The Parallel Algorithm For Multibody System Dynamics Based On The Cluster Platform

With the development of space technology,the size of the spatial structure is constantly increasing,and its structural function tends to be complicated.Using numerical simulation technology to predict the dynamic behavior of large-scale space structures is one of the effective methods for the design and optimization of such structures.However,accurately predicting the dynamics of large-scale space structures involves long-term simulations with many degrees of freedom.Therefore,a single computer performance has been unable to meet the demand of dynamic simulation,which brings opportunities and challenges to the development of the parallel simulation technology.As computer technology continues to evolve,Supercomputers are gradually transformed into loosely coupled computer cluster systems.As a multi-node architecture,the parallel cluster system has scalability,availability,reliability and other advantages.Currently,the parallel cluster system has been widely used in various fields such as defense,finance and scientific research,etc.However,there is no mature literature report on the parallel algorithm based on the cluster system in the field of multibody dynamics.In this dissertation,based on the large-scale space structures,the parallel algorithm for the large-scale multibody system dynamics and the parallel cluster system are studied,and the main research contents are as follows:1.In order to make the parallel algorithm achieve the ideal speedup,the graph theory is employed to represent the flexible multibody system.And,based on the graph partition algorithm,the flexible multibody system is decomposed into several subsystems with the same computational overhead,so that each thread load is balanced;The advantages and disadvantages of several documentary graph partition algorithms are analyzed and compared;In order to decompose the large-scale flexible multi-body system efficiently,a multilevel partitioning method is exploited to assign computational tasks,and several numerical examples are given to validate the effectiveness of the proposed algorithm.2.In order to solve the equation of motion for multi-body systems in parallel,a parallel system based on cluster is designed and constructed;In the Linux environment,NFS shared file system,network interconnection system and Open MPI-ifort simulation system are established;The network stability and validity of the cluster system are tested by using the network testing tools Iperf and several parallel numerical examples based on MPI.3.Based on the Lagrange equations of the first kind,the basic kinetic equations of multibody dynamics are established.The finite element tearing and interconnecting(FETI)method is extended to solve the dynamics of flexible multibody system in parallel by the Schur complement method.The overall computation flow and the detailed iterative process are given;By using Natural Coordinate Formulation(NCF)and Absolute Nodal Coordinate Formulation(ANCF),a dynamic model of a rigid-flexible coupled crank slider is established and the dynamics simulation is preformed through the constructed cluster system.The consuming time of every step in the computational flow is counted and analyzed.The speedup,the complexity of algorithm and the parallel scalability are also studied.