基于OpenMP和Pardiso的柔性多体系统动力学并行计算

Parallel computing studies of flexible multibody system dynamics using OpenMP and Pardiso

为加快大型、复杂柔性多体系统的动力学仿真的速度,对多体系统动力学的并行算法进行研究。首先分析了微分代数方程(differential algebraic equations,DAEs)在数值计算求解过程中主要的计算量。据此,提出采用OpenMP并行计算系统的刚度矩阵、右端项和采用并行的稀疏线性方程组求解器Pardiso对线性方程组进行求解的并行策略。将这两种并行策略应用到自主开发的柔性多体系统动力学软件THUSolver中,实现了对多体系统动力学的并行计算。通过两个工程算例的仿真得到并行的加速比和计算效率,结果表明:采用的两种并行策略都有很高的计算效率,能大幅提高多体系统动力学仿真的速度。

Parallel algorithms were developed for multibody dynamic systems to accelerate the speed of dynamic simulations of large complex flexible multibody systems. The main computational loads during the numerical solution of the differential algebraic equations （DAEs） were analyzed with OpenMP then used to compute the system tangent matrix and residual vector in parallel and with a parallel sparse linear equation solver, Pardiso, used for the linear equations. Both of the parallel strategies were implemented in an in-house multibody algorithm for parallel analyses of multibody dynamic systems. The parallel speedup and the computational efficiency of the algorithm are quite good so the simulation speeds of multibody dynamic analyses are increased substantially.