See discussions, stats, and author profiles for this publication at: https://www.researchgate.

net/publication/355034614

NSPFEM2D: A lightweight 2D node-based smoothed particle finite element

method code for modeling large deformation

Article  in  Computers and Geotechnics · October 2021

DOI: 10.1016/j.compgeo.2021.104484

CITATIONS READS

0 136

2 authors:

Ning Guo Zhongxuan Yang

Zhejiang University Zhejiang University
34 PUBLICATIONS   1,318 CITATIONS    112 PUBLICATIONS   1,909 CITATIONS   

SEE PROFILE SEE PROFILE

Some of the authors of this publication are also working on these related projects:

Modeling large deformation and discontinuity View project

TFEL/MFront View project

All content following this page was uploaded by Ning Guo on 03 October 2021.

The user has requested enhancement of the downloaded file.

 Contents lists available at ScienceDirect

Computers and Geotechnics

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

Research paper

NSPFEM2D: A lightweight 2D node-based smoothed particle finite element

method code for modeling large deformation
N. Guo ∗, Z.X. Yang
Computing Center for Geotechnical Engineering, Engineering Research Center of Urban Underground Space Development of Zhejiang Province, Department of Civil
Engineering, Zhejiang University, Hangzhou, Zhejiang 310058, China

ARTICLE INFO ABSTRACT

Keywords: The study presents a lightweight code that implements 2D node-based smoothed particle finite element
Node-based smoothed particle finite element method (NSPFEM) to fulfill the requirement of modeling large deformation in a variety of research fields and
method (NSPFEM) engineering applications. NSPFEM discretizes the problem domain with a set of Lagrangian nodes and employs
Large deformation
Delaunay triangulation to build/re-build mesh on the nodes at each time step. It is therefore able to tackle
Multiscale modeling
large deformation without suffering from mesh entanglement. The method also adopts the nodal integration
Retrogressive slope failure
Anchor pullout
technique to avoid interpolation of material variables after remeshing which can eliminate interpolation-
Trapdoor induced errors. The code has been built using hybrid C++ and Python programming, and is thus efficient
and versatile to couple with other well-developed codes, such as MFront for the incorporation of various
constitutive models and YADE, a discrete element method code, for multiscale simulations. The implementation
of NSPFEM has first been validated using two examples, i.e., axial vibration of an elastic cantilever beam and
a rolling cylinder down incline. Next, the capability and flexibility of the code has been demonstrated by
three examples, namely retrogressive slope failure, pullout of a strip plate anchor, and multiscale modeling of
trapdoor.

1 1. Introduction spatially fixed mesh (Benson, 1992). The backbone of the methods is 22
an operator-split strategy which comprises two calculation phases: a 23
2 Many research topics and engineering practices involve dealing with Lagrangian phase to update the configuration of solid by solving the 24
3 large deformations of materials. For example, the metal cutting process balance of momentum equation, followed by a convection phase to map 25
4 in machining will induce large deformation in the workpieces (Ro- field variables from the deformed frame to the fixed Eulerian mesh. One 26
5 dríguez et al., 2017). Specifically in geotechnical engineering, large example of such methods is the coupled Eulerian–Lagrangian (CEL), 27
6 deformations of geomaterials take place in most natural disasters, available as a built-in module in the commercial FEM software Abaqus 28
7 e.g., landslides (Chen et al., 2021) and snow avalanches (Li et al., (Dassault Systèmes, 2016). Because of its ready availability, CEL has 29
8 2020), as well as in construction processes, e.g., soil excavation (Peng long been used in offshore geotechnical engineering to study the capac- 30
9 et al., 2017), pipeline burying (Li et al., 2021) and pile/caisson in- ity of spudcan foundations and anchoring plates (Qiu et al., 2011; Qiu 31
10 stallation (Yang et al., 2020; Jin et al., 2019). Numerical modeling and Grabe, 2012; Chen et al., 2013). The tracking of boundary surface 32
11 is indispensable to both advance our understanding of the physics of
or multi-material interfaces in CEL, however, is cumbersome usually 33
12 these phenomena and facilitate the economic and safe design in various
having to resort to the volume fraction of each element. Consequently, 34
13 engineering sectors. Meanwhile, it also faces enormous challenges as
the application of pressure and non-zero displacement boundary con- 35
14 traditional mesh-based numerical methods such as the finite element
ditions in CEL must be achieved indirectly. In fact to avoid mesh 36
15 method (FEM) will easily suffer from mesh entanglement at large
distortion, the mesh used in the convection phase may not necessarily 37
16 deformations.
be fixed in the background but can be optimized after the Lagrangian 38
17 In the past few decades, several numerical techniques have been
18 developed to address the aforementioned mesh-related challenges. The phase with boundaries/interfaces properly conserved, which leads to 39
19 Eulerian methods, which are the mainstream in computational fluid the arbitrary Lagrangian–Eulerian (ALE) method. In ALE, the mesh can 40
20 dynamics, have been applied to solve large deformations of solids be totally regenerated, e.g., the remeshing and interpolation technique 41
21 by treating the solid material as a fluid that ‘‘flows’’ through the by small strain (RITSS) (Hu and Randolph, 1998) or refined using 42

∗ Corresponding author.
E-mail address: [email protected] (N. Guo).

https://doi.org/10.1016/j.compgeo.2021.104484
Received 6 August 2021; Received in revised form 20 September 2021; Accepted 21 September 2021
 N. Guo and Z.X. Yang

1 the 𝑟-adaptive scheme, e.g., the efficient ALE (EALE) (Nazem et al., the method less vulnerable to volumetric locking (Liu et al., 2009) and 67
2 2006). Detailed comparisons of CEL, RITSS, and EALE in terms of mesh sensitivity (Guo et al., 2021b). 68
3 their remeshing and mapping strategies, computational efficiencies, Regarding the approachability of the methods, CEL, ALE and SPH 69
4 and user-friendliness can be found in the review article (Wang et al., are all available in the well-recognized commercial software Abaqus 70
5 2015a). (Dassault Systèmes, 2016) and LS-DYNA (Hallquist, 2006), whereas 71
6 Another category of numerical methods to solve large deformation MPM is under active development by a few open-source projects, 72
7 problems is the particle-based methods, among which the smoothed e.g., NairnMPM (http://osupdocs.forestry.oregonstate.edu/index.php/ 73
8 particle hydrodynamics (SPH) (Gingold and Monaghan, 1977; Lucy, NairnMPM), Uintah (http://uintah.utah.edu) and Cb-Geo (https:// 74
9 1977), the material point method (MPM) (Sulsky et al., 1994), and the github.com/cb-geo/mpm). Besides, PFEM is also contained in the open- 75
10 particle finite element method (PFEM) (Idelsohn et al., 2004; Oñate source software Kratos (https://www.cimne.com/kratos) and can be 76
11 et al., 2004) are most representative. SPH was initially developed for easily implemented in Abaqus with additional meshing and map- 77
12 astrophysical problems (Gingold and Monaghan, 1977; Lucy, 1977) and ping functions, as demonstrated by Li and Zhang (2020) and Yuan 78
13 has been popular in solving solid mechanics problems since Libersky et al. (2021). On the contrary, NSPFEM is currently developed in- 79
14 and Petschek (1991). SPH is a pure meshless method, in which field house only and no NSPFEM code is publicly accessible yet. Due to 80
15 variables are approximated by the kernel function instead of the shape the use of a smoothed displacement–deformation matrix in NSPFEM, 81
16 function as in FEM. Examples of SPH application in large-deformation its implementation in existing FEM codes is not as straightforward 82
17 geomechanical problems can be found in Bui and Fukagawa (2013) as PFEM. The objective of the present study is therefore to offer a 83
18 and Peng et al. (2015), as well as multiphysics problems in Bai et al. lightweight implementation of NSPFEM based on a number of open- 84
19 (2018). In contrast to SPH, MPM combines the strengths of both source libraries to make the method more approachable to interested 85
20 particle- and mesh-based methods by discretizing the continuum with a researchers. The code NSPFEM2D adopts hybrid C++ and Python 86
21 set of Lagrangian particles and meantime retaining the Eulerian mesh to programming and is hence efficient and versatile to couple with other 87
22 solve the governing momentum equation. In light of this, MPM has been well-developed codes, such as MFront (Helfer et al., 2015) for using 88
23 gaining favor with the communities of geomechanics (Soga et al., 2016; various constitutive models and YADE (Šmilauer et al., 2015), a discrete 89
24 Li et al., 2021) and computer graphics (Wang et al., 2020), as well as element method (DEM) code, for multiscale simulations. Despite the 90
25 the film industry (Stomakhin et al., 2013). Despite notable successes primary goal aforementioned, other novelties of the study include a 91
26 of SPH and MPM in modeling large deformation problems, the two predictor–corrector algorithm first introduced to NSPFEM for modeling 92
frictional contact that is deemed more efficient than the penalty-based 93
27 methods have their own respective shortcomings. For example, the SPH
algorithm for the explicit scheme, and the trapdoor problem studied 94
28 rEsearch and engineeRing International Community (SPHERIC) has
using a coupled NSPFEM/DEM approach where the soil arching effect 95
29 recently identified 5 grand challenges for SPH to address including sta-
has been analyzed with enriched multiscale perspectives. 96
30 bility issues and enforcement of boundary conditions (Vacondio et al.,
The rest of the paper is structured as follows. The fundamentals 97
31 2021). MPM, on the other hand, has been criticized for its inaccuracy,
of NSPFEM are first introduced in Section 2, followed by the im- 98
32 particularly stress-related oscillation and inaccuracy, compared with
plementation details in Section 3. Then, two validation tests solving 99
33 FEM solutions (González Acosta et al., 2020). In these regards, PFEM
axial vibration of an elastic cantilever beam and a rolling cylinder 100
34 emerges to be a worthy alternative to the other methods, as it is closely
down incline are presented in Section 4, before three demonstrative 101
35 related to FEM and inherits many merits from the latter.
applications of NSPFEM modeling retrogressive slope failure, pullout 102
36 PFEM was originally proposed by Oñate and coworkers for fluid
of a strip plate anchor, and multiscale trapdoor in Section 5. The study 103
37 dynamics and fluid–structure interaction problems (Oñate et al., 2004;
is finally concluded in Section 6. 104
38 Idelsohn et al., 2004, 2006), and later finds a broad application in large-
39 deformation solid mechanics (Dávalos et al., 2015; Rodríguez et al., 2. Approach and formulation 105
40 2017; Zhang et al., 2013) and multiphysics problems (Aubry et al.,
41 2005; Rodriguez et al., 2016). A state of the art review of the method NSPFEM is an integration of PFEM (Oñate et al., 2004) and NSF 106
42 is provided in Cremonesi et al. (2020). The basic idea of PFEM is EM (Liu et al., 2009). It hence inherits the merits of both methods. 107
43 that it employs a cloud of nodes (or called particles) to discretize the For example, the remeshing operation from PFEM enables NSPFEM to 108
44 continuum body and applies Delaunay triangulation for the efficient model large deformations. Meanwhile, the nodal integration technique 109
45 generation of mesh on these nodes. The governing equation is then from NSFEM allows the material physical variables to be stored at 110
46 solved following standard FEM using the mesh, and the nodes will the nodes so that interpolation of the variables after remeshing is not 111
47 move according to the FEM solution of nodal displacements. Next, required in NSPFEM, which eliminates errors that could be introduced 112
48 a new mesh will be created on the nodes with updated positions by the interpolation process. Since NSPFEM can be regarded as a 113
49 and the field variables stored at the Gauss quadrature points of the variant of FEM and most FEM formulations still apply in NSPFEM, the 114
50 old mesh need to be mapped to the new mesh. PFEM is hence still latter will be introduced first. 115
51 largely Lagrangian FEM in which mesh distortion at large deformations
52 is averted by frequent remeshing. Recently, a node-based smoothed 2.1. Governing equation and FEM discretization 116
53 version of PFEM (called NSPFEM) (Zhang et al., 2018a; Yuan et al.,
54 2019; Franci et al., 2020; Jin et al., 2020; Yuan et al., 2020; Franci, The governing equation for a mechanical problem is given by the 117
55 2021; Guo et al., 2021b; Zhang et al., 2021) was developed aiming to balance of momentum equation: 118
56 skip the mapping operation in PFEM, which is known to be error-prone,
𝜎𝑖𝑗,𝑗 + 𝑏𝑖 = 𝜌𝑢̈ 𝑖 (1) 119
57 by adopting the nodal integration technique borrowed from the node-
58 based smoothed FEM (NSFEM) and meshfree methods (Liu et al., 2009). where 𝜎𝑖𝑗 is the stress tensor; 𝑏𝑖 is the body force per unit volume; 𝜌 120
59 In NSPFEM, material properties are stored at the nodes instead of the is the bulk density of the material; 𝑢𝑖 , 𝑢̇ 𝑖 and 𝑢̈ 𝑖 are the displacement, 121
60 Gauss points, thus any update of the material variables (e.g., stress the velocity, and the acceleration, respectively. The variational form of 122
61 and strain) is performed at the nodes. It is worth noting that NSPFEM the above equation can be obtained by multiplying a test function 𝑣𝑖 on 123
62 possesses several advantages compared with the other methods men- both sides of Eq. (1) and integrating over the problem domain 𝛺. After 124
63 tioned above. It is superior to SPH and MPM in terms of accuracy, applying the divergence theorem, one can reach the following integral 125
64 convergency, and imposition of essential boundary condition, thanks form 126
65 to the inheritance from FEM. In addition, the smoothing operation
66 enables NSPFEM to eliminate interpolation-induced errors and makes 𝜌𝑢̈ 𝑖 𝑣𝑖 𝑑𝛺 = 𝑡𝑖 𝑣𝑖 𝑑𝛤 + 𝑏𝑖 𝑣𝑖 𝑑𝛺 − 𝜎𝑖𝑗 𝑣𝑖,𝑗 𝑑𝛺 (2) 127
∫𝛺 ∫𝛤𝑡 ∫𝛺 ∫𝛺
 N. Guo and Z.X. Yang

Fig. 1. Illustration of the smoothing area around a node.

∑
1 where 𝑡𝑖 is the traction force applied on the Newmann boundary 𝛤𝑡 . ̃𝑗 = 1
𝐁 𝐴𝑘 𝐁𝑘 (10) 41
2 In the context of FEM, the unknowns are discretized into nodal 𝐴̃ 𝑗 𝑘∈sur.tri.
3 values and interpolated by a shape function 𝐍 in the problem domain. where the summation is over the surrounding triangle elements sharing 42
4 The discrete form can be obtained from Eq. (2) the node 𝑗; 𝐴𝑘 and 𝐁𝑘 are the weighting area and the displacement– 43
5 𝐌𝐮̈ = 𝐟 ext − 𝐟 int (3) deformation matrix of the triangle 𝑘, respectively. It is noted that 𝐴𝑘 44
equals to 1/3 of the area of the triangle 𝑘. 45
6 where The strain at the node 𝑗 is evaluated as 46
7 𝐌 = 𝜌𝐍𝑇 𝐍 𝑑𝛺 (4) ̃ 𝑗𝐮
𝜺𝑗 = 𝐁 (11) 47
∫𝛺

8 𝐟 ext = 𝐍𝑇 𝐭 𝑑𝛤 + 𝐍𝑇 𝐛 𝑑𝛺 (5) And the internal force vector given by Eq. (6) will be reformulated as 48
∫𝛤𝑡 ∫𝛺
𝑁
∑𝑗
9 𝐟 int = 𝐁𝑇 𝝈 𝑑𝛺 (6) 𝐟 int = ̃ 𝑇 𝝈 𝑑𝛺 =
𝐁 ̃ 𝑇 𝝈 𝑗 𝐴̃ 𝑗
𝐁 𝑗 (12) 49
∫𝛺 ∫𝛺 𝑗
10 where 𝐁 is the displacement–deformation matrix. The consistent mass where 𝑁𝑗 is the number of nodes in the domain. 50
11 matrix 𝐌 calculated by Eq. (4) is usually replaced with the diagonalized
12 one using the HRZ-lumping (Hinton et al., 1976) so its inverse can 2.3. Frictional contact 51
13 be easily computed. In NSPFEM, the initial nodal mass is obtained by
14 multiplying the nodal area by the material density and keeps constant The contact is treated using a predictor–corrector algorithm similar 52
15 afterwards so that the mass conservation of the system is automatically
to that in the material point method (Bardenhagen et al., 2000) and the 53
16 satisfied.
smoothed particle hydrodynamics method (Wang and Chan, 2014). The 54
17 For the time integration of Eq. (3), the velocity Verlet scheme
nodal velocity is first calculated without considering any contact and 55
18 (Swope et al., 1982) is applied which has been commonly used in
will be regularized after the nodal displacement is updated. A contact 56
19 simulation of dynamic systems and possesses second-order accuracy:
is activated when a node and a boundary segment (assumed to be rigid 57
1
20 𝐮̇ 𝑛+1 = 𝐮̇ 𝑛 + 𝛥𝑡(̈𝐮𝑛 + 𝐮̈ 𝑛+1 ) (7) and either stationary or moving with a prescribed velocity in the study 58
2 for simplicity) have non-zero penetration. The normal contact force 𝐟𝑛𝑐 59
1
21 𝐮𝑛+1 = 𝐮𝑛 + 𝛥𝑡𝐮̇ 𝑛 + 𝛥𝑡2 𝐮̈ 𝑛 (8) is assumed to be able to completely suppress the normal penetrating 60
2
velocity of the node during the time step such that 61
22 where the superscripts 𝑛 and 𝑛+1 denote the variable at two consecutive
23 time steps; 𝛥𝑡 is the time step which must be chosen sufficiently small 𝑚 × 𝑢̇ 𝑛 𝑐
𝐟𝑛𝑐 = − 𝐧 (13) 62
24 to satisfy the Courant–Friedrichs–Lewy (CFL) condition (Courant et al., 𝛥𝑡
25 1928). 𝑐
where 𝑚 is the nodal mass; 𝑢̇ 𝑛 (= 𝐮̇ 𝑟 ⋅ 𝐧 ) is the normal projection of 63
the relative velocity between the node and the segment 𝐮̇ 𝑟 and must be 64
26 2.2. Smoothed particle FEM negative at a contact; 𝐧𝑐 is the normal vector of the segment. Similarly, 65
the tangential contact force 𝐟𝑡𝑐 is calculated by first assuming a sticking 66
27 In line with PFEM, NSPFEM discretizes the problem domain with a condition such that 67
28 set of Lagrangian nodes and constructs the mesh on the nodes by Delau-
𝑚 × 𝑢̇ 𝑡 𝑐
29 nay triangulation. The nodes will move according to the deformation 𝐟𝑡𝑐 = − 𝐭 (14) 68
𝛥𝑡
30 and displacement of the material following the Lagrangian description.
𝑐
where 𝑢̇ 𝑡 (= 𝐮̇ 𝑟 ⋅ 𝐭 ) is the tangential projection of the relative velocity 69
31 As a result, remeshing will be invoked to tackle large deformations.
32 Since the mapping of variables on the Gauss points from the old mesh to between the node and the segment; 𝐭 𝑐 is the tangential vector of the 70
33 the new one will produce extra errors, to avoid this operation, NSPFEM segment. The tangential force is also subjected to the Coulomb friction, 71
34 adopts the nodal integration technique borrowed from NSFEM and which poses a limit |𝐟𝑡𝑐 | ≤ 𝜇|𝐟𝑛𝑐 | on the magnitude of the tangential force 72
35 meshfree methods (Puso et al., 2008), where material stress updates are where 𝜇 is the frictional coefficient. 73
36 performed at the nodes. The smoothing area of a node 𝑗 is illustrated in Consequently, the relative velocity at a contact is modified by the 74
37 Fig. 1, which is constructed by connecting the centroids of triangles and additional contact force according to 75
38 the midpoints of edges around the node. Its area 𝐴̃ 𝑗 and the smoothed
̃ 𝑗 are calculated as 𝐮̇ 𝑟 ← 𝐮̇ 𝑟 − 𝑢̇ 𝑛 𝐧𝑐 − min(|𝑢̇ 𝑡 |, −𝜇 𝑢̇ 𝑛 ) sign(𝑢̇ 𝑡 ) 𝐭 𝑐 (15) 76
39 displacement–deformation matrix 𝐁
∑ The rationality behind the regularization method can be understood 77
40 𝐴̃ 𝑗 = 𝐴𝑘 (9)
as follows: when the tentative sliding velocity is small, the frictional 78
𝑘∈sur.tri.
 N. Guo and Z.X. Yang

Fig. 2. Mesh optimization by centroidal patch triangulation (CPT).

1 force can suppress the tendency completely (sticking condition); but 2016a,b; Guo et al., 2021a) where the material constitutive behaviors 43
2 when the node is already sliding (|𝑢̇ 𝑡 | is large), the offset provided are captured by microscopic models, e.g., DEM. In the conventional 44
3 by the frictional force is −𝜇 𝑢̇ 𝑛 sign(𝑢̇ 𝑡 ) (sliding condition). It is also approach, an open-source code generator MFront (Helfer et al., 2015) 45
4 noteworthy that the current predictor–corrector method involves no is employed for the convenient implementation of various constitutive 46
5 additional parameters and has no influence on the time step size, which models based on a high-level domain specific language (DSL). MFront 47
6 is in contrast to the penalty method where the choice of additional is able to generate dynamic libraries which can then be interfaced 48
7 penalty parameters requires special carefulness and could lead to a with other codes written in either Fortran, C++ or Python through 49
8 small time step since the contact stiffness must be sufficiently large to MFrontGenericInterfaceSupport (Helfer et al., 2020). In the 50
9 guarantee impenetrability at the contact. multiscale approach, the NSPFEM implementation will be coupled with 51
a popular open-source DEM package YADE (Šmilauer et al., 2015) 52
10 2.4. Summary of solution procedure via their respective Python modules. Then, the material constitutive 53
responses are obtained from DEM solution of representative volume 54
11 The solution procedure of NSPFEM is summarized below: elements (RVEs) embedded in the nodes of NSPFEM. For the detailed 55
algorithms in coupling NSPFEM and DEM and the applications of 56
12 (i) Discretize the problem domain with a set of nodes;
the multiscale approach, interested readers are referred to Guo et al. 57
13 (ii) Construct the mesh on the nodes by applying Delaunay triangu-
(2021b). 58
14 lation;
Last, step (vi) is necessary when the mesh quality deteriorates 59
15 (iii) Use alpha shapes (Edelsbrunner and Mücke, 1994) to remove
significantly due to severely non-uniform distribution of nodes at large 60
16 non-physical elements in the domain;
deformations (see Fig. 2). To improve the mesh quality as well as the 61
17 (iv) Solve the nodal displacement and update the positions of nodes
result accuracy, mesh smoothing techniques can be applied. In the 62
18 accordingly;
current study, the method of centroidal patch triangulation (CPT) avail- 63
19 (v) Detect if there is any contact and regularize the velocity at the
able from optimesh (https://pypi.org/project/optimesh) has been 64
20 contact;
adopted, which can be viewed as the generalized Laplacian smooth- 65
21 (vi) Optimize the nodal positions when necessary (interpolation of
ing (Chen and Holst, 2011). After the redistribution of nodes, the 66
22 variables is required with this optional step);
variables associated with the nodes need to be interpolated from the old 67
23 (vii) Go back to step (ii) for the next loading step until completion of
triangulation to the new one. Notably, the mesh optimization process 68
24 all the steps.
is only needed when the mesh quality is severely deteriorated. Thus, 69
25 It is worth mentioning that the Delaunay triangulation in step (ii) the interpolation of variables is only called occasionally in NSPFEM 70
26 can be achieved very efficiently by using some off-the-shelf libraries, in contrast to that required at every time step in PFEM. Simple linear 71
27 for example, CGAL (Hert and Seel, 2020) has been used in the current interpolation function in SciPy (https://www.scipy.org), LinearND- 72
28 study. However, as Delaunay triangulation always results in a convex Interpolator, has been used in the study, which performs linear 73
29 domain, alpha shapes (Edelsbrunner and Mücke, 1994) are typically barycentric interpolation. Fig. 2 illustrates the smoothing of a triangu- 74
30 used in step (iii) to determine if an element belongs to the true domain lation by CPT and the linear interpolation of an arbitrary field variable. 75
31 and to remove those non-physical. Following Zhang et al. (2014), an The improvement in mesh quality is notable, i.e., the node spacing 76
32 element will be treated as non-physical and removed from the mesh if becomes much more uniform after CPT smoothing. Meanwhile, the 77
33 its circumradius is greater than 𝛼ℎ, where ℎ is the characteristic spacing spatial distribution character of the field variable has been reasonably 78
34 of the nodes, and 𝛼 is a user-defined parameter typically ranging maintained. Step (vi) will however not be enabled in the multiscale 79
35 between 1.2 and 1.6. 𝛼 = 1.5 is used exclusively in the paper unless NSPFEM/DEM approach since the interpolation of the microscopic 80
36 otherwise stated. RVEs after remeshing is generally infeasible. 81
37 It is also important to note that in step (iv) when Eq. (3) is solved, a
38 material constitutive model is required to complete the equation system 3. Code implementation 82
39 by updating the stress field of the problem domain. Two methods
40 to prescribe the material constitutive behaviors are explored in the With the aid of CGAL to build mesh based on Delaunay triangula- 83
41 study: the conventional way using a constitutive model, e.g., the von tion, the pivotal step in NSPFEM computation is to find the smoothed 84
42 Mises plastic model, and the multiscale approach (Guo and Zhao, 2014, displacement–deformation matrix 𝐁 ̃ 𝑗 for each node using Eq. (10), 85
 N. Guo and Z.X. Yang

1 which requires the knowledge of the conventional 𝐁𝑘 matrix for a 23 } 65

2 triangle element 𝑘 given by 24 } 66
25 for( size_t i=0; i<npts; i++) // tilde_B = 67
⎡ 𝑦 − 𝑦3 0 𝑦3 − 𝑦1 0 𝑦1 − 𝑦2 0 ⎤ sum(A_k*B_k)/ node_area 68
1 ⎢ 2
𝐁𝑘 = 0 𝑥3 − 𝑥2 0 𝑥1 − 𝑥3 0 𝑥2 − 𝑥1 ⎥ 26 bmatrices [i] /= (area[i] + 1.e -20); 69
6𝐴𝑘 ⎢ ⎥ } 70
⎣𝑥3 − 𝑥2 𝑦2 − 𝑦3 𝑥1 − 𝑥3 𝑦3 − 𝑦1 𝑥2 − 𝑥1 𝑦1 − 𝑦2 ⎦ 27
71
3 (16) ̃
Listing 1: C++ codes to construct mesh and find 𝐁
4 where 𝑥𝑖 and 𝑦𝑖 (𝑖 = 1, 2, 3) are the coordinates of the three vertices of
72
5 the triangle whose area equals 3𝐴𝑘 (see Fig. 1). 1 py :: list NSPFEM :: calcStrain ( const Eigen :: 73
6 This part has been implemented in C++ as it is computationally VectorXd & du) { 74
7 intensive. We use the library Eigen (http://eigen.tuxfamily.org) to 2 Eigen :: Vector3d eps; 75
8 deal with large sparse matrix operations and Boost (https://www. 3 py :: list eps_lst ; 76
9 boost.org) to expose the C++ class as Python extensions. The three 4 for( size_t i=0; i<npts; i++) { 77
10 functions, namely, updateMeshBmat to construct the mesh and find 5 eps = bmatrices [i] * du; 78
̃ 6 eps_lst . append (py :: make_tuple (eps [0], 79
11 the smoothed 𝐁-matrix (Eq. (10)), calcStrain to calculate the strain
eps [1] ,0. , eps [2]/ std :: sqrt (2))); // strain 80
12 (Eq. (11)), and calcFint to calculate the internal force (Eq. (12)) vector in MFront convention 81
13 are public members of a class NSPFEM, which has been compiled to 7 } 82
14 a shared library NSPFEMexp.so. The code snippet to triangulate the 8 return eps_lst ; 83
15 mesh and find 𝐁 ̃ is presented in Listing 1, and the codes implementing 9 } 84
85
16 Eqs. (11) and (12) to calculate 𝜺 and 𝐟 int are given in Listings 2 and
17 3, respectively. The dynamic library NSPFEMexp.so can then be Listing 2: C++ codes to find 𝜺
18 imported into Python (see Listing 4).
86
1 Eigen :: VectorXd NSPFEM :: calcFint ( const Eigen 87
19 :: MatrixXd & sig) { 88
20 1 py :: tuple NSPFEM :: updateMeshBmat ( const Eigen 2 Eigen :: VectorXd fint = Eigen :: VectorXd :: 89
21 :: MatrixXd & pts , const double charlen ) { Zero(npts *2); 90
22 2 // skipped codes for data processing 3 Eigen :: Vector3d stress ; 91
23 3 Delaunay dt( points . begin () ,points .end ()); 4 for( size_t i=0; i<npts; i++) { 92
24 // Delaunay triangulation 5 stress << sig(i ,0) ,sig(i ,1) ,sig(i ,3)/ 93
25 4 for( Finite_faces_iterator fit=dt. std :: sqrt (2); // conversion from MFront 94
26 finite_faces_begin ();fit != dt. stress vector 95
27 finite_faces_end ();fit ++) { // loop over 6 fint += bmatrices [i]. transpose () * 96
28 all triangles in dt stress * area[i]; 97
29 5 for( size_t i=0; i <3; i++) { // loop over 7 } 98
30 vertices of each triangle 8 return fint; 99
31 6 v[i] = fit -> vertex (i)->point (); // 9 } 100
101
32 vertex coordinates
33 7 ind[i] = fit -> vertex (i)->info (); // Listing 3: C++ codes to find 𝐟 int
34 vertex index
35 8 } The usage of NSPFEMexp.so library in Python has been facilitated 102
36 9 circumrad = CGAL :: sqrt(CGAL ::
by constructing a Python class, which wraps up all necessary func- 103
37 squared_radius (v[0] ,v[1] ,v[2])); //
38 triangle circumradius tions for a simulation. Any particular task from users (e.g. modeling 104
39 10 atri = CGAL :: area(v[0] ,v[1] ,v[2]); // retrogressive slope failure) can be conveniently built on top of the 105
40 triangle area class. Its implementation coupling with MFront using a conventional 106
41 11 if( circumrad < charlen && atri > 1.e constitutive law is partly shown in Listing 4. The coupling with YADE 107
42 -10) { // apply alpha shapes for multiscale modeling requires only minor modifications. 108
43 12 for( size_t i=0; i <3; i++) {
44 13 area[ind[i]] += atri / 3.; // 109
45 smoothed nodal area 1 from NSPFEMexp import NSPFEM # import compiled 110
46 14 for( size_t j=0; j <3; j++) { NSPFEMexp library 111
47 15 /* bmatrices is a vector of 2 import mgis. behaviour as mgis_bv # 112
48 Eigen sparse matrices . MFrontGenericInterfaceSupport 113
49 16 The following block finds sum( 3 114
50 A_k*B_k) for each node.*/ 4 constitutive = mgis_bv .load(’src/ libBehaviour 115
51 17 bmatrices [ind[i]]. coeffRef (0, .so ’,’MisesVocePlasticity ’,mgis_bv . 116
52 ind[j]*2) += (v[(j+1) %3][1] - v[(j+2) Hypothesis . PlaneStrain ) # assume plane 117
53 %3][1]) / 6.; strain 118
54 18 bmatrices [ind[i]]. coeffRef (1, 5 119
55 ind[j ]*2+1) += (v[(j+2) %3][0] - v[(j+1) 6 class nspfem_mfront ( object ): 120
56 %3][0]) / 6.; 7 def __init__ (self ,pts ,mater , sigma =None , 121
57 19 bmatrices [ind[i]]. coeffRef (2, gravity =[0. ,0.] , charlen =1.): 122
58 ind[j]*2) += (v[(j+2) %3][0] - v[(j+1) 8 """ 123
59 %3][0]) / 6.; 9 pts: type numpy ndarray , shape = (npts , 124
60 20 bmatrices [ind[i]]. coeffRef (2, 2) , points coordinates 125
61 ind[j ]*2+1) += (v[(j+1) %3][1] - v[(j+2) 10 mater : type structure , material 126
62 %3][1]) / 6.; properties 127
63 21 } 11 sigma : type numpy ndarray , initial 128
64 22 } stress , shape = (4 ,) or (npts , 4) 129
 N. Guo and Z.X. Yang

1 12 gravity : type vector , size = 2, gravity problem considering the first mode vibration can be obtained using 67
2 constant separation of variables and is given by (Bardenhagen, 2002) 68
3 13 charlen : type double , characteristic
4 spacing ̇ 𝑡) = 𝑢̇ 0 cos(𝜔1 𝑡) sin(𝛽1 𝑥)
𝑢(𝑥, (17) 69
5 14 """ 𝑢̇
𝑢(𝑥, 𝑡) = 0 sin(𝜔1 𝑡) sin(𝛽1 𝑥) (18) 70
6 15 self. __npts = len(pts) 𝜔1
7 16 self.__pts = pts
where 𝑢(𝑥,
̇ 𝑡) and 𝑢(𝑥, 𝑡) are the horizontal velocity and displacement, 71
8 17 self. __nspfem = NSPFEM ()
9 18 self. __materData = mgis_bv . respectively, at coordinate 𝑥 and time 𝑡; the first-mode eigenvalue and 72
10 MaterialDataManager ( constitutive , self. frequency are 𝛽1 = 𝜋∕(2𝐿) and 𝜔1√= 𝛽1 𝑐, respectively, where 𝐿 = 10 m 73
11 __npts ) is the length of the beam and 𝑐 = 𝐸∕𝜌 is the wave speed. 𝑢̇ 0 = 0.1 m/s 74
12 19 # skipped codes for material state is the velocity amplitude used to describe the initial velocity field 75
13 initialization , etc ̇ 0) = 𝑢̇ 0 sin(𝛽1 𝑥).
𝑢(𝑥, 76
14 20 For the numerical simulation, the beam is discretized using 306 77
15 21 def __updateMeshBmatrixFint (self): nodes with a characteristic spacing 0.2 m. The predicted velocity and 78
16 22 tri ,area , minAlt = self. __spfem . displacement at the right side of the beam are shown in Fig. 3 for one 79
17 updateMeshBmat ( MatrixX (self. __pts ),self. period 𝑇 = 2𝜋∕𝜔1 = 4 s. The numerical predictions using NSPFEM are 80
18 __charlen )
found generally agreeing well with the analytical solutions. 81
19 23 stress = self. __materData .s1.
20 thermodynamic_forces .copy ()
21 24 self. __fint = numpy . array (self. __spfem . 4.2. Rolling cylinder down incline 82
22 calcFint ( MatrixX ( stress )))
23 25 return .9 * minAlt / self. __wavespeed # This example validates the frictional contact algorithm presented 83
24 guarantee CFL condition in Section 2.3 by simulating an elastic cylinder (assumed plane strain) 84
25 26 with radius 1.6 m rolling down a slope inclined at an angle 𝜃 = 60◦ , 85
26 27 def __updateStrainStress (self ,du): as shown in Fig. 4(a). The Young’s modulus and the Poisson’s ratio of 86
27 28 dstrain = self. __spfem . calcStrain ( the cylinder are 𝐸 = 4.2 MPa and 𝜈 = 0.3, respectively. The density 87
28 VectorX (du)) is taken as 𝜌 = 1000 kg/m3 . The cylinder is discretized with 280 nodes 88
29 29 self. __materData .s1. gradients [: ,:] +=
(see Fig. 4(a)). The interface frictional coefficient between the slope and 89
30 numpy.array( dstrain )
31 30 mgis_bv . integrate (pool ,self. __materData the cylinder is 𝜇 = 0.3 or 0.9, corresponding to the slip and the stick 90
32 ,IT ,self.__dt) condition at the contact, respectively (see Eq. (19)). For an initially 91
33 31 mgis_bv . update (self. __materData ) stationary rigid cylinder, the analytical solution of its center-of-mass 92
34 32 displacement along the slope is given by (Bardenhagen et al., 2000) 93
35 33 def solve(self): #main function
⎧1 2
36 implementing velocity Verlet scheme ⎪ 𝑔𝑡 (sin 𝜃 − 𝜇 cos 𝜃), tan 𝜃 > 3𝜇 (slip)
37 34 self.__vel += .5 * self. __acc * self. 𝑢(𝑡) = ⎨ 2 (19) 94
38 __dt ⎪ 1 𝑔𝑡2 sin 𝜃, tan 𝜃 ≤ 3𝜇 (stick)
⎩3
39 35 du = self. __vel * self.__dt
40 36 self. __disp += du where 𝑔 = 10 m/s2 is the gravitational constant. The results are plotted 95
41 37 self.__pts += du. reshape ( -1 ,2) in Fig. 4(b), from which it is seen that the numerical predictions are 96
42 38 self. __updateStrainStress (du) # update slightly larger than the analytical solutions. Nevertheless, the NSPFEM 97
43 strain first (USF) results are reasonably acceptable. 98
44 39 dt = self. __updateMeshBmatrixFint ()
45 40 self.__acc = numpy . divide (self. __fext + 5. Demonstrative applications 99
46 self. __fbody - self.__fint , self. __mass )
47 41 self.__acc = self. __acc * numpy .
48 logical_not (self. __dmsk ) # comply with 5.1. Retrogressive slope failure 100
49 boundary condition
50 42 self.__vel += .5 * self. __acc * self. The first demonstrative application models the retrogressive failure 101
51 __dt of a long clayey slope in the plane strain condition. The slope geometry 102
52 43 self. __updateContact () #if there is any is 25 m in length and 5 m in height with a slope angle of 45◦ , as 103
53 contact shown in Fig. 5. The domain is initially discretized into 9410 nodes 104
54 44 self.__T += self.__dt # total elapsed with a characteristic spacing ℎ = 0.12 m. The material is sensitive clay, 105
55 time for which large-scale landslides are commonly observed in Scandinavia 106
56
57 45 self.__dt = dt and eastern Canada. According to Locat et al. (2011), a strain-softening 107
soil under totally undrained condition is necessary for the occurrence 108
Listing 4: Python class coupling with MFront
of progressive slope failure. Following previous studies (Zhang et al., 109
2017, 2018b; Islam et al., 2019; Yuan et al., 2020; Jin et al., 2020), 110
58 4. Validation tests the undrained condition is assumed by treating the material pressure 111
independent with an undrained shear strength 𝜏 for simplicity. The 112
59 4.1. Axial vibration of elastic cantilever beam von Mises yield criterion is adopted with a Voce isotropic hardening 113
law (Voce, 1948), the implementation of which in MFront DSL is given 114
60 The implementation of NSPFEM is first validated by simulating an in Listing 5. The undrained shear strength 𝜏 of the material is expressed 115
61 elastic cantilever beam under axial vibration. The beam is 10 m in as 116
62 length and 1 m in height, and fixed at the left side. The material 𝑝
𝜏 = 𝜏𝑟 + (𝜏0 − 𝜏𝑟 )𝑒−𝑏𝜀 (20) 117
63 parameters are assigned following previous studies (Yuan et al., 2019;
64 Zhang et al., 2021) as: the Young’s modulus 𝐸 = 100 Pa, the Poisson’s where 𝜏0 and 𝜏𝑟 are the initial and the residual yield stresses, respec- 118
65 ratio 𝜈 = 0, and the density 𝜌 = 1 kg/m3 . Plane strain condition is tively; 𝑏 is the softening exponent; and 𝜀𝑝 is the equivalent plastic 119
66 assumed for the current 2D simulation. The analytical solution of the strain. It is noted that a similar expression of strength degradation with 120
 N. Guo and Z.X. Yang

Fig. 3. Axial vibration of an elastic beam: (a) velocity and (b) displacement at the right side.

Fig. 4. Rolling cylinder down incline: (a) geometry and discretization, and (b) center-of-mass displacement along the slope.

1 shear strain is adopted in Tran and Sołowski (2019), whereas Dey et al. 18 isotropic_hardening : " Voce " {R0 : " 26
2 (2015) and Wang and Hawlader (2017) used the shear displacement tau0 " , Rinf : " taur " , b : " b " } 27
19 } 28
3 along the shear band instead of the shear strain in the formulation.
20 }; 29
30
4 While the strain-rate effect is considered in Tran and Sołowski (2019)
5 and proved to affect the run-out distance, the factor has been neglected Listing 5: Implementing the von Mises model with Voce hardening in
6 in the present study for simplicity. MFront

7 To illustrate the model performance, a plane-strain uni-axial com- 31

8 1 @DSL Implicit ; // implicit stress integration pression test is conducted where the material element is subjected to an 32
9 scheme axial compressive strain in the 𝑦-axis while keeping zero strain and zero 33
10 2 @Behaviour MisesVocePlasticity ; stress in the 𝑧- and 𝑥-directions, respectively. The von Mises stress–axial 34
11 3 @Algorithm NewtonRaphson ; strain curve is shown in Fig. 6 using the following material parameters: 35
12 4 @Epsilon 1.e -14;
Young’s modulus 𝐸 = 1 MPa, Poisson’s ratio 𝜈 = 0.33, 𝜏0 = 20 kPa, 36
13 5 @Theta 1;
14 6
𝜏𝑟 = 4 kPa, and 𝑏 = 5. The value of Poisson’s ratio in the present study 37
15 7 @MaterialProperty stress tau0; follows that in Zhang et al. (2017) and Yuan et al. (2020). It should be 38
16 8 tau0. setEntryName ( " InitYieldStress " ); noted, however, that a value around 0.49 is more appropriate for the 39
17 9 @MaterialProperty stress taur; undrained condition (Dey et al., 2015; Wang and Hawlader, 2017; Tran 40
18 10 taur. setEntryName ( " ResidualYieldStress " ); and Sołowski, 2019). The response of a more commonly used piecewise 41
19 11 @MaterialProperty real b; linear hardening von Mises model (Wang et al., 2016) with the same 42
20 12 b. setEntryName ( " SoftenExponent " );
initial and residual yield stresses and a hardening slope of −50 kPa is 43
21 13
22 14 @Brick StandardElastoViscoPlasticity { superimposed in the figure for comparison. It is seen that both models 44
23 15 stress_potential : " Hooke " , can capture the rapid strength degradation with cumulative plastic 45
24 16 inelastic_flow : " Plastic " { strain, which is essential for the retrogressive slope failure to occur. 46
25 17 criterion : " Mises " , It should also be noted that the adoption of the Voce softening law 47
 N. Guo and Z.X. Yang

Fig. 5. Geometry and initial mesh of the slope.

whereas the sliding plane (𝑥 > 25 m, 𝑦 = 0 m) is frictional with a 5

frictional coefficient 𝜇 = 0.3. The left surface (𝑥 = 0 m) of the slope 6
has degrees of freedom only in the 𝑦-axis. The typical bulk unit weight 7
for sensitive clays in eastern Canada is reported between 16 kN/m3 8
and 18 kN/m3 (Locat et al., 2013; Dey et al., 2015; Locat et al., 2015; 9
Tran and Sołowski, 2019). Thakur and Degago (2012) reported a bulk 10
unit weight of around 20 kN/m3 for the Norwegian sensitive clay. 11
Without being specific in the present study, the bulk unit weight is set 12
to 𝛾 = 20 kN/m3 following other numerical studies (Wang et al., 2016; 13
Islam et al., 2019; Yuan et al., 2020; Zhang et al., 2018b, 2021). The 14
initial stress condition for the slope is assumed to be 𝑘0 for simplicity, 15
which means that the vertical and the horizontal stresses at a material 16
point are calculated as 𝜎𝑣 = 𝛾𝐷 and 𝜎ℎ = 𝑘0 𝜎𝑣 , respectively, where 17
𝐷 is the depth of the point measured from the slope surface and the 18
coefficient of earth pressure at rest is set to 𝑘0 = 0.6 in the study. With 19
the above simplified initial condition and the same material parameters 20
presented in Fig. 6 (Voce hardening), the slope is inherently unstable 21
and will flow under gravity. A total time of 𝑡 = 5 s is simulated for this 22
Fig. 6. Stress–strain curves of the von Mises strain-softening model.
case. 23
The retrogressive failure process of the slope is presented in Fig. 7. 24
As aforementioned that the slope is unstable and expected to fail, a 25
1 in the study roots in its readiness in MFront so its implementation is few slip surfaces originating from the slope toe and extending to the 26
2 effortless as demonstrated by Listing 5. top surface appear almost simultaneously, among which two (labeled 27
3 For the retrogressive slope failure modeling, the original bottom as s1 and s2) become more prominent than the others at 𝑡 = 0.6 s. Af- 28
4 (0 m ≤ 𝑥 ≤ 25 m, 𝑦 = 0 m) of the slope is fixed in both 𝑥- and 𝑦-axes, terwards, s1 and s2 develop continuously into full-fledged slip surfaces, 29

Fig. 7. Retrogressive failure process of the slope.

 N. Guo and Z.X. Yang

Fig. 8. Mesh for the failed slope with CPT optimization.

1 and another two surfaces in front of and behind the two, s3 and s4
2 respectively, also form completely at 𝑡 = 1.5 s. Four apparent sliding
3 blocks in the slope arise from the four slip surfaces. The domino effect
4 of the preceding failure leads to the development of the 5th slip surface,
5 s5, as shown at 𝑡 = 3 s in the figure. Although the computational domain
6 is limited to 25 m long and the simulation terminates at 𝑡 = 5 s, the
7 retrogressive failure process has not finished at this moment as new
8 slip surfaces (e.g. s6) still emerge successively. The runout distance
9 measured from the flow front to the original slope toe is about 26.5 m,
10 and the retrogression distance measured from the original toe to the
11 current slope crest (formed due to s5) is about 18 m.
12 The mesh for the failed slope at 𝑡 = 5 s is illustrated in Fig. 8.
13 It is seen that the mesh generally possesses good quality with CPT
14 optimization. Nevertheless, adding extra nodes to the domain especially
15 at regions with severe mesh distortion could further improve the overall
16 mesh quality. It should be noted that a combination of the two opera-
17 tions (i.e., mesh optimization plus additional nodes) might be necessary
18 in some cases (Cremonesi et al., 2020).
19 The front velocity (i.e., the velocity of the headmost node) and the Fig. 9. Front velocity and kinetic energy of the slope during time.
20 kinetic energy of the slope during the sliding are plotted in Fig. 9.
21 Both quantities first rise rapidly and show a decreasing trend at the
22 late stage of the computation. The front velocity reaches its peak value
23 of 7.57 m/s at 𝑡 = 2.6 s, which is earlier than the occurrence of the
24 maximum kinetic energy of 6.76 × 105 J per unit length at 𝑡 = 4.05 s
25 (note that the front velocity is not necessarily the maximum velocity of
26 the slope).

27 5.2. Pullout of strip plate anchor

28 The vertical pullout of a strip plate anchor is modeled in this

29 subsection following Song et al. (2008), in which RITSS was used to
30 handle large deformations. Similar results have been presented in a
31 more recent study by Liang et al. (2021) using coupled MPM and DEM.
32 The strip plate anchor has a width of 𝐵 = 1 m and a thickness of 0.2 m,
33 and is embedded in the soil with a depth 𝐻∕𝐵 = 3. Due to symmetry
34 of the problem, only half of the domain is simulated which has been
35 discretized using 3024 nodes with a characteristic spacing of 0.1 m, as
36 shown in Fig. 10 along with the dimension and the boundary condition
37 of the domain. The soil is assumed weightless and elasto-perfectly
38 plastic with the Tresca yield criterion implemented in MFront. The
39 material parameters are: the undrained shear strength 𝑠𝑢 = 10 kPa, the
40 Young’s modulus 𝐸 = 500𝑠𝑢 , the Poison’s ratio 𝜈 = 0.49, and the density
41 𝜌 = 1600 kg/m3 . The interface of the soil and the anchor is treated by
42 Coulomb frictional contact presented in Section 2.3 with a frictional
43 coefficient 𝜇 = 0.3. The anchor is pulled upward vertically with a Fig. 10. Domain discretization, dimension, and boundary condition for anchor pullout.
44 constant speed of 0.02 m/s for 5 s, corresponding to a total movement
45 of 0.1 m (0.1𝐵). As separation between the soil and the bottom of the
46 anchor is allowed, the result of the present study is expected to be pressure stays steady afterwards with mild fluctuations and the peak 52
47 comparable with that under the so-called vented condition in Song et al.
value can be chosen as the ultimate resistance pressure as 𝑞𝑢 = 51 kPa. 53
48 (2008). The evolution of the pullout resistance applied on
49 the anchor is plotted in Fig. 11(a). It is seen that the resistance pressure The ultimate capacity factor is defined as (Song et al., 2008) 54
50 gradually increases and reaches a plateau after a pullout distance of 𝑞𝑢
51 about 0.04 m (0.04𝐵) similar to that observed in Liang et al. (2021). The 𝑁𝑐 = (21) 55
𝑠𝑢
 N. Guo and Z.X. Yang

Fig. 11. Pullout resistance: (a) evolution with pullout distance and (b) ultimate capacity factor.

Fig. 12. Deformation contours at the final state: (a) vertical displacement and (b) equivalent strain.

Fig. 13. (a) Geometry and boundary condition, (b) initial vertical stress profile, and (c) RVE packing at node  of the trapdoor problem.

1 𝑁𝑐 is known to increase with the ratio of the embedding depth and Fig. 11(b), where the present result 𝑁𝑐 = 5.1 at 𝐻∕𝐵 = 3 is superim- 4
2 the anchor width 𝐻∕𝐵. Their relation for weightless Tresca soil under
3 the vented condition obtained by Song et al. (2008) is illustrated in posed for comparison. Given that the anchor thickness in the present 5
 N. Guo and Z.X. Yang

Fig. 14. Contours at 𝑡 = 0.6 s: (a) vertical displacement, (b) accumulated equivalent strain, and (c) incremental equivalent strain (from 𝑡 = 0.594 s to 0.6 s).

Table 1 freedom only in the vertical direction; and free surface is assumed on 38
RVE model parameters for the trapdoor problem.
the top of the domain. The domain is discretized into 2449 nodes, each 39
Radius (mm) 𝜌𝑝 (kg/m3 ) 𝐸𝑐 (MPa) 𝜈𝑐 𝜙𝑝 (rad)
associated with an RVE packing. As a result, the computational domain 40
3–7 2650 600 0.3 0.5 contains a total of 2449 RVE packings and the overall number of dis- 41
crete element particles amounts to approximately 1 million. The initial 42
stress state is generated by allowing the material to settle under gravity. 43
1 study is larger than theirs,1 a slightly bigger 𝑁𝑐 is predicted (Rowe, The bulk unit weight of the material is assumed to be 𝛾 = 16 kN/m3 . 44
2 1978). The profile of the initial vertical stress 𝜎𝑣 is presented in Fig. 13(b). 45
3 The contours of the vertical displacement and the equivalent plastic The coefficient of earth pressure at rest 𝑘0 on the bottom is found 46
4 strain at the final state are given in Fig. 12. It is observed that massive to be around 0.64. An RVE packing extracted from node  (location 47
5 soils above the anchor have been mobilized forming two distinct shear marked in Fig. 13(a)) on the center line and buried at a depth of 1.2 m 48
6 bands emanating from the right edge of the anchor. One propagates (𝜎𝑣 ≈ 19.1 kPa and 𝑘0 ≈ 0.68 for this RVE) is visualized in Fig. 13(c), 49
7 upper-leftward to intersect the centerline and the other to the free where the superimposed force chains have widths proportional to the 50
8 surface. The deformation pattern is consistent to that observed in loose magnitudes of the inter-particle normal contact forces, indicative of 51
9 soil and shallow-to-medium embedding depth conditions (Liang et al., the force transmission pattern inside the packing. It should be noted 52
10 2021). that the multiscale simulation is significantly more time consuming 53
than the retrogressive slope failure test due to local RVE computations, 54
11 5.3. Multiscale modeling of trapdoor therefore, a relatively coarse mesh is adopted for this case and we 55
terminate the simulation at 𝑡 = 0.6 s, which corresponds to a trapdoor 56
12 In the third example, we conduct a multiscale simulation of the downward movement of 0.06 m (i.e., 10%𝐵). 57
13 trapdoor problem by coupling NSPFEM with DEM. Similar work on
The final displacement and deformation contours of the problem 58
14 NSPFEM/DEM multiscale modeling has been presented in Guo et al.
domain are depicted in Fig. 14. The results are in general consistent 59
15 (2021b), which analyzed the failure of a small-scale slope and a footing.
with other published results from experimental, analytical, and DEM 60
16 The RVE (see Fig. 13(c)) possesses a periodic structure and is com-
17 posed of 400 cohesionless circular particles with radii ranging between studies (Liang and Xu, 2019; Ali et al., 2020). It is found by Liang 61
18 3 mm and 7 mm. The particle density 𝜌𝑝 is 2650 kg/m3 . The inter- and Xu (2019) that when 𝐻∕𝐵 < 2, the sliding surfaces of the moving 62
19 particle contact behavior is described by the linear force–displacement particles are roughly vertical, whereas the shape of sliding boundaries 63
20 law as well as the Coulomb friction criterion, which involves three is close to elliptic when 𝐻∕𝐵 = 3. In the current study 𝐻∕𝐵 = 2.5, 64
21 user-defined parameters, 𝐸𝑐 and 𝜈𝑐 for the normal and the tangential the contour of the vertical displacement shown in Fig. 14(a) resembles 65
22 contact stiffnesses, respectively, and 𝜙𝑝 for the particle friction angle bulb-like shapes, which is in agreement with their findings. Multiple 66
23 (FrictMat material in YADE Šmilauer et al., 2015). The values of curvilinear shear bands are observed from the accumulated equivalent 67
24 the parameters for the RVE model is summarized in Table 1 which strain contour (Fig. 14(b)) that originate from the edge of the trap- 68
25 are consistent with our previous studies. The bulk elastic moduli of the door and propagate upper-right to intersect the centerline at different 69
26 RVE have also been obtained previously from small-strain shear tests on heights. Among these shear bands, one is found dominant with the 70
27 the RVE (Guo and Zhao, 2016c; Guo et al., 2016), whose magnitudes others less distinct. However, from the incremental equivalent strain 71
28 are found to be in the order of 10 MPa. Accordingly, the sound wave contour (Fig. 14(c)), it is seen that one of the tributary shear bands 72
29 velocity of the medium is estimated to be less than 100 m/s, which can developed at a later stage becomes dominant, suggesting a progressive 73
30 be used to determine the time step 𝛥𝑡 following the CFL condition. deformation inside the domain (Han et al., 2017). 74
31 The original problem domain is 3 m in length and 1.5 m in depth The normalized vertical pressure 𝑃 ∕𝑃0 , where 𝑃0 is the initial pres- 75
32 (denoted by 𝐻). But due to symmetry, only half of the domain is sure, applied on the trapdoor throughout the downward movement is 76
33 simulated as shown in Fig. 13(a). The half width of the trapdoor, which
depicted in Fig. 15. The curve shows an almost instant abrupt decrease 77
34 is located in the center of the bottom surface, is 𝐵∕2 = 0.3 m. For the
when the trapdoor starts moving downward. The minimum 𝑃 ∕𝑃0 = 0.25 78
35 boundary condition, the left and the bottom surfaces are fixed in space
attained at the movement of 0.005 m is close to that reported in Liang 79
36 except the trapdoor that will move downward at a constant speed of
and Xu (2019). After the lowest point, 𝑃 ∕𝑃0 increases slightly to 0.33 80
37 0.1 m/s; the right surface is supported by rollers thus has degrees of
at the final state due to the soil arching effect, a typical phenomenon 81
observed in many studies (Iglesia et al., 2014; Han et al., 2017). 82
1
If the anchor thickness (gap) is too small, the alpha shape method fails Cross-scale analysis is one of the major strengths provided by the 83
to capture the boundary accurately. multiscale approach. For the demonstration purpose, we examine the 84
 N. Guo and Z.X. Yang

Fig. 15. Reduction of the vertical pressure on the trapdoor.

1 fabric of each RVE packing inside the domain. The contact normal-
2 based fabric is described by a second-order tensor 𝜙𝑖𝑗 defined be-
3 low (Oda, 1982)
1 ∑ 𝑐 𝑐
4 𝜙𝑖𝑗 = 𝑛 𝑛 (22) Fig. 16. Fabric field showing the principal directions of inter-particle contacts (line
𝑁𝑐 𝑁 𝑖 𝑗 segments with red and gray colors indicate major and minor principal directions,
𝑐
respectively).
5 where 𝑁𝑐 is the total number of contacts within the packing; 𝑛𝑐𝑖 is the
6 unit vector along the normal direction of a contact. The directional
7 distribution of contact normals can then be fitted by the following
6. Conclusions 39
8 function
1 We have presented a lightweight code, NSPFEM2D, that implements 40
9 𝐸(𝛩) = (1 + 𝑎𝑐𝑖𝑗 𝑛𝑐𝑖 𝑛𝑐𝑗 ) (23)
2𝜋 the node-based smoothed particle finite element method (NSPFEM) in 41
10 where 𝑎𝑐𝑖𝑗 is the deviatoric fabric tensor given by 2D based on a number of open-source libraries. NSPFEM is specially 42
( ) targeted at solving large-deformation problems by tracking the material 43
1
11 𝑎𝑐𝑖𝑗 = 4 𝜙𝑖𝑗 − 𝛿𝑖𝑗 (24) deformation using a set of Lagrangian nodes and frequent remeshing 44
2
on these nodes using Delaunay triangulation. In addition, the nodal 45
12 with 𝛿𝑖𝑗 being the Kronecker delta. Fig. 16 illustrates the fabric field of integration technique borrowed from NSFEM is employed to avoid 46
13 the whole domain at 𝑡 = 0.6 s, where each fabric tensor is visualized interpolation of material variables after remeshing. 47
14 as a cross. The red and the gray line segments represent the major Owing to the hybrid programming of C++ and Python, NSPFEM2D 48
15 and the minor principal directions of a tensor, respectively, with their is highly efficient and versatile to couple with other well-developed 49
16 length and width proportional to the corresponding eigenvalues. A codes, such as MFront for the implementation of constitutive models 50
17 transition of the major principal direction of inter-particle contacts and YADE for multiscale simulations. Three problems, namely retro- 51
18 from vertical at the left side to horizontal at the centerline can be gressive slope failure, pullout of a strip plate anchor, and multiscale 52
19 vividly observed, giving rise to the arching effect around the trapdoor. modeling of trapdoor, are investigated in the study to demonstrate 53
20 Similar illustrations of the fabric field in granular media can be found the capability and flexibility of the code. The code is hosted in the 54
21 elsewhere (Meier et al., 2008; Rycroft et al., 2009). Note that in the public domain (https://github.com/cenguo/SPFEM) to provide access 55
22 DEM simulation, measuring cells are typically used to homogenize the for interested researchers and to serve as a minimal example showcas- 56
23 fabric at different locations of the problem domain (Rycroft et al., ing the development of customized codes on the strength of available 57
24 2009), which are unnecessary in the present multiscale approach. open-source libraries. 58
The smoothing operation in NSFEM can help mitigate the over- 59
25 The two RVE packings associated with the nodes  and  (their
stiffness of FEM when linear triangle elements are used, leading to 60
26 locations marked in Fig. 13(a)) at 𝑡 = 0.6 s are shown in Fig. 17. By
more accurate stress solutions, higher convergence rates, and less mesh- 61
27 comparing with Fig. 13(c), we found that the deformation at node 
dependency (Zeng and Liu, 2018). However, NSFEM gives overly soft 62
28 is insignificant (i.e., the shape of the RVE remains almost unaltered),
behaviors and is usually used for upper bound solution of strain energy 63
29 but the strong force chains in this RVE switch directions from vertical-
in force driving problems. To improve the accuracy of NSFEM, hybrid 64
30 dominant to horizontal-dominant. In contrast, node  has been severely methods that selectively combine NSFEM and edge-based smoothed 65
31 sheared in the vertical direction with a shear mode close to simple- FEM could be exploited (Jiang et al., 2015). Another drawback of 66
32 shear. The contact normal distributions of the two packings at the NSFEM is that it may suffer from temporal instability in solving dy- 67
33 initial and the final states are also presented in the figure, where the namic problems (Zeng and Liu, 2018). Some stabilization techniques, 68
34 smooth curves are fittings of numerical data by Eq. (23). It is seen e.g., adding squared-residual or gradient terms (Zhang and Liu, 2010; 69
35 that initially both packings are slightly anisotropic but finally develop Wang et al., 2015b) have been proved effective in remedying temporal 70
36 strong anisotropy. The major principal directions of the two packings instability. Given all these considerations, it will be the next pursuit 71
37 at the end of simulation tilt at an angle of 16◦ and 31◦ in relative to of the authors to develop the corresponding stabilized NSPFEM for 72
38 the horizontal direction, respectively. large-deformation geotechnical problems in the future. 73
 N. Guo and Z.X. Yang

Fig. 17. Final RVE packings of nodes  and , and the rose diagrams showing the distribution of inter-particle contact orientations.

1 CRediT authorship contribution statement Bardenhagen, S.G., Brackbill, J.U., Sulsky, D., 2000. The material-point method for 29
granular materials. Comput. Methods Appl. Mech. Engrg. 187, 529–541. http: 30
2 N. Guo: Conceptualization, Methodology, Data curation, Funding //dx.doi.org/10.1016/S0045-7825(99)00338-2. 31
Benson, D.J., 1992. Computational methods in Lagrangian and Eulerian hydrocodes. 32
3 acquisition, Writing – original draft, Visualization. Z.X. Yang: Concep- Comput. Methods Appl. Mech. Engrg. 99, 235–394. http://dx.doi.org/10.1016/ 33
4 tualization, Funding acquisition, Writing – review & editing. 0045-7825(92)90042-I. 34
Bui, H.H., Fukagawa, R., 2013. An improved SPH method for saturated soils and 35
5 Declaration of competing interest its application to investigate the mechanisms of embankment failure: Case of 36
hydrostatic pore-water pressure. Int. J. Numer. Anal. Methods Geomech. 37, 31–50. 37
http://dx.doi.org/10.1002/nag.1084. 38
6 The authors declare that they have no known competing finan- Chen, L., Holst, M., 2011. Efficient mesh optimization schemes based on optimal 39
7 cial interests or personal relationships that could have appeared to Delaunay triangulations. Comput. Methods Appl. Mech. Engrg. 200, 967–984. 40
8 influence the work reported in this paper. http://dx.doi.org/10.1016/j.cma.2010.11.007. 41
Chen, Z., Tho, K.K., Leung, C.F., Chow, Y.K., 2013. Influence of overburden pressure 42
and soil rigidity on uplift behavior of square plate anchor in uniform clay. Comput. 43
9 Acknowledgments Geotech. 52, 71–81. http://dx.doi.org/10.1016/j.compgeo.2013.04.002. 44
Chen, X., Zhang, L., Zhang, L., Zhou, Y., Ye, G., Guo, N., 2021. Efficient mesh 45
10 The authors are grateful to the two anonymous reviewers for their optimization schemes based on optimal Delaunay triangulations. Comput. Geotech. 46
11 constructive comments on the paper. The study has been financially 129, 103877. http://dx.doi.org/10.1016/j.compgeo.2020.103877. 47
Courant, R., Friedrichs, K., Lewy, H., 1928. Über dIe partiellen differenzengleichungen 48
12 supported by the National Natural Science Foundation of China (Nos. der mathematischen physik. Math. Ann. 100, 32–74, (in German). 49
13 52078456 and 52020105003) and the Fundamental Research Funds for Cremonesi, M., Franci, A., Idelsohn, S., Oñate, E., 2020. A state of the art review 50
14 the Central Universities (No. 2021FZZX001-14). of the particle finite element method (PFEM). Arch. Comput. Methods Eng. 27, 51
1709–1735. http://dx.doi.org/10.1007/s11831-020-09468-4. 52
Dassault Systèmes, 2016. Abaqus Analysis User’s Guide, two thousand and sixteenth 53
15 References ed. Dassault Systèmes Simulia Corp.Johnston, Rhode Island. 54
Dávalos, C., Cante, J., Hernández, J., Oliver, J., 2015. On the numerical modeling of 55
16 Ali, U., Otsubo, M., Ebizuka, H., Kuwano, R., 2020. Particle-scale insight into soil granular material flows via the particle finite element method (PFEM). Int. J. Solids 56
17 arching under trapdoor condition. Soils Found. 60, 1171–1188. http://dx.doi.org/ Struct. 71, 99–125. http://dx.doi.org/10.1016/j.ijsolstr.2015.06.013. 57
18 10.1016/j.s{andf}.2020.06.011. Dey, R., Hawlader, B., Phillips, R., Soga, K., 2015. Large deformation finite-element 58
19 Aubry, R., Idelsohn, S., Oñate, E., 2005. Particle finite element method in fluid- modelling of progressive failure leading to spread in sensitive clay slopes. 59
20 mechanics including thermal convection–diffusion. Comput. Struct. 83, 1459–1475. Géotechnique 65, 657–668. http://dx.doi.org/10.1680/geot.14.P.193. 60
21 http://dx.doi.org/10.1016/j.compstruc.2004.10.021. Edelsbrunner, H., Mücke, E.P., 1994. Three-dimensional alpha shapes. ACM Trans. 61
22 Bai, B., Rao, D., Xu, T., Chen, P., 2018. SPH-FDM Boundary for the analysis of Graph. 13, 43–72. http://dx.doi.org/10.1145/174462.156635. 62
23 thermal process in homogeneous media with a discontinuous interface. Int. J. Heat Franci, A., 2021. Lagrangian finite element method with nodal integration for fluid– 63
24 Mass Transfer 117, 517–526. http://dx.doi.org/10.1016/j.ijheatmasstransfer.2017. solid interaction. Comput. Part. Mech. 8, 389–405. http://dx.doi.org/10.1007/ 64
25 10.004. s40571-020-00338-1. 65
26 Bardenhagen, S.G., 2002. Energy conservation error in the material point method for Franci, A., Cremonesi, M., Perego, U., Oñate, E., 2020. A Lagrangian nodal integration 66
27 solid mechanics. J. Comput. Phys. 180, 383–403. http://dx.doi.org/10.1006/jcph. method for free-surface fluid flows. Comput. Methods Appl. Mech. Engrg. 361, 67
28 2002.7103. 112816. http://dx.doi.org/10.1016/j.cma.2019.112816. 68
 N. Guo and Z.X. Yang

1 Gingold, R.A., Monaghan, J.J., 1977. Smoothed particle hydrodynamics: theory and Li, H., Zhang, S., 2020. Implement the particle finite element method in ABAQUS. Jpn. 81
2 application to non-spherical stars. Mon. Not. R. Astronimical Soc. 181, 375–389. Geotech. Soc. Spec. Publ. 8, 70–75. http://dx.doi.org/10.3208/jgssp.v08.c16. 82
3 http://dx.doi.org/10.1093/mnras/181.3.375. Liang, L., Xu, C., 2019. Numerical and theoretical research on stress distribution in the 83
4 González Acosta, P.J., Remmerswaal, G., Hicks, M.A., 2020. An investigation of stress loosening zone of the trapdoor problem. Int. J. Numer. Anal. Methods Geomech. 84
5 inaccuracies and proposed solution in the material point method. Comput. Mech. 43, 1426–1447. http://dx.doi.org/10.1002/nag.2906. 85
6 65, 555–581. http://dx.doi.org/10.1007/s00466-019-01783-3. Liang, W., Zhao, J., Wu, H., Soga, K., 2021. Multiscale modeling of anchor pullout 86
7 Guo, N., Chen, L.F., Yang, Z.X., 2021a. Multiscale modelling and analysis of footing in sand. J. Geotech. Geoenviron. Eng. 147, 04021091. http://dx.doi.org/10.1061/ 87
8 resting on an anisotropic sand. Géotechnique http://dx.doi.org/10.1680/jgeot.20. (ASCE)GT.1943-5606.0002599. 88
9 P.306, (in press). Libersky, L.D., Petschek, A.G., 1991. Smooth particle hydrodynamics with strength of 89
10 Guo, N., Yang, Z., Yuan, W., Zhao, J., 2021b. A coupled SPFEM/DEM approach for materials. In: Trease, H.E., Fritts, M.F., Crowley, W.P. (Eds.), Advances in the Free- 90
11 multiscale modeling of large-deformation geomechanical problems. Int. J. Numer. Lagrange Method Including Contributions on Adaptive Gridding and the Smooth 91
12 Anal. Methods Geomech. 45, 648–667. http://dx.doi.org/10.1002/nag.3175. Particle Hydrodynamics Method. In: Lecture Notes in Physics, vol. 395, Springer, 92
13 Guo, N., Zhao, J., 2014. A coupled FEM/DEM approach for hierarchical multiscale Berlin, Heidelberg, pp. 248–257. http://dx.doi.org/10.1007/3-540-54960-9_58. 93
14 modelling of granular media. Internat. J. Numer. Methods Engrg. 99, 789–818. Liu, G., Nguyen-Thoi, T., Nguyen-Xuan, H., Lam, K., 2009. A node-based smoothed 94
15 http://dx.doi.org/10.1002/nme.4702. finite element method (NS-FEM) for upper bound solutions to solid mechanics 95
16 Guo, N., Zhao, J., 2016a. 3D Multiscale modeling of strain localization in granular problems. Comput. Struct. 87, 14–26. http://dx.doi.org/10.1016/j.compstruc.2008. 96
17 media. Comput. Geotech. 80, 360–372. http://dx.doi.org/10.1016/j.compgeo.2016. 09.003. 97
18 01.020. Locat, A., Jostad, H.P., Leroueil, S., 2013. Numerical modeling of progressive failure 98
19 Guo, N., Zhao, J., 2016b. Multiscale insights into classical geomechanics problems. Int. and its implications for spreads in sensitive clays. Can. Geotech. J. 50, 961–978. 99
20 J. Numer. Anal. Methods Geomech. 40, 367–390. http://dx.doi.org/10.1002/nag. http://dx.doi.org/10.1139/cgj-2012-0390. 100
21 2406. Locat, A., Leroueil, S., Bernander, S., Demers, D., Jostad, H.P., Ouehb, L., 2011. 101
22 Guo, N., Zhao, J., 2016c. Parallel hierarchical multiscale modelling of hydro-mechanical Progressive failures in eastern Canadian and Scandinavian sensitive clays. Can. 102
23 problems for saturated granular soils. Comput. Methods Appl. Mech. Engrg. 305, Geotech. J. 48, 1696–1712. http://dx.doi.org/10.1139/t11-059. 103
24 37–61. http://dx.doi.org/10.1016/j.cma.2016.03.004. Locat, A., Leroueil, S., Fortin, A., Demers, D., Jostad, H.P., 2015. The 1994 landslide at 104
25 Guo, N., Zhao, J., Sun, W.C., 2016. Multiscale analysis of shear failure of thick-walled Sainte-Monique, Quebec: geotechnical investigation and application of progressive 105
26 hollow cylinder in dry sand. Géotechnique Lett. 6, 77–82. http://dx.doi.org/10. failure analysis. Can. Geotech. J. 52, 490–504. http://dx.doi.org/10.1139/cgj-2013- 106
27 1680/jgele.15.00149. 0344. 107
28 Hallquist, J.O., 2006. LS-DYNA Theory Manual, rtwelveth ed. Livermore Software Lucy, L.B., 1977. A numerical approach to the testing of the fission hypothesis. Astron. 108
29 Technology Corporation, Livermore, California. J. 82, 1013–1024. http://dx.doi.org/10.1086/112164. 109
30 Han, J., Wang, F., Al-Naddaf, M., Xu, C., 2017. Progressive development of two- Meier, H.A., Schlemmer, M., Wagner, C., Kerren, A., Hagen, H., Kuhl, E., Steinmann, P., 110
31 dimensional soil arching with displacement. Int. J. Geomech. 17, 04017112. http: 2008. Visualization of particle interactions in granular media. IEEE Trans. Vis. 111
32 //dx.doi.org/10.1061/(ASCE)GM.1943-5622.0001025. Comput. Graphics 14, 1110–1125. http://dx.doi.org/10.1109/TVCG.2008.65. 112
33 Helfer, T., Bleyer, J., Frondelius, T., Yashchuk, I., Nagel, T., Naumov, D., 2020. Nazem, M., Sheng, D., Carter, J.P., 2006. Stress integration and mesh refinement 113
34 The mFrontGenericInterfaceSupport project. J. Open Source Softw. 5 (2003), http: for large deformation in geomechanics. Internat. J. Numer. Methods Engrg. 65, 114
35 //dx.doi.org/10.21105/joss.02003. 1002–1027. http://dx.doi.org/10.1002/nme.1470. 115
36 Helfer, T., Michel, B., Proix, J.M., Salvo, M., Sercombe, J., Casella, M., 2015. Introduc- Oñate, E., Idelsohn, S.R., Pin, F.Del., Aubry, R., 2004. The particle finite element 116
37 ing the open-source mfront code generator: Application to mechanical behaviours method – an overview. Int. J. Comput. Methods 1, 267–307. http://dx.doi.org/ 117
38 and material knowledge management within the pleiades fuel element modelling 10.1142/S0219876204000204. 118
39 platform. Comput. Math. Appl. 70, 994–1023. http://dx.doi.org/10.1016/j.camwa. Oda, M., 1982. Fabric tensor for discontinuous geological materials. Soils Found. 22, 119
40 2015.06.027. 96–108. http://dx.doi.org/10.3208/s{andf}1972.22.4_96. 120
41 Hert, S., Seel, M., 2020. DD convex hulls and delaunay triangulations. In: CGAL User Peng, C., Wu, W., Yu, H.S., Wang, C., 2015. A SPH approach for large deformation 121
42 and Reference Manual, 5.0.2 ed. CGAL Editorial Board. analysis with hypoplastic constitutive model. Acta Geotech. 10, 703–717. http: 122
43 Hinton, E., Rock, T., Zienkiewicz, O.C., 1976. A note on mass lumping and related //dx.doi.org/10.1007/s11440-015-0399-3. 123
44 processes in the finite element method. Earthq. Eng. Struct. Dyn. 4, 245–249. Peng, C., Zhou, M., Wu, W., 2017. Large deformation modeling of soil-machine 124
45 http://dx.doi.org/10.1002/eqe.4290040305. interaction in clay. In: Papamichos, E., Papanastasiou, P., Pasternak, E., Dyskin, A. 125
46 Hu, Y., Randolph, M.F., 1998. A practical numerical approach for large deformation (Eds.), Bifurcation and Degradation of Geomaterials with Engineering Applications. 126
47 problems in soil. Int. J. Numer. Anal. Methods Geomech. 22, 327–350. http://dx. Springer, pp. 249–257. http://dx.doi.org/10.1007/978-3-319-56397-8_32. 127
48 doi.org/10.1002/(SICI)1096-9853(199805)22:5<327::AID-NAG920>3.0.CO;2-X. Puso, M.A., Chen, J.S., Zywicz, E., Elmer, W., 2008. Meshfree and finite element 128
49 Idelsohn, S.R., Oñate, E., Del Pin, F., 2004. The particle finite element method: a nodal integration methods. Internat. J. Numer. Methods Engrg. 74, 416–446. 129
50 powerful tool to solve incompressible flows with free-surfaces and breaking waves. http://dx.doi.org/10.1002/nme.2181. 130
51 Internat. J. Numer. Methods Engrg. 61, 964–989. http://dx.doi.org/10.1002/nme. Qiu, G., Grabe, J., 2012. Numerical investigation of bearing capacity due to spudcan 131
52 1096. penetration in sand overlying clay. Can. Geotech. J. 49, 1393–1407. http://dx.doi. 132
53 Idelsohn, S.R., Oñate, E., Del Pin, F., Calvo, N., 2006. Fluid–structure interaction using org/10.1139/t2012-085. 133
54 the particle finite element method. Comput. Methods Appl. Mech. Engrg. 195, Qiu, G., Henke, S., Grabe, J., 2011. Application of a coupled Eulerian–Lagrangian ap- 134
55 2100–2123. http://dx.doi.org/10.1016/j.cma.2005.02.026. proach on geomechanical problems involving large deformations. Comput. Geotech. 135
56 Iglesia, G.R., Einstein, H.H., Whitman, R.V., 2014. Investigation of soil arching with 38, 30–39. http://dx.doi.org/10.1016/j.compgeo.2010.09.002. 136
57 centrifuge tests. J. Geotech. Geoenviron. Eng. 140, 04013005. http://dx.doi.org/ Rodriguez, J.M., Carbonell, J.M., Cante, J.C., Oliver, J., 2016. The particle finite 137
58 10.1061/(ASCE)GT.1943-5606.0000998. element method (PFEM) in thermo-mechanical problems. Internat. J. Numer. 138
59 Islam, N., Hawlader, B., Wang, C., Soga, K., 2019. Large-deformation finite-element Methods Engrg. 107, 733–785. http://dx.doi.org/10.1002/nme.5186. 139
60 modelling of earthquake-induced landslides considering strain-softening behaviour Rodríguez, J.M., Carbonell, J.M., Cante, J.C., Oliver, J., 2017. Continuous chip forma- 140
61 of sensitive clay. Can. Geotech. J. 56, 1003–1018. http://dx.doi.org/10.1139/cgj- tion in metal cutting processes using the particle finite element method (PFEM). 141
62 2018-0250. Int. J. Solids Struct. 120, 81–102. http://dx.doi.org/10.1016/j.ijsolstr.2017.04.030. 142
63 Jiang, C., Zhang, Z.Q., Liu, G.R., Han, X., Zeng, W., 2015. An edge-based/node-based Rowe, R.K., 1978. Soil Structure Interaction Analysis and its Application to the 143
64 selective smoothed finite element method using tetrahedrons for cardiovascu- Prediction of Anchor Behaviour (Ph.D. thesis). University of Sydney, Sydney, 144
65 lar tissues. Eng. Anal. Bound. Elem. 59, 62–77. http://dx.doi.org/10.1016/j. Australia. 145
66 enganabound.2015.04.019. Rycroft, C.H., Kamrin, K., Bazant, M.Z., 2009. Assessing continuum postulates in 146
67 Jin, Z., Yin, Z.Y., Kotronis, P., Jin, Y.F., 2019. Numerical investigation on evolving simulations of granular flow. J. Mech. Phys. Solids 57, 828–839. http://dx.doi. 147
68 failure of caisson foundation in sand using the combined Lagrangian-SPH method. org/10.1016/j.jmps.2009.01.009. 148
69 Mar. Georesour. Geotechnol. 37, 23–35. http://dx.doi.org/10.1080/1064119X. Šmilauer, ., et al., 2015. Using and programming. In: Yade Documentation, second ed. 149
70 2018.1425311. The Yade Project, http://dx.doi.org/10.5281/zenodo.34043. 150
71 Jin, Y.F., Yin, Z.Y., Yuan, W.H., 2020. Simulating retrogressive slope failure using two Soga, K., Alonso, E., Yerro, A., Kumar, K., Bandara, S., 2016. Trends in large- 151
72 different smoothed particle finite element methods: A comparative study. Eng. Geol. deformation analysis of landslide mass movements with particular emphasis on 152
73 279, 105870. http://dx.doi.org/10.1016/j.enggeo.2020.105870. the material point method. Géotechnique 66, 248–273. http://dx.doi.org/10.1680/ 153
74 Li, W.L., Guo, N., Yang, Z.X., Helfer, T., 2021. Large-deformation geomechanical jgeot.15.LM.005. 154
75 problems studied by a shear-transformation-zone model using the material point Song, Z., Hu, Y., Randolph, M.F., 2008. Numerical simulation of vertical pullout of 155
76 method. Comput. Geotech. 135, 104153. http://dx.doi.org/10.1016/j.compgeo. plate anchors in clay. J. Geotech. Geoenviron. Eng. 134, 866–875. http://dx.doi. 156
77 2021.104153. org/10.1061/(ASCE)1090-0241(2008)134:6(866). 157
78 Li, X., Sovilla, B., Jiang, C., Gaume, J., 2020. The mechanical origin of snow avalanche Stomakhin, A., Schroeder, C., Chai, L., Teran, J., Selle, A., 2013. A material point 158
79 dynamics and flow regime transitions. Cryosphere 14, 3381–3398. http://dx.doi. method for snow simulation. ACM Trans. Graph. 32 (102), http://dx.doi.org/10. 159
80 org/10.5194/tc-14-3381-2020. 1145/2461912.2461948. 160
 N. Guo and Z.X. Yang

1 Sulsky, D., Chen, Z., Schreyer, H.L., 1994. A particle method for history-dependent Yang, Z.X., Gao, Y.Y., Jardine, R.J., Guo, W.B., Wang, D., 2020. Large deformation 39
2 materials. Comput. Methods Appl. Mech. Engrg. 118, 179–196. http://dx.doi.org/ finite-element simulation of displacement-pile installation experiments in sand. J. 40
3 10.1016/0045-7825(94)90112-0. Geotech. Geoenviron. Eng. 146, 04020044. http://dx.doi.org/10.1061/(ASCE)GT. 41
4 Swope, W.C., Andersen, H.C., Berens, P.H., Wilson, K.R., 1982. A computer simulation 1943-5606.0002271. 42
5 method for the calculation of equilibrium constants for the formation of physical Yuan, W.H., Liu, K., Zhang, W., Dai, B., Wang, Y., 2020. Dynamic modeling of large 43
6 clusters of molecules: Application to small water clusters. J. Chem. Phys. 76, deformation slope failure using smoothed particle finite element method. Landslides 44
7 637–649. http://dx.doi.org/10.1063/1.442716. 17, 1591–1603. http://dx.doi.org/10.1007/s10346-020-01375-w. 45
8 Thakur, V., Degago, S.A., 2012. Quickness of sensitive clays. Géotechnique Lett. 2, Yuan, W.H., Wang, H.C., Zhang, W., Dai, B.B., Liu, K., Wang, Y., 2021. Particle finite 46
9 87–95. http://dx.doi.org/10.1680/geolett.12.0008. element method implementation for large deformation analysis using abaqus. Acta 47
10 Tran, Q.A., Sołowski, W., 2019. Generalized interpolation material point method Geotech. 16, 2449–2462. http://dx.doi.org/10.1007/s11440-020-01124-2. 48
11 modelling of large deformation problems including strain-rate effects – Application Yuan, W.H., Wang, B., Zhang, W., Jiang, Q., Feng, X.T., 2019. Development of an 49
12 to penetration and progressive failure problems. Comput. Geotech. 106, 249–265. explicit smoothed particle finite element method for geotechnical applications. 50
13 http://dx.doi.org/10.1016/j.compgeo.2018.10.020. Comput. Geotech. 106, 42–51. http://dx.doi.org/10.1016/j.compgeo.2018.10.010. 51
14 Vacondio, R., Altomare, C., De Leffe, X., Le Touzé, S., Marongiu, J.C., Marrone, S., Zeng, W., Liu, G.R., 2018. Smoothed finite element methods (S-FEM): An overview and 52
15 Rogers, B.D., Souto-Iglesias, A., 2021. Grand challenges for smoothed particle recent developments. Arch. Comput. Methods Eng. 25, 397–435. http://dx.doi.org/ 53
16 hydrodynamics numerical schemes. Comput. Part. Mech. 8, 575–588. http://dx. 10.1007/s11831-016-9202-3. 54
17 doi.org/10.1007/s40571-020-00354-1. Zhang, X., Krabbenhoft, K., Pedroso, D.M., Lyamin, A.V., Sheng, D., Vicente da 55
18 Voce, E., 1948. The relationship between stress and strain for homogeneous Silva, M., Wang, D., 2013. Particle finite element analysis of large deformation 56
19 deformation. J. Inst. Metals 74, 537–562. and granular flow problems. Comput. Geotech. 54, 133–142. http://dx.doi.org/10. 57
20 Wang, D., Bienen, B., Nazem, M., Tian, Y., Zheng, J., Pucker, T., Randolph, M.F., 2015a. 1016/j.compgeo.2013.07.001. 58
21 Large deformation finite element analyses in geotechnical engineering. Comput. Zhang, X., Krabbenhoft, K., Sheng, D., 2014. Particle finite element analysis of the 59
22 Geotech. 65, 104–114. http://dx.doi.org/10.1016/j.compgeo.2014.12.005. granular column collapse problem. Granul. Matter 16, 609–619. http://dx.doi.org/ 60
23 Wang, J., Chan, D., 2014. Frictional contact algorithms in SPH for the simulation of 10.1007/s10035-014-0505-5. 61
24 soil–structure interaction. Int. J. Numer. Anal. Methods Geomech. 38, 747–770. Zhang, Z.Q., Liu, G.R., 2010. Temporal stabilization of the node-based smoothed finite 62
25 http://dx.doi.org/10.1002/nag.2233. element method and solution bound of linear elastostatics and vibration problems. 63
26 Wang, G., Cui, X.Y., Feng, H., Li, G.Y., 2015b. A stable node-based smoothed finite Comput. Mech. 46, 229–246. http://dx.doi.org/10.1007/s00466-009-0420-5. 64
27 element method for acoustic problems. Comput. Methods Appl. Mech. Engrg. 297, Zhang, X., Sheng, D., Sloan, S.W., Bleyer, J., 2017. Lagrangian modelling of large defor- 65
28 348–370. http://dx.doi.org/10.1016/j.cma.2015.09.005. mation induced by progressive failure of sensitive clays with elastoviscoplasticity. 66
29 Wang, C., Hawlader, B., 2017. Numerical modeling of three types of sensitive clay slope Internat. J. Numer. Methods Engrg. 112, 963–989. http://dx.doi.org/10.1002/nme. 67
30 failures. In: Proceedings of the 19th International Conference on Soil Mechanics and 5539. 68
31 Geotechnical Engineering, Seoul. pp. 871–874. Zhang, X., Sloan, S.W., Oñate, E., 2018b. Dynamic modelling of retrogressive landslides 69
32 Wang, X., Qiu, Y., Slattery, S.R., Fang, Y., Li, M., Zhu, S.C., Zhu, Y., Tang, M., with emphasis on the role of clay sensitivity. Int. J. Numer. Anal. Methods 70
33 Manocha, D., Jiang, C., 2020. A massively parallel and scalable multi-GPU material Geomech. 42, 1806–1822. http://dx.doi.org/10.1002/nag.2815. 71
34 point method. ACM Trans. Graph. 39 (30), http://dx.doi.org/10.1145/3386569. Zhang, W., Yuan, W., Dai, B., 2018a. Smoothed particle finite-element method for 72
35 3392442. large-deformation problems in geomechanics. Int. J. Geomech. 18, 04018010. 73
36 Wang, B., Vardon, P.J., Hicks, M.A., 2016. Investigation of retrogressive and progressive http://dx.doi.org/10.1061/(ASCE)GM.1943-5622.0001079. 74
37 slope failure mechanisms using the material point method. Comput. Geotech. 78, Zhang, W., Zhong, Z.H., Peng, C., Yuan, W.H., Wu, W., 2021. GPU-Accelerated 75
38 88–98. http://dx.doi.org/10.1016/j.compgeo.2016.04.016. smoothed particle finite element method for large deformation analysis in ge- 76
omechanics. Comput. Geotech. 129, 103856. http://dx.doi.org/10.1016/j.compgeo. 77
2020.103856. 78

View publication stats