Computers and Geotechnics 75 (2016) 126–134 Contents lists available at ScienceDirect Computers and Geotechnics journal homepage: www.elsevier.com/locate/compgeo Research Paper Numerical modelling of seismic slope failure using MPM Tushar Bhandari a,1, Fursan Hamad b, Christian Moormann b,⇑, K.G. Sharma a, Bernhard Westrich b a b Department of Civil Engineering, Indian Institute of Technology-Delhi, Hauz Khas, New Delhi, India Institute of Geotechnical Engineering, University of Stuttgart, Stuttgart, Germany a r t i c l e i n f o Article history: Received 26 September 2015 Received in revised form 12 December 2015 Accepted 17 January 2016 Available online 17 February 2016 Keywords: Material Point Method Non-zero kinematic condition Large deformation Landslides Slope failure a b s t r a c t The Finite Element Method (FEM) is widely used in the simulation of geotechnical applications. Owing to the limitations of FEM to model problems involving large deformations, many efforts have been made to develop methods free of mesh entanglement. One of these methods is the Material Point Method (MPM) which models the material as Lagrangian particles capable of moving through a background computational mesh in Eulerian manner. Although MPM represents the continuum by material points, solution is performed on the computational mesh. Thus, imposing boundary conditions is not aligned with the material representation. In this paper, a non-zero kinematic condition is introduced where an additional set of particles is incorporated to track the moving boundary. This approach is then applied to simulate the seismic motion resulting in failure of slopes. To validate this simulation procedure, two geotechnical applications are modelled using MPM. The first is to reproduce a shaking table experiment where the results of another numerical method are available. After validating the present numerical scheme for relatively large deformation problem, it is applied to simulate progression of a large-scale landslide during the Chi-Chi earthquake of Taiwan in which excessive material deformation and transportation is taking place. Ó 2016 Elsevier Ltd. All rights reserved. 1. Introduction Slopes stability has been considered as one of the most significant topic to study for the geotechnical society over many years. In most cases, the stability of slope is what one aims to analyse. For this reason, the evolution of analysis techniques has taken place in the regime of mesh based methods that can take into account the limiting deformations as well as the effect of supporting structures. In some cases, however, it is important to study the behaviour of the slopes beyond failure. In these cases, slope failure is inevitable and the impact of the sliding mass on low-lying areas becomes important to investigate. The physical significance of slope failures can be gauged in the form of landslides leading to a massive flow of debris. Landslide-debris flow is a very rapid and massive flow-like motion of soil and fragmented rock. The material mobility and the impact of the avalanche can cause ⇑ Corresponding author at: Pfaffenwaldring 35, 70569 Stuttgart, Germany. E-mail addresses: bhandaritushar1390@gmail.com (T. Bhandari), fursan.hamad@igs.uni-stuttgart.de (F. Hamad), christian.moormann@igs. uni-stuttgart.de (C. Moormann), kgsharmaiitd@gmail.com (K.G. Sharma), bmwest@gmx.de (B. Westrich). 1 Present address: SMEC India Pvt. Ltd., 5th Floor, Bldg. 8, Tower C, DLF Cyber City Phase II, Gurgaon, Haryana, India. http://dx.doi.org/10.1016/j.compgeo.2016.01.017 0266-352X/Ó 2016 Elsevier Ltd. All rights reserved. significant damage to human life and engineering structures that come in the way of the flow. Slope failures and landslides, in particular, can be triggered by many factors including heavy rainfall, imposed loads, strength degradation due to weathering and seismic excitation. Among these factors, seismic excitation or earthquake has been recognised to be the major causes of slope failures [1]. Therefore it becomes more important to analyse the response of a slope to a seismic event. Many methods have been developed to address this issue. Jibson [2] classifies the methods for assessing the performance of a slope during earthquakes fall into three phases: (1) pseudo static analysis, (2) permanent displacement analysis, and (3) stress–deformation analysis. Pseudo static analysis, used as a preliminary analysis method can only indicate the safety against slope failure using a limit equilibrium method in which the seismic shaking is represented by a constant inertial force applied on a sliding mass. As a significant improvement to pseudo static analysis, permanent displacement analysis provides more quantitative measure to evaluate the performance of slopes during earthquakes. A common example of the permanent displacement analysis is the Newmark rigid-block analysis [3] where the permanent slope deformation induced by earthquakes is estimated by the permanent displacement of the rigid block sliding along the inclined plane under a base acceleration. Both pseudo static 127 T. Bhandari et al. / Computers and Geotechnics 75 (2016) 126–134 Notations q r g u a e X n mass density the Cauchy stress tensor gravitational acceleration vector displacement nodal displacement vector strain tensor volume unit normal vector analysis and permanent displacement analysis are based on highly simplified geometric and material models. However, these methods cannot be relied upon to evaluate earthquake-induced slope deformations under complex geological conditions. For this purpose, stress–deformation analysis is used which can account for complex soil behaviours (e.g., non-linear response to dynamic loading, strain softening and strain rate dependence of material strength) and geometric conditions. It follows the approach of computing stresses in a material and its response in form of deformations based on a defined constitutive relationship between stress and strain. A stress–deformation analysis is frequently performed using numerical methods like the Finite Element Method (FEM) or Finite Difference Method (FDM). However, the mesh based methods (e.g. FEM) have difficulties in modelling large deformations due to problems of mesh distortion and entanglement. As a result, stress–deformation analysis is currently limited to estimating relatively small seismicallyinduced slope deformations [2]. The drawback of these methods to deal with large deformations therefore considerably impedes their application in the analysis of earthquake-induced slope deformations. In order to overcome these drawbacks in stress–deformation analysis, various mesh-free methods have been proposed. The Material Point Method (MPM) is one of these methods, which has shown its applicability to model granular materials like soil in different geotechnical applications [4,5]. Previous MPM research has focused on modelling of collapsing slopes and landslides using strength degradation [6]. Andersen and Andersen [7] also studied collapse of slopes in which the slide is initiated by increasing the density of material, corresponding to the behaviour during heavy rainfall. MPM has also been applied to model failure of geotextile-reinforced slope [8]. Although MPM represents the continuum by material points, solution is performed on the computational mesh. Thus, imposing boundary conditions is not aligned with the material representation. A non-zero kinematic condition is introduced in this paper where an additional set of particles is incorporated to track the moving boundary. The MPM procedure is applied to simulate the seismic excitation and dynamic response of a slope. The seismic history is introduced to the MPM model via the rigid boundary condition introduced by Hamad et al. [9,10]. Also, the proposed simulation approach is tested to model a shaking table experiment and to compare the results with corresponding numerical simulation from Hiraoka et al. [11] using another numerical method called the Smooth Particle Hydrodynamics – SPH method. Finally, MPM is applied to simulate progression of a large-scale landslide during the 1999 Chi-Chi earthquake of Taiwan. In this paper, the effect of water is not considered. In many cases, water can be a triggering factor for landslides and may impact the simulation results significantly. However, the simulation of dry landslide can be very relevant in cases of dry debris flow and rock avalanches where the controlling factor is the seismic motion. M B F N C1 Vp mass matrix gradient of the shape function nodal force vector shape function matrix relaxation coefficient p-wave velocity horizontal component of velocity time vx t 2. Brief description of MPM MPM can be viewed as an extension of the classical finite element procedure, in which the continuum body is discretised by Lagrangian material points that can move through a fixed computational mesh as shown in Fig. 1. The momentum equation is solved on the computational mesh which provides a convenient means of calculating discrete derivatives. 2.1. Spatial discretisation We start with the conservation of linear momentum, which reads qu€ ¼ r r þ qg ð1Þ where r(x, t) is the Cauchy stress tensor at position x and time t, q(x, t) is the mass density, g is the gravitational acceleration vector, u(x, t) is the displacement with the superposed dot denoting differentiation with time. By taking the virtual displacement du as test function for a domain of volume X surrounded by boundary S, the weak form of the momentum equation can be written as Z X € dX ¼ duT q u Z X deT r dX þ Z X Z duT q g dX þ duT t dC ð2Þ Ct where t ¼ r n is the prescribed traction on boundary Ct, n is the outward unit normal and e is the strain tensor represented in vector form. The superscript T denotes the transpose. Similar to the standard finite element method, the value of a variable inside the element can be based on the nodal values and the nodal shape functions. Using these definitions and discretizing the momentum equation, it takes the form (e.g., [4]) €¼F Ma ð3Þ € the nodal acceleration where M is the consistent mass matrix, a vector, and (F = Fext–Fint) with Fext and Fint being the external and internal nodal force vectors, respectively. In practice, the lumped mass matrix is preferred over the consistent mass matrix. This simplifies the computations at the expense of introducing a slight amount of numerical dissipation [12]. Referring to Eq. (3), the internal force vector is given by Fint ¼ np X xp BT ðxp Þrp ð4Þ p¼1 where the quotient of the material point mass and density is the volume of the material point, xp = mp/qp and B is the gradient of the shape function, as also used in standard finite element method [4], rp is a vector containing the stress components at the material point p. The external nodal force vector is given by Fext ¼ Z np X mp NT ðxp Þg þ NT t dC p¼1 Ct ð5Þ 128 T. Bhandari et al. / Computers and Geotechnics 75 (2016) 126–134 corrected using the velocity of the coupled bodies following Coulomb friction. This solution scheme uses the concept of comparing the single and combined body velocities for a contact node and defining its behaviour accordingly. The algorithm is able to detect whether two bodies in contact are approaching or separating from each other. If the two bodies are separating, the algorithm allows free separation where each body moves according to its own equation of motion. For more details about the algorithm, the reader is referred to Ref. [15]. 2.4. Prescribed velocity in MPM Fig. 1. MPM representation of a continuum. where N is the global shape function matrix and np is the total number of material points. 2.2. Time integration The discretised momentum equation (Eq. (3)) needs to be solved for discrete time intervals. With the mass matrix being a diagonal matrix, the system of equations can be solved using the Euler-forward time integration scheme, i.e. €t ; a_ tþDt ¼ a_ t þ Dt a €t ¼ ½Mtl a 1 Ft ð6Þ where Dt is the current time increment, a_ t and a_ tþDt are the nodal velocities at time t and (t + Dt), respectively and Ml is the lumped mass matrix. The incremental nodal displacement is obtained by integrating the nodal velocity by the Euler-backward rule (see for e.g., Jassim et al. [13]) DatþDt ¼ Dt a_ tþDt ð7Þ and the positions of the particles are subsequently updated from xptþDt ¼ xtp þ Np D atþDt In traditional dynamic FEM, the prescribed velocity is defined over nodes. These nodes always define part of the Lagrangian body boundary. On the other hand, in MPM the continuum is defined by Lagrangian particles which might change position from one element to another. Hence, there is no defined interface surface where prescribed velocity are applied. Within the framework of MPM, prescribed velocity can be applied directly on the material points of a rigid body representing the moving boundary for simple one-dimensional problem. For applications with axial movement, part of the mesh can be displaced as a moving mesh having a prescribed value while the rest is stretched uniformly [16]. This becomes very complicated and inconvenient when applied to applications like imposing seismic motion for a slope problem. As an alternative, an additional set of particles (Fig. 2) is introduced which tracks the moving boundary by carrying the time-dependent boundary evolution [9]. Following the same methodology, prescribed velocities are applied as a boundary condition for the rigid wall and a contact is defined between the rigid wall and soil. ð8Þ where xtp and xtp+Dt are the particle positions at time t and (t + Dt) respectively. For the present MPM solution procedure, a slightly different algorithm has been adopted for updating the particles velocity following Sulsky et al. [14]. By solving the equation of motion for the nodes, the elements deform and the material points in the interior of the element move in proportion to the motion of the nodes, based on the nodal shape functions. The position of the material points is updated using a single-valued continuous velocity field and hence the interpenetration of material is precluded. This automatic feature of the algorithm allows simulations of no-slip contact between different bodies without the need for special interface tracking and contact algorithms. After getting the nodal velocities, the strain increment of a material point p is calculated. The constitutive model is applied at the material points to get the incremental strain, which allows direct evaluation and tracking of history-dependent variables. At the end of time step the material point variables are updated and a new cycle begins using the information carried by the material points to initialise nodal values on the computational mesh. Note that at this stage, a new computational mesh can be defined since all the state variables are carried by the material points. In practice, however, it is more efficient to use the original mesh. 2.3. Contact algorithm The frictional contact algorithm proposed by Bardenhagen et al. [15] is used in this paper. It can be seen as a predictor–corrector scheme formulated in explicit manner, in which the velocity is predicted from the solution of each body separately and then 2.4.1. Non-zero kinematic conditions In this paper, the non-zero kinematic condition is developed as shown in Fig. 2 where an additional set of particles is introduced, which tracks the moving boundary by carrying the timedependent boundary evolution. At the beginning of a time step, the velocity a_ p ðxp ; tÞ of the prescribed particle p is assigned. Next, the prescribed velocity must be mapped (via the shape functions) from the prescribed particles to the computational nodes, where the discrete equations are solved. Nodes belonging to the elements where the prescribed particles are located are then tagged to be boundary nodes. It should be appreciated that the thickness of the boundary corresponds to one computational element. The prescribed values are assigned directly at the boundary nodes. As an alternative, a weighted mapping procedure can be used, which is more consistent with the principles of MPM, with the nodal velocity a_ i of boundary node i being obtained from [9,10] P _ p Ni ðnp Þwp ap a_ i ¼ P p Ni ðnp Þwp ð9Þ where a_ p is the prescribed velocity of material point p, Ni(np) is the shape function of i being evaluated at p, and wp is a weighting function (e.g. mass or volume of p). The summations in this equation run over the number of prescribed particles. Depending on the location of the boundary particles, the number of the boundary nodes is updated constantly as well as their values from Eq. (9). 2.4.2. Validation case: prescribed velocity with contact To validate the procedure of introducing prescribed velocity particles involving contact, a (1 1 m) square with a unit weight of 10 kN/m3 supported by 45 prescribed particles with zero velocity is considered. After calculating the initial gravitational stresses, the layer of prescribed particles underneath is moved suddenly with a horizontal velocity of 2 m/s. Fig. 3 shows three 129 T. Bhandari et al. / Computers and Geotechnics 75 (2016) 126–134 Soil 1.0 m Frictional Contact 0.35 0.53 Wall Rigid Wall 0.2 11 (a) Two Bodies in Contact Prescribed Velocity Contact Interface 0.5 45o 2 2 (b) MPM Representation 0.9 Fig. 2. Prescribed velocity in MPM. Shaking Table Fig. 4. Experimental model (after Hiraoka et al. [11]). scenarios: if the standard MPM (top) is applied, the body travels together with the bottom; for the case where a rough contact is introduced (middle), the glue condition is broken if the bottom moves fast enough; finally, the absence of the frictional resistance in the smooth contact case (bottom) leads to the early separation of the two bodies as shown. 3. Numerical application In the geotechnical field, dynamic process of slope failures subjected to seismic loads is often investigated by means of physical modelling [17–19]. Slope failure under seismic excitation is implemented by a box filled with soil and mounted on a shaking table. These experiments play a vital role in the calibration of numerical models for similar applications. For assessing the performance of MPM to simulate seismic excitation in geo-mechanical problems, a shaking table experiment is considered here and the simulated response is compared with published results [11] based on Smooth Particle Hydrodynamics (SPH) method. The objective here is to test the proposed numerical scheme in MPM (including the non-zero kinematic condition) with a simple Mohr–Coulomb failure criterion for simulating a dynamic test in comparison with more established numerical methods like SPH. The shaking table experiment under consideration consists of a small-scale cut slope as shown in Fig. 4. A steel box is mounted on top of the shaking table. The soil slope model (0.9 0.6 0.5 m) was set in the shaking box, and the slope angle was made as 45°. The soil used in the experiment was Masa soil which is weathered granite commonly found in Kansai area in Japan. Laser displacement sensors were used to measure displacement within the slope. The slope model was subjected to the seismic wave loading shown in Fig. 5. The test runs for 14 s until the slope completely collapsed. More details about the test setup and the experiment can be found in Ref. [11]. 3.1. Reference solutions In order to test the proposed treatment of boundary conditions in MPM, a comparison with other numerical methods namely the Finite Element Method (FEM) and the Smooth Particle Hydrodynamics (SPH) method is provided in this paper. The FEM model used in this research is suitable for the failure initiation where small deformation theory is applicable. On the other hand, the SPH model is more appropriate for the large deformation analysis. Fig. 3. Prescribed velocity boundary with contact: (top) standard MPM, (middle) rough contact, and (bottom) smooth contact. 130 T. Bhandari et al. / Computers and Geotechnics 75 (2016) 126–134 Velocity, vx (m/s) 0.15 0.10 0.05 0.00 -0.05 -0.10 -0.15 0 2 4 6 8 10 12 14 Time (s) Fig. 5. Velocity–time history of the shaking table (after Hiraoka et al. [11]). [%] 1.5 3.1.1. FEM model To set a benchmark for the proposed MPM simulation approach, dynamic analysis for the slope is carried out using the FEM software, PLAXIS 2D. Keeping in mind that the FE software does not count for large deformations, the FE simulation is performed up to 10 s of seismic excitation. The soil is modelled using the Mohr–Coulomb failure criterion. Soil properties are considered as given by Hiraoka et al. [11] for SPH model. The density of the soil is 1680 kg/m3 with Young’s modulus of 2570 kPa and Poisson’s ratio of 0.33. The stiffness parameters are chosen upon the construction process of the slope, which is built in phases of 5 cm thick compacted layers. The shear strength of the soil is defined by cohesion of 0.78 kPa and friction angle of 23°. A non-associated constitutive model is adopted with zero dilatancy angle. The problem is solved in two phases, first for gravity and then the dynamic phase. In PLAXIS, different boundary conditions are used for solving the static and dynamic phase. During the static phase, roller boundary at side and fixed boundary at bottom are considered for the calculation of gravity stresses. These boundary conditions are applied by introducing prescribed boundary displacements. The boundary fixities are removed during the transition from static to dynamic phase which means that the displacement boundary conditions are replaced with another set of boundary conditions defined for the dynamic phase. By removing these boundary fixities, the boundary starts moving as a result of initial stresses. To prevent this, the original boundary stress is converted to an initial (virtual) boundary velocity. When calculating the stress in the dynamic phase, the initial boundary velocity is subtracted from the real velocity: rn ¼ C 1 qV p ðu_ n u_ on Þ ð10Þ where rn is the normal stress on boundary, C1 is relaxation coefficient, q is the density of material, Vp is the p-wave velocity, u_ n is real velocity and u_ on is the initial velocity. Prescribed velocities at side as well as bottom are used to induce the seismic motion as per the velocity–time history (Fig. 5). The deformed mesh to true scale and shear strain plot are shown in Fig. 6. The scale for the shear strain plots is set to be same as in Ref. [11] to facilitate comparison of results. Large deformation is clearly present at t = 10 s as seen in Fig. 6 and therefore finite element analysis beyond this deformation is not considered. 3.1.2. SPH model A 2D-SPH (Smooth Particle Hydrodynamics) model was used by Hiraoka et al. [11] to simulate the dynamic behaviour of the slope model. Soil parameters used for the elastic–plastic Drucker–Prager constitutive model are the same as given in Section 3.1.1. Boundary conditions were free-roller at the vertical and full-fixity at the base. A total of 3245 particles were used to create the slope model. Details of the modelling framework and simulation approach for SPH can be found in Ref. [11]. 0 Fig. 6. Finite element analysis at t = 10 s: (top) deformed mesh and (bottom) shear strain. 3.2. Simulation approach for seismic slope modelling in MPM MPM is a relatively new method for geotechnical applications and hence the suitability of the method to simulate specific geotechnical problems (like the considered ones) has not been verified. Since, MPM differs from FEM and SPH in its numerical framework, similar simulation approach cannot be applied for such applications. For instance, unlike FEM, prescribed velocities cannot be applied directly as a boundary condition to the soil particles. Therefore, in this paper, an approach is proposed to deal with such applications in MPM as discussed in the subsequent sections. 3.2.1. Boundary conditions In order to model the shaking table problem, rigid particles are defined to play the role of velocity carriers as shown in Fig. 7. The rigid wall particles can be related to the steel box as generally used in the experimental setup. The soil is modelled using the Mohr– Coulomb constitutive law. Furthermore, two different types of contacts are defined between the soil and the walls. A smooth contact is provided for side wall to allow free settlement of soil along the glass wall in vertical direction and a rough contact is provided for the bottom wall. 3.2.2. Calculation phases As in most of the numerical methods, the analysis is done in different steps. The model is first solved for gravitational stresses. The rigid walls are assigned zero velocity and the stresses are built-up under the self-weight of soil. A quasi static solution is obtained with the use of local damping. Local damping is applied by assuming that the damping force F damp is proportional to unbalanced forces [16]. F damp ¼ ajF ext F int jsignðv Þ ð11Þ where a being the damping coefficient used as 0.7 for the gravity phase, and the sign of velocity at the degree of freedom (i) is defined as signðv Þ ¼ v i =jv i j. Eq. (11) is added to the right hand side of Eq. (3). To check that equilibrium is achieved, the kinetic energy and the out of balance forces are evaluated at all computational nodes and compared to a tolerance, which is 0.005 for the present cases. 131 T. Bhandari et al. / Computers and Geotechnics 75 (2016) 126–134 Smooth contact µ=0 Rigid Particle Rigid Wall M-C Soil M-C Soil Rough contact µ=1 Fig. 8. Initial configuration of the MPM model. [%] 1.5 vx(t)= prescribed velocity (Fig. 5) vy(t)= 0 Fig. 7. Dynamic boundary conditions in MPM. After obtaining an equilibrium state of gravity stresses, next step for calculation is defined for dynamic analysis. The local damping is switched off and prescribed velocities (Fig. 5) are assigned to the rigid walls which simulates dynamic motion of the steel box and consequent behaviour of the soil it contains taking into account the external forces, body forces and the inertial forces. 3.2.3. Simulated results To model the seismic excitation of the soil slope in MPM, the simulation approach as discussed in Sections 3.2.1 and 3.2.2 is used. Mesh and particle discretisation are illustrated in Fig. 8. Shear strain at t = 10 s and total displacement at t = 14 s are shown in Fig. 9. The dynamic response of the slope can be described in the form of displacement history of specific points on the slope. The displacement histories of specific control points 1 and 2 as outlined in Fig. 4 are used to compare the experimental results with the dynamic response predicted by different numerical models. The comparison for the vertical displacement history for point 1 and horizontal displacement history for point 2 are shown in Fig. 10. It is evident that even with limited time i.e. t = 10 s, FEM in unable to predict the trend of deformations towards failure. The deformation behaviour of the slope as predicted by MPM is in fair comparison with that predicted by SPH and also close to the experimental results. Sensitivity of the analysis results to mesh discretization was also assessed by solving three cases for different mesh sizes. The details of these cases are presented in Appendix A. It is noted that like other mesh based methods, MPM results are also dependent on the size of mesh. With a variation of ±60% in the mesh size, the maximum total displacement varies up to around ±7%. In the present analyses, the shape of the sliding surface is predicted as almost circular, which is in agreement with the other continuum-based model using SPH. However, the sliding surface as observed from the experiment was a curved line with higher curvature angle. Hiraoka et al. [11] suggest a lack of clarity about this failure mechanism observed in the experiment and attribute it to possible technical errors while removing the collapsing soil to specify the failure surface in the experiments. The SPH analysis [11] also suggests to assign a non-zero dilatancy angle (w = //2 and /) in order to come closer to the experiment. Although this assumption improves the prediction of the failure surface, it over-predicts the plastic volumetric expansion like the soil is heavily compacted which contradicts the initial state of the considered 0 [m] 0.21 0 Fig. 9. MPM analysis: (top) shear strain at t = 10 s and (bottom) norm of total displacement at t = 14 s. soil. Consequently, large run-out distance is observed in the numerical model. Therefore, these analyses are excluded from this paper. Advanced constitutive models that consider the effect of soil degradation with the stress and density evolutions are supposed to perform better than the simple elasto-plastic Mohr–Coulomb model being used. In principle, the implementation of these models in MPM is straightforward, while the related numerical stability problems are under study and the formulation under development. Considering that the soil in the experiment is reported to have 10% water content, the advanced constitutive model should be combined with partially saturated soil model for better simulation of the progressive failure of the experiment. 4. Earthquake induced landslide debris flow The dynamic process of landslides induced by earthquakes is very complex in its nature. Various numerical methods are used to simulate different activities involved in the whole process of evolution of a landslide starting from the development of a critical sliding surface to initiation and triggering of failure to disintegration of the sliding mass to debris flow and finally deposition. In this section, MPM is used to simulate the last part of the process i.e. progression of the sliding mass and deposition. For this, the Chiu-fen-erh-shan landslide of 1999 is used as a case study example. 132 T. Bhandari et al. / Computers and Geotechnics 75 (2016) 126–134 Time (s) 2 4 6 Pre-slide Topography 8 10 12 Post-slide Topigraphy 14 0 -25 500 (m) -50 -75 Fig. 11. Topography of landslide (after Wu et al. [23]). -100 -125 -150 Experimental SPH PLAXIS MPM 1.0 Velocity, vx (m/s) Horizontal Displacement (mm) Vertical Displacement (mm) 0 200 150 0.8 0.6 0.4 0.2 0.0 -0.2 -0.4 100 0 20 40 60 80 100 Time (s) 50 0 Fig. 12. Time history of the horizontal velocity (after Wu et al. [23]). 0 2 4 6 8 10 12 14 Time (s) Experimental PLAXIS SPH MPM Fig. 10. Displacement history of the control points: (top) Point 1 and (bottom) Point 2. 4.1. The Chiu-fen-erh-shan landslide The Chiu-fen-erh-shan landslide is one of the major landslides caused by the disastrous Mw 7.6 Chi-Chi earthquake of 1999 in Taiwan. The landslide has been characterised as a translational rockblock slide on dip slope and known to be disintegrated into fragmented rock avalanche travelling long distance at high velocity [20]. The landslide debris travelled a distance of more than 1 km and covered an area of 1.95 km2. Many studies have been conducted since then to understand the mechanism and movement of the landslide [20–23]. These studies concluded that the slope had undergone gravitational creep and was unstable even prior to the earthquake. A slip surface had been developed as a result of flexural folding. Observations by Wang et al. [22] suggested the presence of clay seams between the alternating beds of shale and sandstone in the slope. The existence of the clay seam provides a very smooth surface for the slide to occur. The slope was retained in its position because of a sandstone bed that formed resistant ridges at the foot of the slope. The sandstone bed was most probably damaged seriously during the earthquake giving way to the massive landslide. The topography of the slope before and after the slope is depicted in Fig. 11. Detailed geological description of the landslide is available in Ref. [22]. 4.2. Continuum modelling of the rock-block slide According to Wu et al. [23], the mechanical properties of the rock in sliding mass, composed of highly jointed rock mass, are: Fig. 13. Progression of landslide. density of 2550 kg/m3, Young’s modulus of 7.57 GPa, Poisson’s ratio of 0.19, whereas the strength parameters of the weak planes of rock are adopted from the direct shear experimental study conducted by Chen [24]. These strength parameters of the rock joints are considered to mainly control the strength behaviour of the 133 T. Bhandari et al. / Computers and Geotechnics 75 (2016) 126–134 sliding mass as movement is likely to occur along the existing weak planes in the sliding mass and thus these parameters are used directly to define the shear strength of the continuum. The values adopted for the cohesion and friction are 20 kPa and 24° respectively. In the present numerical scheme, the aim is to model the rock joint masses as a continuum material. Therefore, the elastic modulus of the equivalent medium was estimated to be 500 MPa with a geological strength index range of 25–30 [25]. [%] 1.5 0 4.3. Material point analysis of the landslide [%] For the seismic motion, horizontal component of the velocity– time history of the Chi-Chi earthquake as shown in Fig. 12 was used. This history was used by Wu et al. [23] in their discrete analysis using velocities projected to the slope direction and corrected for baseline correction. The sliding surface in MPM is modelled using rigid particles and the velocities are imposed on these particles. 1.5 4.4. Simulated results 0 The progression of the flow resulting from the landslide is depicted in Fig. 13. The progressive displacement of the front of [%] 1.5 Table A.1 Discretization details for cases of mesh sensitivity analysis. S. no. Item Case-1 Case-2 Case-3 1. 2. 3. 4. 101,100 16,174 32,827 1.05 103 m 47,640 7503 15,336 1.89 103 m 24,540 4029 8315 2.97 103 m Number of particles Number of elements Number of nodes Average element size 0 Fig. A.2. Shear strain at t = 10 s for cases 1, 2 and 3 (top to bottom). the sliding mass is marked for different time steps. It can be seen from the stabilizing displacement profile that the sliding mass reaches an equilibrium state in 120 s. The final configuration as simulated by MPM is compared with the actual scenario in Fig. 13 which suggests a fair match between the shape of the debris deposit except for a slightly different shape at the rare end of the debris flow. 5. Conclusion and outlook Fig. A.1. Mesh and particle discretization for cases 1, 2 and 3 (top to bottom). An important geotechnical application has been modelled using the MPM. Simulation approach to induce seismic motion in MPM is tested by modelling a shaking table experiment and comparing the results with other numerical methods. The deformation behaviour of the slope as predicted by MPM is in fair comparison with that predicted by SPH. A variation of the order of 5% in results is observed and attributed to the fact that both MPM and SPH differ in their basic formulation. Also, the failure criteria and implementation of boundary conditions used in both the methods are different. Compared to SPH, MPM is seen to predict the dynamic response of the slope closer to the experimental results. However, better simulation of progressive failure can be achieved by incorporating advanced constitutive models in the current formulation of MPM. The limitation of conventional FEM to model large deformation problems is also emphasised by comparing the results with FEM simulation using PLAXIS. While this limitation can largely be overcome by Lagrangian–Eulerian methods, remeshing may still be required and state variables associated with material points need to be remapped. This is of particular concern when the history of the material must be taken into account. 134 T. Bhandari et al. / Computers and Geotechnics 75 (2016) 126–134 [m] Appendix A 0.23 To assess the sensitivity of the model to mesh discretization, three cases for different mesh sizes were solved. The discretization details of these cases are presented in Table A.1. The mesh for the three cases is shown in Fig. A.1. Analysis results in form of contour plots for shear strain at t = 10 s and total displacement at 14 s are presented in Figs. A.2 and A.3 respectively for the three cases. 0 References [m] 0.21 0 [m] 0.20 0 Fig. A.3. Total displacement at t = 14 s for cases 1, 2 and 3 (top to bottom). Finally, the potential of the MPM to model extensive deformation in form of landslide debris-flow is demonstrated. MPM is seen to simulate the progression of the landslide and generate a reasonable post-failure configuration. The proposed simulation approach holds well for the post failure scenario but detailed study is required to understand the modelling of complete evolution process of landslides which may include a better contact algorithm to model the brittle behaviour of rock joints and precise equivalent continuum modelling of rock mass. Moreover, the present approach can be extended with further studies to include the effect of water and improve constitutive modelling to take into account the behaviour of rock in dynamic conditions. Acknowledgements The authors acknowledge the help of ‘‘German Academic Exchange Service (DAAD)” and ‘‘Institute of Geotechnical Engineering (IGS), Stuttgart” for providing the financial and physical resources required to carry out this research. We would also like to acknowledge ‘‘Deltares, The Netherlands” for providing access to their MPM source code, which was further developed in this paper. [1] Keefer DK. Landslides caused by earthquakes. Geol Soc Am Bull 1984;95 (4):406–21. [2] Jibson RW. Methods for assessing the stability of slopes during earthquakes – a retrospective. Eng Geol 2011;122(1):43–50. [3] Newmark NM. Effects of earthquakes on dams and embankments. Geotechnique 1965;15(2):129–60. [4] Wie˛ckowski Z, Youn S-K, Yeon J-H. A particle-in-cell solution to the silo discharging problem. Int J Numer Meth Eng 1999;45(9):1203–25. [5] Coetzee C, Vermeer P, Basson A. The modelling of anchors using the material point method. Int J Numer Anal Meth Geomech 2005;29(9):879–95. [6] Andersen S, Andersen L. Modelling of landslides with the material-point method. Comput Geosci 2010;14(1):137–47. [7] Andersen S, Andersen L. Material-point-method analysis of collapsing slopes. In: Proceedings of the 1st international symposium on computational geomechanics (ComGeo I), Juan-les-Pins, France; 2009. p. 817–28. [8] Hamad F, Vermeer P, Moormann C. Failure of a geotextile-reinforced embankment using the material point method. In: Proceedings of the 3rd international conference on particle-based methods-fundamentals and applications, Stuttgart, Germany; 2013. p. 498–509. [9] Hamad F, Vermeer P, Moormann C. Development of a coupled FEM-MPM approach to model a 3D membrane with an application of releasing geocontainer from barge. In: Proceedings of the 3rd international conference on installation effects in geotechnical engineering, Rotterdam, The Netherlands; 2013. p. 176–83. [10] Hamad F, Stolle D, Moormann C. Material point modelling of releasing geocontainers from a barge. J Geotext Geomembranes 2016;44(3):308–18. [11] Hiraoka N, Oya A, Bui HH, Rajeev P, Fukagawa R. Seismic slope failure modelling using the Mesh-free SPH method. Int J GEOMATE 2013;5:660–5. [12] Burgess D, Sulsky D, Brackbill J. Mass matrix formulation of the FLIP particlein-cell method. J Comput Phys 1992;103(1):1–15. [13] Jassim I, Stolle D, Vermeer P. Two-phase dynamic analysis by material point method. Int J Numer Anal Meth Geomech 2013;37(15):2502–22. [14] Sulsky D, Zhou S-J, Schreyer HL. Application of a particle-in-cell method to solid mechanics. Comput Phys Commun 1995;87(1):236–52. [15] Bardenhagen S, Brackbill J, Sulsky D. The material-point method for granular materials. Comput Methods Appl Mech Eng 2000;187(3):529–41. [16] Jassim I, Hamad F, Vermeer P. Dynamic material point method with applications in geomechanics. In: Proceedings of the 2nd international symposium on computational geomechanics (COMGEO II), CavtatDubrovnik, Croatia; 2011. p. 445–6. [17] Kutter BL. Earthquake deformation of centrifuge model banks. J Geotech Eng 1984;110(12):1697–714. [18] Arulanandan K, Yogachandran C, Muraleetharan K, Kutter B, Chang G. Seismically induced flow slide on centrifuge. J Geotech Eng 1988;114 (12):1442–9. [19] Wartman J, Seed RB, Bray JD. Shaking table modeling of seismically induced deformations in slopes. J Geotech Geoenviron Eng 2005;131(5):610–22. [20] Huang C-C, Lee Y-H, Liu H-P, Keefer DK, Jibson RW. Influence of surface-normal ground acceleration on the initiation of the Jih-Feng-Erh-Shan landslide during the 1999 Chi-Chi, Taiwan, earthquake. Bull Seismol Soc Am 2001;91(5):953–8. [21] Hung J-J. Chi-Chi earthquake induced landslides in Taiwan. Earthq Eng Eng Seismol 2000;2(2):25–33. [22] Wang W-N, Chigira M, Furuya T. Geological and geomorphological precursors of the Chiu-fen-erh-shan landslide triggered by the Chi-chi earthquake in central Taiwan. Eng Geol 2003;69(1):1–13. [23] Wu J-H, Lin J-S, Chen C-S. Dynamic discrete analysis of an earthquake-induced large-scale landslide. Int J Rock Mech Min Sci 2009;46(2):397–407. [24] Chen H. Engineering geological characteristics of Taiwan landslides. SinoGeotechnics 2000;79:59–70. [25] Hoek E, Diederichs MS. Empirical estimation of rock mass modulus. Int J Rock Mech Min Sci 2006;43(2):203–15.