Acceleration of the FEM computations in electromagnetic analysis for switched reluctance motor

Abstract

In this paper an application of parallel computing to accelerate modeling of subsystem in the frame of HiL (Hardware-in-the-Loop) platform is presented. The subsystem is the model of SRM (Switched Reluctance Motor), analyzed with the use of numerical method FEM (Finite Element Method). The paper focuses on potential of the GPU (Graphical Processing Unit) Tesla cards, which are made in a Fermi and Kepler technology to optimize computation time. The FEM is analyzed and possible accelerations of this method are emphasized. A bottleneck of the FEM is a necessity to solve LSE (Linear System of Equations), with very large and sparse matrices, which are results of PDEs (Partial Differential Equations) discretization process. This paper examines parallelism of the sparse matrices solvers on the high-end GPUs. Variations of the iterative CG (Conjugate Gradient) solver were analyzed, due to the CG is the iterative solver used in FEMM, freeware software package for 2D FEA (Finite Element Analysis) of EM (Electro-Mechanical) systems. The Conjugate-Gradient without preconditioner, CG with Jacobi preconditioner, CG with ilu0 preconditioner were analyzed. In parallel computations CULA tools routines for iterative solvers were used.