Computers and Geotechnics 129 (2021) 103820

Contents lists available at ScienceDirect

Computers and Geotechnics

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

Research Paper

A two-dimensional, adaptive finite element approach for simulation of T

backward erosion piping
B.A. Robbinsa,b, , D.V. Griffithsb
⁎

a
U.S. Army Engineer Research and Development Center, 3909 Halls Ferry Road, Vicksburg, MS 39180, USA
b
Department of Civil and Environmental Engineering, Colorado School of Mines, 1500 Illinois Street, Golden, CO 80401, USA

ARTICLE INFO ABSTRACT

Keywords: Backward erosion piping (BEP) is a type of internal erosion that significantly contributes to internal erosion
Internal erosion being a leading cause of levee and dam failures, second only to overtopping. Despite its significance, very few
Backward erosion piping numerical or analytical tools exist for assessing the progression of backward erosion pipes. Further, existing tools
Finite elements do not automatically refine the mesh near the pipe tip, nor do they automatically search for the weakest erosion
Adaptive mesh refinement
path in the foundation. Both features are critical for assessing piping when conditions along the pipe path vary
due to natural geologic variability or engineered features such as filter trenches and partial cutoff walls. In the
present study, a two-dimensional, auto-adaptive, finite element program has been developed that solves the
coupled groundwater flow and erosion pipe flow equations in a piecewise, steady-state manner to simulate BEP
progression using automated, local mesh refinement. Criteria for pipe advancement and pipe widening are
presented that permit pipe growth through the mesh along the path of least resistance, and an approximation to
account for foundation depth is introduced. Examples are presented in which BEP progression is simulated
beneath a levee to illustrate the analysis method.

1. Introduction profession for tools enabling engineers to assess failure risks due to BEP.
The progression of BEP has been shown to be driven by two distinct
Backward erosion piping (BEP) is a type of internal erosion by processes, i.e., primary erosion at the pipe tip and secondary erosion in
which small, shallow erosion channels propagate upstream through the pipe channel as illustrated in Fig. 1 (Hanses, 1985; Van Beek, 2015).
foundation material along the contact of a pervious, sandy layer and a Primary erosion refers to the erosion of soil particles near the pipe tip
semi-impervious clay layer as illustrated in Fig. 1. The erosion initiates by highly concentrated, local seepage forces. Laboratory measurements
during high water events when concentrated seepage passes through have demonstrated that a critical value of the horizontal, hydraulic
defects in the clay layer caused by natural causes (e.g., animal activity gradient a few centimeters in front of the advancing pipe may provide a
or dessication cracks) or heave induced cracking during a high-water suitable criterion for primary erosion (Robbins et al. 2018b). Secondary
event. The concentrated seepage carries eroded material to the ground erosion refers to the erosion and transport of sand grains in the pipes
surface where it is deposited in a cone shaped mound often referred to that form. The sediment eroded from the pipe tip must be conveyed
as a sand boil. If the water level is high enough, the erosion pipes will through the pipe for the pipe to remain open and erosion to continue.
progress upstream through the foundation (opposite the flow direction) The secondary erosion also controls the dimensions of the pipes that
along the sand/clay interface, potentially undermining the structure form, and, therefore, the head loss that occurs in the pipes. Laboratory
leading to failure. Conditions like that shown in Fig. 1 naturally occur measurements of the hydraulic conditions in BEP channels have de­
along many meandering river systems such as the Mississippi River monstrated that classical incipient motion criteria for sand provides a
(Saucier, 1994). Because this geologic configuration represents ideal suitable secondary erosion criterion for assessing the pipe dimensions
conditions for BEP, BEP is a primary source of failure for levees, second and corresponding head loss (Van Beek et al., 2019).
only to overtopping (Baker, 2018). BEP is also a leading cause of dam While an understanding of the mechanics of BEP has been obtained
failures (Foster et al., 2000, Richards and Reddy, 2007). BEP is there­ through numerous studies over the last 40 years, e.g., De Wit, 1984;
fore an issue of major concern for engineers involved in managing ex­ Hanses, 1985; Sellmeijer, 1988; Schmertmann, 2000; Van Beek, 2015;
isting flood control embankments, and there is a great need in the Robbins et al., 2018b; Vandenboer, 2019, very few tools for assessing

⁎
Corresponding author at: U.S. Army Engineer Research and Development Center, 3909 Halls Ferry Road, Vicksburg, MS 39180, USA.
E-mail address: [email protected] (B.A. Robbins).

https://doi.org/10.1016/j.compgeo.2020.103820
Received 8 March 2020; Received in revised form 24 July 2020; Accepted 31 August 2020
Available online 25 November 2020
0266-352X/ Published by Elsevier Ltd.
B.A. Robbins and D.V. Griffiths Computers and Geotechnics 129 (2021) 103820

Nomenclature L2 width of the base of the levee [m]

L3 distance from levee to downstream boundary [m]
Notation n initial pipe depth multiplier (initial depth as number of
sand grains) [–]
a depth of the erosion pipe [m] Ni FE shape functions
A cross-sectional area of the erosion pipe [m2] p pore pressure [Pa]
d thickness of the sand layer in the z-direction [m] P wetted perimeter of the pipe cross section [m]
d50 sand grain size for which fifty percent is finer by mass [m] q specific discharge [m2/s]
DH hydraulic diameter of the erosion pipe [m] Q total discharge in erosion pipe [m3/s]
f Darcy–Weisbach friction factor [–] {Q} total, net flow quantities at FE nodes [m3/s/m]
g gravitational acceleration [m/s2] Re Reynolds number
h piezometric head [m] t target (minimum) element size [m]
H piezometric head applied to the upstream boundary [m] V mean flow velocity in the erosion pipe [m/s]
Hcr critical value of the piezometric head on the upstream w pipe width [m]
boundary [m] x spatial coordinate in the x-direction [m]
{H } vector of nodal heads [m] X length of analysis domain in the x-direction [m]
i hydraulic gradient [–] Xi x-coordinate of initiation point [m]
ic critical value of the hydraulic gradient for pipe progres­ y spatial coordinate in the y-direction [m]
sion [–] Y length of analysis domain in the y-direction [m]
k hydraulic conductivity [m/s] Yi y-coordinate of initiation point [m]
kx hydraulic conductivity in the x direction [m/s] z spatial coordinate in the direction normal to analysis plane
ky hydraulic conductivity in the y-direction [m/s] roughness of the pipe wall [m]
kpL pipe hydraulic conductivity for laminar flow conditions w unit weight of water [N/m3]
[m/s] µ dynamic viscosity of water [Ns/m3]
kpT pipe hydraulic conductivity for turbulent flow conditions e element subdomain
[m/s] s
soil domain
[K e] global, assembled conductivity matrix for FE problem [m/ p
pipe domain
s] density of water (kg/m3)
L total length of analysis domain or total length of pipe [m] hydraulic boundary shear stress on pipe wall (Pa)
Lp total length of pipe path [m] c critical hydraulic shear stress for incipient motion of sand
L1 distance from levee to upstream boundary [m] grains (Pa)

Fig. 1. Illustration of the backward erosion piping process beneath a typical levee.

BEP progression exist. Current practice relies upon combinations of represented, i.e., (1) models that simply increase pipe zone perme­
theoretical and empirical tools used to extrapolate critical boundary ability within a routine seepage analysis to examine the influence of the
conditions measured in laboratory tests to the field scale (Sellmeijer pipe on the groundwater flow, (2) multi-phase continuum models in
et al., 2011; Schmertmann, 2000). Because these methods rely on ad­ which erosion and transport of eroded particles are explicitly accounted
justing critical boundary conditions measured in the lab to equivalent for, and (3) discrete element method (DEM) simulations typically
field values using semi-empirical factors, they have trouble accounting coupled to a continuum description of fluid flow.
for the influence of scale effects and important variables, such as pro­ The first category of models can be further subdivided based on the
blem geometry and geology, that alter the relationship between approach taken to represent the increased hydraulic conveyance pro­
boundary conditions and local gradients near the pipe tip (Robbins and vided by the erosion pipe. The simplest approach involves representing
Van Beek, 2015). Despite this limitation, these methods remain the only the pipe as a zone of increased permeability in which the hydraulic
methods applied in practice on a regular basis (USBR and USACE, 2015; conductivity of the soil is increased by a constant factor (Hagerty and
Van Beek et al., 2013). Curini, 2004; Jianhua, 1998; Vandenboer et al., 2013; Robbins, 2016;
While no generally accepted approach for modeling BEP exists Fascetti and Oskay, 2019). A more accurate approach is to determine
today, numerous approaches for modeling BEP have been proposed in the conductivity of the pipe zone based on pipe flow equations in which
the literature. Wang et al. (2014) classified modeling approaches into the pipe dimensions are determined according to principals of sediment
three broad categories based upon how the piping process is transport (Sellmeijer, 2006; Zhou et al., 2012; Van Esch et al., 2013;

2
B.A. Robbins and D.V. Griffiths Computers and Geotechnics 129 (2021) 103820

pipe and will therefore predict the maximum erosion extent. Category 1
models only require a criterion for the pipe evolution; whereas category
2 models also require phase transfer relationships which can be difficult
to determine due to the high dependency on the hydromechanical load
path followed (Rochim et al., 2017). For these reasons, the authors
selected a steady state category 1 modelling approach for simulating
BEP.
As pipe progression has been shown to be a highly localized phe­
nomenon (Robbins et al., 2018b; Van Beek, 2015; Vandenboer, 2019),
provisions for auto-adaptive mesh refinement near the erosion pipe
have also been included to allow for efficient computation of the local
hydraulic gradient near the pipe tip. Combined with a threshold hy­
draulic gradient for pipe progression (primary erosion), the pipe can
then be programmed to automatically follow the path of least re­
sistance. An auto-adaptive approach that searches for the path of least
resistance is required for cases where the critical pipe path is not known
a priori, as is the case in variable geologic environments or situations
where engineered features may alter the path (e.g., barriers and relief
wells). In cases where the pipe path is known, a mesh can be generated
with local refinement along the pipe path to permit analysis of pipe
progression, negating the need for auto-adaptivity. However, even in
these cases, the auto-adaptive mesh refinement provides advantages by
(1) permitting the use of any starting mesh (no custom refinement along
features) and (2) using small elements only in the vicinity of the si­
mulated pipe (instead of along the full pipe path for each calculation
step). Liang et al. (2017) also presented an auto-adaptive approach for
simulating internal erosion; however, their moving mesh method only
Fig. 2. (a.) Finite element discretization of x-y analysis plane for two-dimen­ attempted to align mesh boundaries with erosion boundaries and did
sional analysis of pipe progression and (b.) head and pipe depth profile along
not increase the resolution of the solution in the vicinity of the erosion
centerline of the erosion pipe.
pipe as required for assessing BEP progression. The authors are not
aware of further work attempting to model BEP in an auto-adaptive
Robbins and Griffiths, 2018). Models of this first category can make use finite element framework that searches for the path of least resistance.
of both steady state, e.g., Robbins and Griffiths, 2018, or transient e.g., Such a model is presented and demonstrated herein.
Van Esch et al., 2013, groundwater solutions; however, transient as­
pects of the sediment transport in the erosion pipe are not captured.
The second category of models in which multi-phase continuum 2. Model description
formulations are employed can simulate the transient aspects of the
sediment transport in the erosion pipe. Models in this category separate Fig. 2a illustrates a plan view of the analysis plane A-A along the top
the problem into phases representing the pore fluid, soil structure, and of the sand layer in Fig. 1. The domain is subdivided into a soil domain
eroded soil particles. As far as the authors are aware, this approach was and an erosion pipe domain. The upstream and downstream boundaries
first proposed as a means of predicting wellbore sand production are constant, total head boundary conditions. The remaining bound­
(Stavropoulou et al., 1998) but has since been used in various forms by aries are no-flow boundaries. The analysis plane is initially discretized
many authors for simulating internal erosion (Wang et al., 2014; into a structured mesh of quadrilateral elements according to the spe­
Cividini and Gioda, 2004; Becker et al., 2010; Steeb et al., 2007; Liang cified base element size as shown in Fig. 2a.
and Chen, 2011; Fujisawa et al., 2010; Luo, 2013; Zhang et al., 2012; Additionally, a target, minimum element size (t) for the erosion pipe
Rotunno et al., 2019; Callari and Froiio, 2020). This approach requires discretization must be specified. The pipe is initiated at a specified in­
a constitutive relationship defining the rate at which soil particles are itiation point (Xi,Yi) by refining the mesh near this point until the tar­
transferred from the solid soil structure to the eroded particle phase but geted minimum element size is reached. The element in which the in­
provides the advantage of being able to simulate the temporal evolution itiation point falls is the initiation element and represents the initial
of erosion. extent of the pipe domain. As the initiation element represents a see­
The third category of models captures the particle scale micro­ page exit where erosion begins, a boundary condition must be assigned
mechanics of the BEP process by using DEM models to capture the to the nodes of this element. For this work, the initiation element was
movement of individual particles. The fluid forces on the soil particles fixed to a constant head equal to the head applied to the downstream
are typically resolved through coupling of the DEM model of the soil boundary. However, alternate boundary conditions could readily be
skeleton to a continuum model of the fluid phase, e.g., El Shamy and explored. To simulate pipe progression, the upstream head is then in­
Aydin, 2008; Lominé et al., 2013; Maeda et al., 2010; Sari et al., 2011; creased incrementally, the coupled groundwater flow–pipe flow equa­
Wang and Ni, 2013; Zou et al., 2013; Tran et al., 2016, and Froiio et al., tions are solved, the mesh is refined, and the pipe progression is si­
2019. This category of models is computationally expensive and is mulated in a piece-wise, steady state manner. The following sections
therefore used primarily to study the fundamental mechanics of erosion present the details of each of these steps in the analysis.
through simulations of representative elementary volumes. To be useful
in practice, the observed behavior must then be upscaled to continuum 2.1. Governing equations and discretization
models to assess the full-scale behavior. Much remains to be done in
this area before these models are useful in engineering practice. 2.1.1. Groundwater flow
Steady state models of the first category provide a conservative The groundwater flow in the soil domain of the x-y analysis plane
analysis of the pipe progression as the steady state groundwater solu­ (Fig. 2a) is treated as steady state, laminar flow, which is governed by
tion yields the highest hydraulic gradients in the vicinity of the erosion the Laplace Equation given as Eq. (1)

3
B.A. Robbins and D.V. Griffiths Computers and Geotechnics 129 (2021) 103820

h h can thus be satisfied by solving Eq. (1) across the domain of the pipe
kx + ky =0
x x y y (1) with k x = k y = kpL .
For turbulent flow conditions (Re > 2500) , Eqs. (7)–(9) no longer
where k x and k y represent the hydraulic conductivities in the x and y hold true. In the case of turbulent flow, an analytical relation for f does
directions, respectively, and h = z + p/ w denotes the total head (po­ not exist and the head loss must be expressed as a function of f . Eq. (6)
tential) in terms of the elevation head z and the pressure head given in can still be rearranged to match the form of Darcy’s law
terms of the pore pressure p and unit weigh of water w .
1
4a3g dh dh dh
q= = kpT
2.1.2. Pipe Flow f dx dx dx (10)
Under steady state conditions, the flow in pipes of any shape can be
described by the Darcy–Weisbach equation where
1
V2 L 4a3g dh
dh = f kpT =
2g DH (2) f dx (11)
where dh is the head loss in the pipe, f is the friction factor, V is the is the equivalent hydraulic conductivity for turbulent flow in a pipe
mean flow velocity, g is the gravitational acceleration, L is the length of of depth a . The friction factor f must be determined using the em­
the pipe, and DH = 4A/ P is the hydraulic diameter for a pipe with area pirically derived Moody diagram or similar relations. For this work, an
A and wetted perimeter P . The erosion pipes that form are typically explicit relationship proposed by Romeo et al. (2002) was used to ob­
very shallow in depth (order of 1–10 sand grains according to Van Beek, tain f as follows
2015; Vandenboer, 2019) and have pipe widths that are on the order of
0.9924
10–50 times the pipe depth (Van Beek, 2015; Robbins et al., 2018a; 1 DH 5.0272 DH 4.567 DH
Vandenboer, 2019). As such, the erosion pipes that form can be con­ = 2 log log log
f 3.7065 Re 3.827 Re 7.7918
sidered to be rectangular in shape with width-to-depth ratios greater
than 10. For rectangular pipe cross sections of width w and depth a , the
0.9345
hydraulic diameter is +
5.3326
208.815 + Re
4wa (12)
DH =
2(w + a) (3)
where denotes the pipe wall roughness. For this study, it was assumed
Considering the case of w a, DH can be approximated by DH that = d50/2 . As the Reynolds number Re , and in turn the friction
where factor f ,depend on the specific discharge q , kpT cannot explicitly be
DH = 2a (4) solved for. Instead, the turbulent pipe flow is solved iteratively by
solving Eq. (1), but with k x = k y = kpT in the pipe domain, until con­
The average pipe velocity in the pipe is given by vergence on f is obtained. In this manner, the turbulent pipe flow
V = Q / A = Q/ wa = q/ a (5) equations for a pipe with depth a are satisfied.

where Q denotes the total flow rate in the pipe and q is the specific
2.1.3. Finite element discretization
discharge (flow per unit width) in the pipe. Substituting Eqs. (4) and (5)
Using the Galerkin weighted residual method, the finite element
into Eq. (2) and expressing differentially yields the following governing
discretization of the coupled soil and pipe flow is (e.g. Smith and
equation for the pipe flow
Griffiths, 2004)
dh f q2
= 3 [K e ]{H } = {Q} (13)
dx 2a 2g (6)
where {H } and {Q} represent the total head and net inflow/outflow at
Under laminar flow conditions with the assumption that w a, the the FEM nodes and
friction factor is given by
96 96µ Ni Nj N Nj
f= = [K e] = s kx + ky i d s
e
Re VDH (7) s e x x y y
Ni Nj Ni Nj
where Re = VDH / µ is the dimensionless pipe Reynolds number given + p kpipe + d p
e
as a function of the fluid density , mean flow velocity V , hydraulic
p e x x y y (14)
diameter DH , and dynamic fluid viscosity µ . By substituting the laminar where s is the soil domain, p is the pipe domain, Ni are the ele­
friction factor (Eq. (7)) into the pipe flow equation (Eq. (6)) and re­ ment shape functions, the subscript e denotes the element subdomains
arranging, the pipe flow equation takes the form of Darcy’s equation of both the soil and pipe domains, and kpipe denotes the hydraulic
a3 g dh dh conductivity of the pipe elements which is given by Eqs. (9) or (11),
q= = kpL
12µ dx dx (8) depending on the flow regime. In this work, 4-node elements with
linear shape functions are used. Solution of Eq. (13) yields the values of
where total head for the coupled flow solution for an erosion pipe of known
a3 g depth a . However, the depth of the erosion pipe is not known a priori.
kpL = The depth of the erosion pipe that forms is dictated by the critical,
12µ (9)
hydraulic shear stress of the soil being eroded. For flow between par­
is the equivalent hydraulic conductivity for laminar pipe flow in a allel plates, the hydraulic shear stress on the pipe boundary is given by
pipe of depth a . As noted by Harr (1962), this analogue between Darcy’s
a
law and laminar flow relations for flow between parallel plates was = w
h
2 (15)
recognized long ago and formed the basis for viscous flow physical
models of groundwater flow. This relationship was also previously de­ where h is the gradient of h . The critical hydraulic shear stress ( c ) at
rived for BEP pipe flow through simplification of Navier Stokes equa­ which the sand grains in the pipe erode can be estimated from incipient
tions by Sellmeijer (1988). For laminar flow conditions, the pipe flow motion criteria. The approximate relationship

4
B.A. Robbins and D.V. Griffiths Computers and Geotechnics 129 (2021) 103820

c (Pa) = d50 (mm) (16) demonstrated both experimentally (de Wit, 1984) and theoretically
(Sellmeijer, 1988, 2006). The physical explanation for this lies in the
suggested by Briaud et al. (2017) has been shown to yield reason­
fact that thicker soil layers can convey more water towards the pipe,
able estimates of c for fine sands. If > c , the erosion pipe will deepen
which in turn leads to higher flow rates in the pipe, larger pipe di­
further as the sand on the bottom of the pipe is displaced by the flow.
mensions, and lower head losses in the erosion pipe. While this cannot
The erosion pipe is initiated to a depth of a = nd50 when a pipe element
be fully accounted for in the two-dimensional model due to the highly
is activated where n is a user-defined value. To determine the pipe
three-dimensional flow field around the pipe, an approximation can be
depth ai in each pipe element, Picard iterations are conducted over a
made to partially correct for this factor. Consider a plan view analysis as
until < c in all elements in p yielding the spatial function of pipe
illustrated in Fig. 2 for the scenario in Fig. 1 with a sand layer of
depth as shown in Fig. 2b. This will be elaborated on further when
thickness d where k = k x = k y . The specific flow rate through the soil
discussing the simulation algorithm (Section 2.6).
layer of thickness d is given by Darcy’s law as

2.2. Pipe progression qs = dk h (19)

Thus, it is readily seen that the thickness of the sand layer can be
The progression of the erosion pipe is controlled by the local hy­ accounted for in the plan view analysis by scaling the soil hydraulic
draulic gradient as suggested by numerous researchers (Hanses, 1985; conductivity by a factor d . Applying d as a correction to the hydraulic
Schmertmann, 2000; Van Beek, 2015). Robbins et al. (2018b) measured conductivity yields a depth corrected hydraulic conductivity for all
the local hydraulic gradient near the pipe tip in laboratory experiments elements in the soil domain given by
and found that the pipe advanced at a consistent value of the hydraulic
gradient over a distance of 10 cm upstream of the pipe tip for multiple kd = d × k (20)
experiment scales. This suggests that the local hydraulic gradient near where the depth corrected value of hydraulic conductivity (kd ) is ex­
the pipe tip may be a suitable, measurable criteria for assessing pipe pressed in terms of the original soil hydraulic conductivity (k ) and the
progression. For the present study, the analysis plane is a horizontal aquifer depth (d ). By scaling the hydraulic conductivity of all elements
plane, so only the horizontal gradient will be considered, and no cor­ in the soil domain but leaving the pipe elements as previously de­
rections for pipe path inclination will be made. A critical value of the scribed, the flow quantities in the pipe elements increase leading to
local hydraulic gradient, ic , is specified. The pipe can progress into any larger pipe depths as would occur in the situation of deeper aquifers in
soil element adjacent to the pipe domain if the condition three dimensions. While the result of this approximation will be illu­
h 2 > ic (17) strated through examples presented later, it should be reiterated that
the two-dimensional approximation will not fully capture the nature of
is satisfied where 2 denotes the vector magnitude (L2 norm of the the highly three-dimensional problem. Future studies must be con­
gradient vector). ducted using three-dimensional models to evaluate the adequacy of this
approximation.
2.3. Pipe widening
2.5. Mesh refinement
In addition to upstream progression of the pipe, the erosion pipe
will naturally widen as the depth increases. The minimum width of the As mentioned previously, the solution must be refined near the pipe
pipe that can be represented corresponds to the target element size (t ) . to (1) obtain hydraulic gradient estimates over sufficiently small dis­
As this represents the smallest element used, the pipe cannot have di­ tances to assess pipe progression and (2) obtain elements small enough
mensions smaller than this target size. If the pipe width, w , becomes to represent the erosion pipe domain. A myriad of strategies exist for
larger than the target element size, the pipe must be represented by refining finite element solutions. The solution can be locally refined by
multiple elements. This is handled by switching all neighboring soil decreasing the element size (h-refinement), increasing the polynomial
elements to pipe elements if order of the shape functions on select elements (p-refinement), redis­
Ra tributing nodes within an existing mesh (r-refinement), or any combi­
>2 nation of the three approaches. Other strategies such as multigrid
t (18)
techniques may also be used to obtain locally refined solutions. For the
where R = w /a is the user-defined width-to-depth ratio for the erosion present application, the authors chose to use solely h-refinement as this
pipe. As this check is made for all pipe elements in the domain p , the approach yields a refined estimate of the solution surrounding the pipe
neighbors of the newly activated neighbors will also be switched to pipe while providing complete control over the element size being used to
elements if the condition in Eq. (18) is met. In this manner, the pipe can represent the pipe subdomain.
be represented by as many elements as necessary to maintain the pre­ For h-refinement on structured, quadrilateral meshes, the easiest
scribed width-to-depth ratio. approach to implement is an element subdivision algorithm in which
existing quadrilateral meshes are divided into four equally sized ele­
2.4. Foundation depth correction ments as illustrated in Fig. 2a. This approach yields irregular meshes
with “hanging nodes” that are not connected to neighboring elements.
The plan view analysis illustrated in Fig. 2a is formulated per unit To obtain a continuous solution with hanging nodes, it is necessary to
length in the z-direction (perpendicular to the analysis plane). As such, constrain the value of the solution at the hanging nodes as described by
an analysis conducted in this manner computes flow quantities into the Rheinboldt and Mesztenyi (1980) or Bangerth et al. (2007). For the
pipe that are equivalent to flow quantities associated with a 1-m thick mesh in Fig. 2a which has only 1 degree of irregularity, i.e., 1 hanging
soil layer. The analysis assumes that all flow in the soil is horizontal, all node permitted per edge, the solution at the hanging node is con­
flow in the pipe is horizontal, and no flow occurs in the soil beneath the strained to the average value of the neighboring nodes. Let hi be the
erosion pipe. Unfortunately, most BEP analyses are for scenarios with value of the solution at hanging node i . The value of hi is then simply
foundations much thicker than 1 m. As the pipe remains very shallow constrained to the value
relative to the thickness of the sand layer, the flow concentrates
1
strongly towards the pipe in the vertical direction with increasing hi = (hc,1 + hc,2)
2 (21)
concentration of flow as the foundation depth increases. As a result, the
pipe progresses more readily in layers of greater depth, which has been where hc,1 and hc,2 are the values of the solution at the two constraining

5
B.A. Robbins and D.V. Griffiths Computers and Geotechnics 129 (2021) 103820

nodes neighboring the hanging node. This constraint can be enforced The values of ii at the FE nodes are calculated as the average value
either by condensing the global system of equations if assembling the of all values of i at the nodes
full matrices (Bangerth et al., 2007), through a penalty approach (Smith m
1
and Griffiths, 2004), or by enforcing the constraint within an iterative ii = ii, j
solver as suggested by Sheu et al. (1999). This latter approach was m j=1 (24)
taken in this study as an element-by-element preconditioned conjugate
where ii, j denotes the nodal value of i at node i obtained from elements
gradient solver was employed.
j = 1, , m that share node i . An indicator of the relative error in the
For adaptive refinement of the solution, it is also necessary to spe­
element can then be estimated as
cify criteria that define when and where the mesh is refined. Two cri­
teria were used to specify local mesh refinement: (1) all elements ad­ (1/2)
e e T (e ) d e
jacent to pipe elements were adapted until the element size was equal = =
e
e (1/2)
to or less than the target element size and (2) elements with an error i (i )2d e (25)
estimator that was too large were also refined. The first criterion
e

combined with only 1 degree of irregularity being permissible ensured a Limiting values of e can then be defined to obtain further refine­
refined mesh in the vicinity of the erosion pipe. The second criterion ment of the solution in areas of highest error.
allowed for additional refinement to be specified in areas with high
hydraulic gradients to resolve the local hydraulic gradients near the 2.6. Simulation algorithm
pipe tip accurately. For this study, the Zienkiewicz and Zhu (1987)
error indicator was calculated for the gradient field as the local hy­ The simulation of pipe progression is conducted in a piecewise,
draulic gradient was of primary interest. The error in the gradient field steady-state manner. A diagram of the algorithm steps is provided in
is defined as Fig. 3. The mesh is initially refined to the target element size in the
(22) vicinity of the initiation point until the element containing the initia­
e=i i
tion point is smaller than the target element size. In this study, all mesh
where i = h is the exact solution, and i = Ni hi is the approximate refinement was conducted using the SODA Fortran library described in
solution obtained from the finite element formulation. As the exact Robbins and Griffiths (2020). The element containing the initiation
solution is unknown, an estimate i i is made following the averaging point is switched from the soil domain to the pipe domain, and the pipe
approach of Burkley and Bruch (1991) where depth in the initiation element is set to the initial value of n d50 . All soil
elements adjacent to pipe elements are refined until they are equal in
i = Ni ii (23)
size to the pipe elements. Nominal initial boundary conditions are

Fig. 3. Diagram of simulation algorithm for piping analysis.

6
B.A. Robbins and D.V. Griffiths Computers and Geotechnics 129 (2021) 103820

Table 1 0.045
Material properties for example analyses.
0.040
Property Name Value

kx,ky (m/s) Hydraulic conductivity 1.0 × 10-5 0.035

Avg. Gradient
d50 (m) Grain size 2.5 × 10-4
n Initial pipe depth multiplier 3 0.030
ic Critical hydraulic gradient 0.30
τc (Pa) Critical shear stress 0.25
0.025
ρ (kg/m3) Water density 1000
μ (Ns/m3) Water dynamic viscosity 1.0 × 10-3
R Pipe width-to-depth ratio 20 0.020
Run 1
Run 2
0.015 Run 3
assigned, and Picard iterations over the pipe depth are conducted, in­ 0.010
creasing the depth by d50/2 if > c , until the sediment equilibrium 0.3 0.4 0.5 0.6 0.7 0.8 0.9
condition of < c is satisfied in the pipe elements. Once the pipe depth (L-x)/L
is correct in all pipe elements, the error indicator is calculated, and
Fig. 5. Normalized pipe location as a function of average hydraulic gradient.
elements are checked to see if further refinement is necessary. Once the
mesh is adequately refined, the magnitude of the hydraulic gradient in
elements adjacent to the pipe is checked to see if the pipe can progress and the analysis was repeated until the pipe passed through the domain
further. In the present study, only 1 soil element was switched to a pipe or the head on the upstream boundary reached a maximum value.
element during each progression step to ensure a single pipe path is
obtained. A single pipe is preferred as it causes greater flow con­
3. Example analyses
centration thereby resulting in the most conservative analysis results.
The element with max|i ic | was switched to a pipe element during each
A series of example analyses have been performed to demonstrate
progression step. This process was repeated until i < ic in all soil ele­
the analysis approach, assess the method repeatability, and demon­
ments. Upon reaching equilibrium, the upstream head was increased,
strate the result of the approximate depth correction factor introduced.

Fig. 4. Resulting (a) pipe path for example levee simulation with initiation point (185,50) and (b) corresponding depth of the erosion pipe.

7
B.A. Robbins and D.V. Griffiths Computers and Geotechnics 129 (2021) 103820

Fig. 6. Asymmetrical refinement around the erosion pipe leading to pipe Fig. 8. Pipe location curves for the original analysis (Pt 1) and the analysis with
meandering. the initiation point perturbed up and down (Pts 2 and 3).

Fig. 7. Resulting erosion pipe paths obtained for cases with the initiation point (a) perturbed down 10-m in the y-direction (185,40) and (b) perturbed up 10-m in the
y-direction (185,60).

8
B.A. Robbins and D.V. Griffiths Computers and Geotechnics 129 (2021) 103820

The downstream boundary condition is set to h = 0 while the up­

stream boundary condition is initialized to h = 2.65 m corresponding to
an average hydraulic gradient across the domain of 0.01. When pipe
equilibrium is reached, the upstream boundary condition is increased in
1% intervals until the pipe propagates through the entire domain.
The initiation point was arbitrarily selected to be 20 m downstream
of the levee toe at an x-location of 185 m and a y-location of 50 m
(center of domain). The initial, base element size was set to a value of
5.0 m, and the layer depth was 1.0 m (no depth correction). The flow in
the pipe was initially simulated using only laminar flow assumptions
(Eq. (9)) as this led to much faster calculations by eliminating the
iteration on the friction factor f . As will be demonstrated in following
sections, this assumption was reasonable for the layer depth of 1.0 m.
The resulting pipe path and calculated pipe depth for the main
piping channel are illustrated in Fig. 4. The relatively uniform mesh
refinement along the pipe (Fig. 4a) is obtained due to criterion 1 re­
quiring all adjacent elements to be the same size as the pipe elements.
The additional refinement located downstream of the initiation point
and in a few locations along the pipe is due to the calculated error
indicator. Both refinement criteria described in Section 2.5 can be ad­
justed to obtain varying levels of refinement.
The position of the pipe within the domain at equilibrium points is
plotted as a function of the average hydraulic gradient applied across
the domain in Fig. 5. The pipe progresses with each increase in the
upstream head until a critical value is reached. At this point, the pipe
progresses continuously through the domain with no further increase in
the upstream boundary conditions. This result is similar to results ob­
tained by Sellmeijer (2006) and van Esch et al. (2013).
For this homogeneous case, the pipe should progress straight rather
than meander as shown in Fig. 4. Upon further inspection of the mesh
near the upstream end of the erosion pipe (Fig. 6), it was recognized
that the pipe meandering was being caused by variations in the hy­
draulic gradient magnitude due to asymmetric mesh refinement around
the pipe. As shown in Fig. 6, one of the corner elements has a hydraulic
gradient of 0.399, which is higher than the element directly in front of
the pipe, due to the mesh refinement pattern that developed. Because of
this, the pipe progresses to this corner element, activating both the
corner element and element with a gradient of 0.379 to maintain edge
Fig. 9. Influence of aquifer depth on (a) critical average gradient for laminar connectivity of pipe elements. The result of this is the pipe progresses in
and turbulent flow assumptions and (b) calculated Reynolds number. sporadic directions based on the mesh refinement. To investigate the
influence this may have on the results, two tests were run. First, three
identical runs of the example problem were run to ensure the analysis
For the example analyses, consider the levee problem illustrated in approach is reproducible. Second, two additional runs of the example
Fig. 1 for a case in which the embankment is 10 m in height with 3H:1V problem were run with the initiation point perturbed 10 m up and
side slopes and a 5-m crest width. The base of the embankment is thus down in the y-direction to see if the critical value of the average hy­
L2=65 m in width. Letting L1 = L3 = 100m , the total length of the draulic gradient is sensitive to the pipe path taken. For the repeat runs,
analysis plane is L = 265 m in length. Let us consider a domain that is identical pipe patterns to the first run shown in Fig. 4 were obtained in
100 m in width (normal to the plane of Fig. 1) such that the analysis each case (Fig. 5).
plane is 265 m in length and 100 m in width. As fine sand is the primary The results obtained from the two additional runs with the initiation
material of concern for BEP, a fine sand with material properties shown point perturbed are shown in Fig. 7a and b. As seen from these figures,
in Table 1 is used for these example analyses. The hydraulic con­ the resulting pipe paths obtained with the two different initiation points
ductivity was selected based upon d50=0.25 mm. The hydraulic shear were different than the original pipe path illustrated in Fig. 4. Ad­
stress was determined using Eq. (16). For all analyses conducted, the ditionally, minor differences in the pipe location curves (Fig. 8) were
target element size was 0.2 m. This element size was selected for also observed. However, despite these minor differences, the maximum
computational efficiency; use of smaller elements resulted in slower value of the average hydraulic gradient required for full pipe progres­
analyses. The critical gradient, ic , was selected based on available in­ sion (0.041) was identical for all three cases (Fig. 8). This indicates that
formation in the literature. Robbins et al. (2018b) measured values of the critical, average, hydraulic gradient across the domain (or the cri­
ic = 0.35 0.43 for sands with d50 = 0.3 mm over distances of 0.1 m. tical upstream head) may not be sensitive to the exact pipe path taken,
The value of ic will decrease with distance away from the pipe. An at­ and the proposed adaptive meshing approach may be adequate for
tempt to account for this was made by selecting ic = 0.3 for the 0.2 m engineering purposes. Nevertheless, more work is needed to address the
element size used in this study. While a more rigorous procedure for numerical meandering observed in these analyses to make the pipe
determining ic would be necessary in practice, this was deemed suffi­ paths follow the theoretically anticipated paths.
ciently accurate for the sole purpose of demonstrating the analysis In addition to the runs discussed, the example analysis with the
approach in the sections that follow. Lastly, a value of R = 20 was initiation point centered in the domain was also run with various sand
selected for the pipe width-to-depth ratio as it was in the middle of layer thicknesses, d, to assess the performance of the depth correction
measured average width/ depth ratios determined by Van Beek (2015). factor proposed. Increasing d leads to increased flow in the erosion

9
B.A. Robbins and D.V. Griffiths Computers and Geotechnics 129 (2021) 103820

Fig. 10. Analysis results for turbulent flow with d = 1 m showing (a) pipe path at critical pipe length and (b) corresponding Reynolds number and pipe depth in the
main pipe channel.

pipes and higher turbulence levels. As such, analyses were run (1) as­ hydraulic gradient required to cause the pipe to fully progress through
suming solely laminar flow and (2) allowing either laminar or turbulent the domain decreased as the aquifer depth increased for both flow as­
flow based upon the Reynolds number. This was done to also assess the sumptions (Fig. 9a). Further, as the depth increased, the Reynolds
influence of turbulent flow on piping calculations as the laminar flow number in the pipe increased (Fig. 9b) such that the flow transitioned
assumption has often been made in the literature (van Esch et al., 2013; from laminar to turbulent flow at an aquifer depth of approximately
Robbins and Griffiths, 2018). The maximum value of the average 10 m. For depths less than this, the flow was laminar due to the low

10
B.A. Robbins and D.V. Griffiths Computers and Geotechnics 129 (2021) 103820

1.8 4. Conclusions
Sellmeijer
1.6 Laminar
Laminar & Turbulent An auto-adaptive, piecewise steady-state, finite element program
was developed and demonstrated for simulating progression of back­
1.4
ward erosion piping in plan-view for both laminar and turbulent flows
in the erosion pipe. Adaptive mesh refinement permitted analysis of
1.2
H cr/H cr,d=1m

pipe progression at length scales appropriate for assessing both the

1 hydraulics within the erosion pipes and the hydraulic gradients near the
pipe tip that are responsible for pipe progression. While the path of the
0.8 erosion pipe was found to be influenced by the mesh refinement, the
critical upstream head at which the pipe progressed to failure was not
0.6 sensitive to the pipe path in the simulations conducted. As such, the
adaptive meshing approach presented may prove to be a suitable en­
0.4 gineering tool for initial assessment of pipe progression beneath dams
and levees.
0.2 An approximate correction factor for aquifer thickness was found to
-1 0 1 2 3
10 10 10 10 10
yield results similar to prior studies where the critical gradient de­
Depth (m)
creases with increasing aquifer depth. While increasing the aquifer
Fig. 11. Influence of aquifer depth in the present FE model compared to the thickness resulted in increasingly turbulent flows, the influence of
Sellmeijer theory. turbulence on the critical gradient was found to be minor for the case
assessed. While the approach presented provides a step towards effi­
cient computation of BEP progression at the field scale, much more
values of the Reynolds number computed when the pipe was at the research is needed to reliably predict BEP in practice. In particular,
critical pipe length (Fig. 10). For greater depths, the flow was no longer additional studies are needed to
laminar, and the analyses permitting turbulent flow yielded lower cri­
tical average gradients than the analysis conducted assuming solely (1) determine the appropriate value of the critical hydraulic gradient ic
laminar flow. However, the differences were relatively small (max used in the model to predict pipe progression
difference of 0.002 in terms of average gradient, or 0.5 m across the (2) assess the differences between two- and three- dimensional mod­
265 m seepage length) and likely not of significant importance in an eling approaches to determine if the depth correction factor pre­
engineering sense. This suggests that laminar flow assumptions are sented is reliable in all cases
adequate for the input parameter set considered in this study (see (3) refine the adaptive algorithm further (through higher order solu­
Fig. 11). tions or additional criteria for mesh adaptation or pipe progression)
The decreasing critical average gradient computed with increasing to eliminate numerical meandering of the pipe
values of d illustrated in Fig. 9a is consistent with previous studies of (4) validate predictions by modeling case histories and/or experiments
Sellmeijer (1988, 2006) and Schmertmann (2000). To compare the using independently predicted or measured input parameters
results of this study to the results of previous studies, the piping rule of
Sellmeijer et al. (2011) was compared to the results of the present The authors encourage research in these directions to accelerate
study. Sellmeijer et al. (2011) found that the critical average gradient development of tools that can be used in practice to reliably predict
was related to the aquifer depth and length by the expression BEP. Such tools will greatly improve the professions ability to manage
BEP risks in a cost-effective manner.
0.28 + 0.04
Hcr d (d / Lp)2.8 1
0.91 CRediT authorship contribution statement
Lp Lp (26)
B.A. Robbins: Conceptualization, Methodology, Software,
where d denotes the depth of the aquifer, and Lp denotes the final
Validation, Investigation. D.V. Griffiths: Conceptualization,
length of the pipe path (185 m). Because this equation was calibrated to
Methodology, Software, Validation, Supervision.
a series of experiments at different scales, it serves as a useful bench­
mark of the magnitude of the change in critical gradient due to changes
Declaration of Competing Interest
in d and Lp . As sufficient data is not available to validate the depth
correction using field case histories, comparing to Eq. (26) was, in the
authors’ view, the simplest means of evaluating the performance of the The authors declare that they have no known competing financial
FE depth correction proposed in this study. interests or personal relationships that could have appeared to influ­
The results of this study can be compared to the results of Sellmeijer ence the work reported in this paper.
et al. (2011) by comparing the change in the critical hydraulic gradient
(Hcr / L ) obtained for different aquifer depths to the trend predicted by Acknowledgements
Eq. (26). This comparison was made by normalizing the modeled cri­
tical gradients to the case d = 1 m such that Eq. (26) could be plotted This research was financially supported by the Flood and Coastal
normalized to Eq. (26) for d = 1. The results of this comparison are Systems R&D Program of the U.S. Army Corps of Engineers and the
shown in Fig. 11. The reduction in Hcr obtained with the depth cor­ Long-Term Training Program of the U.S. Army Engineer Research and
rection as described for the FE calculations in this study yields a similar Development Center. Permission to publish granted by the Director,
trend to that predicted by Eq. (26). While the numerical calculations Geotechnical and Structures Laboratory.
also yield a decreased value of Hcr with increasing values of d , the
magnitude of reduction obtained is not quite as large as predicted by References
Sellmeijer et al. (2011) for larger values of d . Nevertheless, for values of
d ranging from 1 to 20 m, the approximate depth correction method Baker, C., 2018. USACE conducts first analysis of risk and benefits of USACE levees. Flood
Risk Management Newsletter (Vol. 11). Washington, D.C.
presented in this study may be sufficiently accurate for analysis of BEP. Bangerth, W., Hartmann, R., Kanschat, G., 2007. deal.II–-A general-purpose object-

11
B.A. Robbins and D.V. Griffiths Computers and Geotechnics 129 (2021) 103820

oriented finite element library. ACM Trans. Math. Software 33 (4), 24-es. https://doi. Geotech. 18 (5), 266–279.
org/10.1145/1268776.1268779. Robbins, B.A., Van Beek, V.M., 2015. Backward erosion piping: A historical review and
Becker, C., Kurzeja, P., Steeb, H., 2010. Modelling Internal Erosion of Cohesionless Soils discussion of influential factors. In Association of State Dam Safety Officials, Dam
Using a Microstructural Parameter. PAMM 10 (1), 355–356. Safety, pp. 2015.
Briaud, J.L., Govindasamy, A.V., Shafii, I., 2017. Erosion charts for selected geomaterials. Rotunno, A.F., Callari, C., Froiio, F., 2019. A finite element method for localized erosion
J. Geotech. Geoenviron. Eng. 143 (10), 04017072. in porous media with applications to backward piping in levees. Int. J. Numer. Anal.
Burkley, V.J., Bruch, J.C.J., 1991. Adaptive error analysis in seepage problems. Int. J. Meth. Geomech. 43 (1), 293–316. https://doi.org/10.1002/nag.2864.
Numer. Meth. Eng. 31, 1333–1356. Sari, H., Chareyre, B., Catalano, E., Philippe, P., Vincens, E., 2011. Investigation of
Callari, C., Froiio, F., 2020. A hydro-mechanical finite element formulation for localized Internal Erosion Processes Using a Coupled DEM-Fluid Method. In: Oate, E., Owen, D.
internal erosion in porous media, with application to backward piping in cofferdams. R.J. (Eds.), Particles 2011 II International Conference on Particle-Based Methods,
Int. J. Multiscale Comput. Eng. 18 (2), 181–197. https://doi.org/10.1615/ Barcelona, 1–11.
IntJMultCompEng.2020031422. Saucier, R.T., 1994. Geomorphology and Quarternary Geologic History of the Lower
Cividini, A., Gioda, G., 2004. Finite-Element Approach to the Erosion and Transport of Mississippi Valley. Volume 1. U.S. Army Corps of Engineers, Waterways Experiment
Fine Particles in Granular Soils. Int. J. Geomech. 4 (3), 191–198. Station, Vicksburg, MS.
de Wit, J.M., 1984. Research report on sand boil model tests. Delft, Netherlands. Schmertmann, J.H., 2000. The No-Filter Factor of Safety Against Piping Through Sands.
El Shamy, U., Aydin, F., 2008. Multiscale Modeling of Flood-Induced Piping in River In: Silva, F., Kavazanijian, E.J. (Eds.), Judgement and Innovation. American Society
Levees. J. Geotech. Geoenviron. Eng. 134 (9), 1385–1398. of Civil Engineers, pp. 65–133.
Fascetti, A., Oskay, C., 2019. Dual random lattice modeling of backward erosion piping. Sellmeijer, J.B., 1988. On the mechanism of piping under impervious structures. Delft
Comput. Geotech., 105, May 2018, 265–276. https://doi.org/10.1016/j.compgeo. University of Technology.
2018.08.018. Sellmeijer, J.B., 2006. Numerical computation of seepage erosion below dams (piping).
Foster, M., Fell, R., Spannagle, M., 2000. The statistics of embankment dam failures and International Conference on Scour and Erosion.
accidents. Retrieved from. Can. Geotech. J. 37 (10), 1000–1024. http://www. Sellmeijer, H., López, J., Cruz, D., van Beek, V.M., Knoeff, H., 2011. Fine-tuning of the
nrcresearchpress.com/doi/abs/10.1139/t00-030. backward erosion piping model through small-scale, medium-scale and IJkdijk ex­
Froiio, F., Callari, C., Rotunno, A., 2019. A numerical experiment of backward erosion periments. Eur. J. Environ. Civil Eng. 15 (8), 1139–1154. https://doi.org/10.3166/
piping: kinematics and micromechanics. Meccanica 54 (14), 2099–2117. https://doi. EJECE.15.1139-1154.
org/10.1007/s11012-019-01071-7. Sheu, T.W.H., Fang, C.C., Tsai, S.F., 1999. Application of an element-by-element
Fujisawa, K., Murakami, A., Nishimura, S.-I., 2010. Numerical Analysis of the Erosion and BiCGSTAB iterative solver to a monotonic finite element model. Comput. Math. Appl.
the Transport of Fine Particles within Soils Leading to the Piping Phenomenon. Soils 37 (3), 57–70. https://doi.org/10.1016/S0898-1221(99)00046-2.
Found. 50 (4), 471–482. Smith, I.M., Griffiths, D.V., 2004. Programming the finite element method, 4th ed. Wiley.
Hagerty, D.J., Curini, A., 2004. Impoundment Failure Seepage Analyses. Environ. Eng. Stavropoulou, M., Papanastasiou, P., Vardoulakis, I., 1998. Coupled Wellbore Erosion and
Geosci. 10 (1), 57–68. Stability Analysis. Int. J. Numer. Anal. Meth. Geomech. 22 (9), 749–769.
Hanses, U., 1985. The mechanics of the development of erosion pipes in a layered sub­ Steeb, H., Diebels, S., Vardoulakis, I., 2007. Modeling Internal Erosion in Porous Media.
stratum beneath dams. Technischen Universitat Berlin. In: Geo-Denver 2007: New Peaks in Geotechics, 1–10. Dever, CO.
Harr, M., 1962. Groundwater and seepage. McGraw-Hill Inc. Tran, D., Prime, N., Froiio, F., Callari, C., Vincens, E., 2016. Numerical modelling of
Jianhua, Y., 1998. FE Modeling of Seepage in Embankment Soils with Piping Zone. Chin. backward front propagation in piping erosion by DEM-LBM coupling. Eur. J. Environ.
J. Rock Mech. Eng. 17 (6), 679–686. Civil Eng. 1–28.
Liang, Y., Chen, J., 2011. Mathematical Model for Piping Erosion Based on Fluid-Solid USBR, & USACE. 2015. Best Practices in Risk Assessment for Dams and Levees.
Interaction and Soils Structure. ASCE Spec. Publ. 217, 109–116. Denver, CO.
Liang, Y., Yeh, T.C.J., Wang, J., Liu, M., Zha, Y., Hao, Y., 2017. An auto-adaptive moving Van Beek, V.M., 2015. Backward Erosion Piping: Initiation and Progression. Technische
mesh method for the numerical simulation of piping erosion. Comput. Geotech. 82, Universiteit Delft. https://doi.org/10.1007/s13398-014-0173-7.2.
237–248. https://doi.org/10.1016/j.compgeo.2016.10.011. Van Beek, V.M., Robbins, B.A., Hoffmans, G.J.C.M., Bezuijen, A., van Rijn, L.C., 2019. Use
Lominé, F., Scholtès, L., Sibille, L., Poullain, P., 2013. Modeling of Fluid – Solid of incipient motion data for backward erosion piping models. Int. J. Sedim. Res. 34
Interaction in Granular Media with Coupled Lattice Boltzmann/Discrete Element (5), 401–408. https://doi.org/10.1016/j.ijsrc.2019.03.001.
Methods: Application to Piping Erosion. Int. J. Numer. Analyt. Methods Geomech. Van Beek, V.M., Yao, Q.L., Van, M.A., 2013. Backward Erosion Piping Model Verification
37(December 2011): 577–596. http://doi.org/10.1002/nag. Using Cases in China and the Netherlands. In: Seventh International Conference on
Luo, Y.L., 2013. A Continuum Fluid-Particle Coupled Piping Model Based on Solute Case Histories in Geotechnical Engineering, Chicago, IL. pp. 1–7.
Transport. Int. J. Civil Eng. 11 (1), 1–7. Van Esch, J.M., Sellmeijer, J.B., Stolle, D., 2013. Modeling Transient Groundwater Flow
Maeda, K., Muir Wood, D., Nukudani, E., 2010. Modelling Mechanical Consequences of and Piping under Dikes and Dams. In: 3rd International Symposium on
Erosion. Géotechnique 60 (6), 447–457. https://doi.org/10.1680/geot.2010.60.6. Computational Geomechanics (ComGeo III), 9.
447. Vandenboer, K., 2019. A Study on the Mechanisms of Backward Erosion Piping. Ghent
Rheinboldt, W.C., Mesztenyi, C.K., 1980. On a Data Structure for Adaptive Finite Element University.
Mesh Refinements. ACM Trans. Mathem. Software (TOMS) 6 (2), 166–187. https:// Vandenboer, K., van Beek, V., Bezuijen, A., 2013. 3D FEM Simulation of Groundwater
doi.org/10.1145/355887.355891. Flow During Backward Erosion Piping. In: Fifth International Young Geotechnical
Richards, K.S., Reddy, K.R., 2007. Critical appraisal of piping phenomena in earth dams. Engineering Conference, pp. 5–8.
Bull. Eng. Geol. Environ. 66 (4), 381–402. https://doi.org/10.1007/s10064-007- Wang, D.-Y., Fu, X.-D., Jie, Y.-X., Dong, W.-J., Hu, D., 2014. Simulation of Pipe
0095-0. Progression in a Levee Foundation with Coupled Seepage and Pipe Flow Domains.
Rochim, A., Marot, D., Sibille, L., Thao Le, V., 2017. Effects of Hydraulic Loading History Soils Found. 54 (5), 974–984.
on Suffusion Susceptibility of Cohesionless Soils. J. Geotech. Geoenviron. Eng. 143 Wang, Y., Ni, X., 2013. Hydro-Mechanical Analysis of Piping Erosion Based on Similarity
(7), 04017025. https://doi.org/10.1061/(ASCE)GT.1943-5606.0001673. Criterion at Micro-Level by PFC3D. Eur. J. Environ. Civil Eng. 17 (sup1), s187–s204.
Romeo, E., Royo, C., Monzón, A., 2002. Improved explicit equations for estimation of the https://doi.org/10.1080/19648189.2013.834594.
friction factor in rough and smooth pipes. Chem. Eng. J. 86 (3), 369–374. Zhou, X.-J., Jie, Y.-X., Li, G.-X., 2012. Numerical Simulation of the Developing Course of
Robbins, B.A., 2016. Numerical modeling of backward erosion piping. In: Proceedings of Piping. Comput. Geotech. 44 (June), 104–108. https://doi.org/10.1016/j.compgeo.
the 4th Itasca Symposium on Applied Numerical Modeling, pp. 551–558. 2012.03.010.
Robbins, B.A., Griffiths, D.V., 2018. Modelling of Backward Erosion Piping in Two- and Zhang, X.S., Wong, H., Leo, C.J., Bui, T.-A., Wang, J.X., Sun, W.H., Huang, Z.Q., 2012. A
Three- Dimensional Domains. In: Bonelli, S., Jommi, C., Sterpi, D. (Eds.), Internal Thermodynamics-Based Model on the Internal Erosion of Earth Structures. Geotech.
Erosion in Earth Dams, Dikes, and Levees. Cham, Switzerland: Springer Nature Geol. Eng. 31 (2), 479 492. https://doi.org/10.1007/s10706-012-9600-8.
Switzerland AG, pp. 149–158. https://doi.org/10.1007/978-3-319-99423-9. Zienkiewicz, O.C., Zhu, J.Z., 1987. A simple error estimator and adaptive procedure for
Robbins, B.A., Griffiths, D.V., 2020. SODA: A serial Fortran library for adaptation of practical engineerng analysis. Int. J. Numer. Meth. Eng. 24 (2), 337–357. https://doi.
structured meshes. Proceedings of 16th IACMAG. July 1-4, Torino, Italy. (In review). org/10.1002/nme.1620240206.
Robbins, B.A., Stephens, I.J., Leavell, D.A., López-Soto, J.F., Montalvo-Bartolomei, A.M., Zou, Y.-H., Chen, Q., Chen, X.-Q., Cui, P., 2013. Discrete Numerical Modeling of Particle
2018a. Laboratory piping tests on fine gravel. Can. Geotech. J. 55 (11), 1552–1563. Transport in Granular Filters. Comput. Geotech. 47 (January), 48–56. https://doi.
Robbins, B.A., van Beek, V.M., Lopez-Soto, J., Montalvo-Bartolomei, A.M., Murphy, J., org/10.1016/j.compgeo.2012.06.002.
2018b. A novel laboratory test for backward erosion piping. Int. J. Phys. Modell.

12