 0.1 0.191 99.03
0.002 0.380
2 22 803.75 k Dk (%)
18 100 5.711 86.06
36 10.626
Kd DKd (%)
0.01 900 5.711 11.09
0.10 5.077
Cme 1 DCme (%)
0.025 1.37 5.711 122.57
0.059 12.711
6 43 803.75 k Dk (%)
6.7 100 6.893 87.35
13.4 12.914
Kd DKd (%)
0.01 900 6.893 6.74
0.10 6.429
Cme 1 DCme (%)
0.045 5.99 6.938 68.79
0.076 11.711
8 35 803.75 k Dk (%)
3.7 100 3.871 91.62
7.4 7.418
Kd DKd (%)
0.01 900 3.871 3.53
0.10 3.735
Cme 1 DCme (%)
0.051 2.96 3.636 61.72
0.082 5.880

Figure 6. Space-time evolution of concentration and rate particles deposition.


EUROPEAN JOURNAL OF ENVIRONMENTAL AND CIVIL ENGINEERING 13

Table 3. Comparison of 1D versus 3D model parameters.


Nature of
the parameters Parameter 1D 3D
Invariable parameters in Intrinsic parameters Initial permeability K0 ¼ 0.33104 m/s
both models of sample Initial porosity /0 ¼ 0:35
Test conditions Hydraulic gradient gradH ¼ 2.4
Numerical parameters Coefficient of k ¼ 8:5m1
erosion kinetics
Maximum erodibility Cme ¼ 1.001
coefficient
Initial deposition kinetic Kd ¼ 0:0066 s1
coefficient
Parameters that change Dispersion Longitudinal dispersivity aL ¼ 0:001 aL ¼ 0:001
between the Transverse dispersivity / a ¼ 0:001
2T 3
two models Dispersion coefficient D ¼ aT.u aL 0 0
D ¼ 4 0 aT 0 5juj
0 0 aT
Velocity Darcy velocity qx qx, qy, qz
Erosion Concentration Cx Cx, Cy, Cz

Figure 7. Comparison between 3D and 1D numerical model (Silt 5%, gradH ¼ 2.4, L ¼ 12.3 cm).

from experimental matching (Table 1). The range variation of parameters values aimed the numerical
investigation of the model behaviour, even for far values. Table 2 shows the relative influence of
each parameter on the maximal outlet concentration. The results indicate that Cme is the most influ-
ent parameter in which the low increase of 0.1% leads to a large increase by 99% of maximal con-
centration at outlet. An increase of k by 100% leads to an increase by 76% of maximal
concentration. A very large increase of Kd by 900% leads to a reduction by 11% of maximal concen-
tration. So, Cme was the first parameter calibrated from experimental results providing the relative
eroded mass. The two other parameters (k and Kd) act slightly on the numerical result with the fol-
lowing importance: k then Kd. Comparison of the experimental data and the numerical curves pro-
vides the parameters obtained from the fitted curves. As regards to results a very slight variation of
Cme induces a large evolution of outlet concentration, while a large variation of two other parameters
(k and Kd) is required to reach such variation of particle concentration.
The space–time–variation of the concentration and the rate deposition are shown in Figure 6.
The results denote that the particle deposition increases along the porous medium as the con-
centration, while soil porosity is time dependent mainly.
The particle deposition rate is expressed as the product of three terms (Kd  /  C) and so evolves dur-
ing erosion process through the instantaneous evolution of the concentration C, which in turn depends
on the instantaneous variation of the porosity / (see the second relation of Eq. 2: C ¼ Cð/ðtÞÞ:
14 A. CHETTI ET AL.

Figure 8. Effect of initial fine fraction on suffusion process (at the outlet).

3.2. Adjustment of experimental results


In order to validate our 3D model, 1D experimental data obtained from previous tests (Seghir,
Benamar, & Huaqing, 2014) were matched with the numerical results. The adjustment was per-
formed through the time–mass concentration and the deduced cumulative eroded mass evolu-
tion at the outlet of the sample. Table 3 provides comparison of the parameters between 1D
and 3D models.
Figure 7 shows the comparison between 1D numerical results (Chetti et al., 2016) and those
of 3D numerical based on experimental data obtained for 5% of initial fines contents under
hydraulic gradient of 2.4. As is clear, time–concentration evolution (Figure 7) shows that a quite
better agreement is obtained by using 3D numerical model but discrepancy rises when using 1D
numerical model. It is clear that the bump observed on the experimental curve is a local eccen-
tricity in relation to the numerical simulation curve. It is a measurement artefact, i.e. a singularity
due to experimental measurements and has no relation to the suffusion mechanism.
Figure 8 shows the adjustment between experimental data and numerical results obtained for
three different initial fines contents under different hydraulic gradients. Time–concentration evolution
(Figure 8) shows a rapid increase of the concentration of the effluent reaching a peak whose magni-
tude decreases with increasing initial fines content even if hydraulic gradient also increases. These
results show that the effect of fines content is stronger than that of hydraulic gradient in suffusion
process. Compared to 1D numerical results, 3D numerical model provides a good agreement with
experimental curves even within the residual value of concentration over the test time. This numer-
ical model therefore provides representative evolution of the suffusion process which agree with
experimental data. That explains the inherent 3D nature of suffusion process.
The comparison of the results obtained with diverse fines contents (2%, 6% and 8%) of sam-
ples subject to different hydraulic gradients, shows the increase of eroded mass fraction with
decreasing initial fines content, in spite of the decrease of hydraulic gradient. Independently
from hydraulic conditions, lower is the initial fines content in the soil, more important is the
eroded fraction of fine particles, as reported by Seghir et al. (2014). As regards to engineering
dam, Wan and Fell (2004) reported that the embankments designed with at least 20% of fines
are safe against suffusion. The geotechnical explanation for the effect of fines content can be
supported by the soil cohesion concept (shear strength) involved by higher fraction of fines
which lead to make particles attached together and to the matrix. The sensitivity of such param-
eter (initial fines content) shows that its slight variation causes a large variation in the ultimate
eroded mass and so more vulnerability to suffusion. It is then concluded that the larger the ini-
tial fine content, the smaller the erosion susceptibility.
EUROPEAN JOURNAL OF ENVIRONMENTAL AND CIVIL ENGINEERING 15

4. Summary and conclusion


In this article, we have developed a 3D numerical model of the suffusion process, based on a
macroscopic approach. It includes solving an advection–dispersion–deposition equation, using
upwind finite volume scheme. It involves fine particles detachment and 3D advection–disper-
sion–deposition processes leading to particles transport within the porous medium. In this paper,
we have introduced dispersion and deposition processes within an erosion 3D model for better tak-
ing into account the inherent mechanism of suffusion. This model takes into account the instantan-
eous variation of porosity and 3D seepage rate during erosion process. Through sensitivity analysis
one notices that the 3D processes of dispersion and deposition exert a slight effect on soil suffusion
involved within a short medium. The influence of deposition kinetics may be significant for long sam-
ples but also for short samples subjected to severe hydraulic conditions (high gradients). Resistant
effect of initial fines fraction against the suffusion phenomenon is dominant compared to that of the
hydraulic gradient which favours the suffusion process. Numerical results indicate that the suffusion
process is strongly linked to the geotechnical and hydraulic parameters. Compared to 1D numerical
model, the 3D numerical model fits better with the experimental results because of the three-dimen-
sional nature of the suffusion process. Unlike 1D, 3D model allows to study anisotropic soils (i.e. soil
multilayers). Even if studied by simplified hypotheses, such a model permits to take into account the
3D effects of suffusion process (preferential leakage 3D paths due to cracks or holes) when evaluat-
ing the vulnerability of hydraulic structures and road embankments against the susceptibility to
internal erosion.

List of symbols
C Volume concentration of eroded particles
Ck constant in Kozeny–Carman equation
Cme maximum erodibility coefficient
DL longitudinal dispersion coefficient (m2 =s)
DT transverse dispersion coefficient (m2 =s)
gradH Hydraulic gradient (m/m)
H hydraulic head (m)
k Hydraulic permeability (m/s)
kd initial deposition kinetic coefficient (s1 )
m0 Initial fines mass (mg)
m eroded cumulative mass (mg)
m=m0 eroded mass fraction (relative cumulative mass)
q Darcy velocity (m  s1 )
u pore velocity (m/s)
ufs velocity of fluidized solid (eroded fines particles) (m/s)
U conservated variable
F(U) flux function
S(U) source term

Greek letters
a dispersivity coefficient (m)
cw Specific weight of water (N  m3 )
l Instantaneous dynamic viscosity of water (Kg  m1  s1 )
l0 initial dynamic viscosity of water (kg  m1  s1 )
qf Fluid density (kg=m3 )
qs Solid density (kg=m3 )
k coefficient of erosion kinetics (m1 )
/ Instantaneous porosity of soil
/0 initial porosity of soil
Dt time increment
16 A. CHETTI ET AL.

Dx; Dy; Dz space increments

References
Benamar, A., & Seghir, A. (2017). Physical and analytical modeling of internal erosion of fine particles in cohesion-
less soils. Journal of Porous Media, 20(3), 205–216. doi:10.1615/JPorMedia.v20.i3.20
Bonelli, S., & Marot, D. (2012). Micromechanical modeling of internal erosion. European Journal of Environmental
and Civil Engineering, 15(8), 1207–1224. doi:10.3166/ejece.15.1207-1224
Chetti, A., Benamar, A., & Hazzab, A. (2016). Modeling of particles migration in porous media: Application to soil
suffusion. Transport in Porous Media, 113(3), 591–606. doi:10.1007/s11242-016-0714-y
Engemoen, W. O., & Redlinger, C. G. (2009). Internal erosion incidents at bureau of reclamation dams. Paper prepared
for 2009 USSD Meeting in Nashville, TN.
Engemoen, W. O. (2011). Bureau of Reclamation experiences with internal erosion incidents. Paper prepared for 2011
European Working Group on Internal Erosion meeting in Brno, Czech Republic, April 2011.
Ferguson, K. A. (2012). Investigation and Evaluation of Seepage Conditions and Potential Failure Modes Around Outlet Conduits.
Proceedings of the 2012 Annual Conference, United States Society on Dams (USSD), New Orleans, LA, April, 2012.
Fry, J., Degoutte, G., & et Goubet, A. (1997). L’ erosion interne: typologie, detection et reparation. Barrages &
Reservoirs, N6 p. 126.
Golay, F., & Bonelli, S. (2012). Numerical modeling of suffusion as an interfacial erosion process. European Journal of
Environmental and Civil Engineering, 15(8), 1225–1241. doi:10.1080/19648189.2011.9714850
Leveque, R. J. (2004). Finite volume method for hyperbolic problems. Cambridge, MA: Cambridge University Press.
Sail, Y., Marot, D., Sibille, L., & Alexis, A. (2011). Suffusion tests on cohesionless granular matter. European Journal of
Environmental and Civil Engineering, 15(5), 799–817. doi:10.1080/19648189.2011.9693366
Scheidegger, A. E. (1961). General theory of dispersion in porous media. Journal of Geophysical Research, 66(10),
3273–3278. doi:10.1029/JZ066i010p03273
Schmertmann, J. H. (2000). The non-filter factor of safety against piping through sands. In F. Silva and E
Kavazanjian (Eds.). ASCE Geotechnical Special Publication No. 111, Judgment and Innovation (pp. 65–132), ASCE,
Reston, VA.
Seghir, A., Benamar, A., & Huaqing, W. (2014). Effects of fine particles on the suffusion of cohesionless soils.
Transport in Porous Media, 103(2), 233–247. doi:10.1007/s11242-014-0299-2
Townsend, F. C., Schmertmann, J. H., Logan, T. J., Pietrus, T. J., & Wong, Y. W. (1981). An Analytical and
Experimental Investigation of a Quantitative Theory for Piping in Sand, Final Report. University of Florida.
Vandenboer, K., Van Beek, V., & Bezuijen, A. (2014). 3D finite element method (FEM) simulation of groundwater
flow during backward erosion piping. Frontiers of Structural and Civil Engineering, 8(2), 160–162. doi:10.1007/
s11709-014-0257-7
Wan, C. F., & Fell, R. (2004). Investigation of rate of erosion of soils in embankment dams. Journal of Geotechnical
and Geoenvironmental Engineering, 130(4), 373–380. doi:10.1061/(ASCE)1090-0241(2004)130:4(373)
Wong, Y. W. (1981). Three dimensional finite element analysis of a quantitative piping theory. Report Presented to
the Department of Civil Engineering at the University of Florida in Partial Fulfillment of the Requirements for the
Degree of Master of Engineering.
Zhen-Ming, S., Hong-Chao, Z., Song-Bo, Y., Ming, P., & Tao, J. (2018). Application of CFD-DEM to investigate seepage
characteristics of landslide dam materials. Computers and Geotechnics, 101, 23–33.

You might also like