Three-dimensional deformable distinct element method with polyhedral elements and cloud GPGPU acceleration

Abstract

Discontinuous deformation analysis (DDA), as a kind of deformable distinct element method (DDEM), features rigorous formulations but bears the Achilles’ heel of low efficiency. In this study, the explicit discontinuous deformation analysis (EDDA) is adapted for efficient large-scale computation. This EDDA/DDEM takes linear elastic polyhedral blocks as basic elements. The cell-mapping neighbor search and high-fidelity direct search are utilized to identify potential contacts. Explicit contact forces decouple the global equation into independent block-wise ones. Moreover, the data structures are carefully designed to reduce memory usage and enhance cache locality. GPU acceleration is exploited, endowing EDDA/DDEM the efficiency that is comparable with the state-of-art parallel distinct element method using rigid elements. In this study, the batch computation on a cloud GPU frees the user’s computer after launching the large-scale computation. The performance of EDDA/DDEM is explored using several large-scale tests of masonry walls and densely packed blocks. A test with over a million blocks is run for 100 thousand steps and only takes 2.5 days, proving the capability of the newly developed EDDA/DDEM to handle engineering analysis.