Three-dimensional modeling of dynamic interaction between soils and multibody system with GPU acceleration

Ling Zhan,Chong Peng,Bingyin Zhang

2019-01-01

Abstract:Dynamic interaction between soils and multibody system widely presents in geotechnical problems. For instance, in debris flows, soil-water mixtures carry large amount of rocks and flow in high speed. The rocks have the size several orders of magnitude bigger than that of the soil particles and can lead to severe bed erosion and damage to structures. There is a tremendous need to understand the process of dynamic interaction between soils and multibody systems, and provide a practical tool for safety assessment and hazard mitigation. In contrast of costly laboratory and in-situ tests, numerical simulations can be carried out conveniently and can give more details about the dynamic interaction and the motion of different objects. However, this process involves large deformation, moving interface, moving boundary with complex geometry and prohibitively heavy computational burdens, which requires further investigation for numerical simulations. A GPU-accelerated massively parallel numerical framework for three-dimensional simulation of soil-multibody interaction in geotechnical engineering based on smoothed particle hydrodynamics (SPH) is presented in this study. The geotechnical material is simulated using conventional Eulerian kernel based SPH with the Dracker-Prager model, while the solid structures are treated as rigid bodies and discretized with SPH particles. The interactions between the soil and rigid bodies, and among the rigid bodies, are calculated using a particle-to-particle contact algorithm based on ideal frictional contact. A kernel gradient correction technique is employed to alleviate the particle deficiency near the contact boundary. Furthermore, the advanced GPU-acceleration is employed to achieve high efficiency in three-dimensional simulation. The computationally extensive operations, including neighboring search, friction force calculation, particle interaction and time evolution of soil and multibody system are all executed on GPU. Owing to the acceleration provided by GPU, complex multibody dynamic problems can be solved within a rather short calculation time. The numerical results are compared with experimental data. The good agreement indicates that the presented framework can have a wider application to problems involving dynamic soil-multibody interaction. Figure 1: (a) The initial configuration of the system: Retaining wall collapse simulation. Dimensions are in cm; (b) Comparison of the final failure pattern between the experiment [1] and the SPH simulation.