A hybrid parallel framework for computational solid mechanics

Abstract

A novel, hybrid parallel C++ framework for computational solid mechanics is developed and presented. The modular and extensible design of this framework allows it to support a wide variety of numerical schemes including discontinuous Galerkin formulations and higher order methods, multiphysics problems, hybrid meshes made of different types of elements and a number of different linear and non-linear solvers. In addition, native, seamless support is included for hardware acceleration by Graphics Processing Units (GPUs) via NVIDIA's CUDA architecture for both single GPU workstations and heterogenous clusters of GPUs. The capabilities of the framework are demonstrated through a series of sample problems, including a laser induced cylindrical shock propagation, a dynamic problem involving a micro-truss array made of millions of elements, and a tension problem involving a shape memory alloy with a multifield formulation to model the superelastic effect.