High-performance Computing in modeling of Landslide Post-failure Stage using Material Point Method

The post-failure of landslide is a stage where large deformations are present. It is difficult to properly resolve such large deformations using traditional mesh-based numerical methods. Meshless methods, such as the material point method (MPM), can resolve such problems by reducing the dependence on the mesh. However, the time-consuming mapping procedure between the material points and background nodes exists at each time step of MPM, consequently, one needs an efficient implementation taking advantage of modern computer hardware architectures for a high-resolution computational model. In the present study, we develop a high-performance MPM simulation package using Julia language to simulate the landslide post-failure stage. We show both the 2D and 3D computation models. The parallel algorithm on the GPU version is based on the features of MPM through CUDA.jl, a library that natively supports CUDA computing in Julia. To validate the performance of the present simulation package, we perform benchmarks on both CPU and GPU versions of the package. Furthermore, we use the uniform Generalized Interpolation MPM (uGIMP) and apply it to resolve a real problem to demonstrate the capabilities of this package.  The simulation result is in good agreement with the ground truth. HPC simulation is not only reproducing the run-out process but also provides us with a better understanding of the complex mechanisms involved in landslide movements.