Development Of Multi-Core Computational Modules For Structural Topology Optimization Based On SiPESC

In the past decades,topology optimization theory has made extensive development and progress.At the same time,human production and manufacturing capacity has been rapidly developed.With the rapid development of production capacity,the design of topology optimization method proposed the design requirements of higher precision and efficiency,from the attention of academia and industry,3D topology optimization design method has been a hot spot.In order to promote the topology optimization of better service to the engineering application,attention should be paid to improve the computational efficiency of topology optimization for large scale problems.At present SiPESC-a CAE platform which is independently developed by Dalian university of technology is still a serial topology optimization module,and has limitations for large-scale topological optimization problems such as time consuming,low calculation efficiency.These limitations seriously hinder its development.Based on the existing serial topology optimization module,parallel topological optimization modules are developed by the employment of multi-core techniques.The efforts improve SiPESC platform computational efficiency for structural topology optimization,and in the future will help to solve large-scale topological optimization problems.Based on SiPESC,this paper develops the parallel topology optimization module by using OpenMP parallel processing technology.OpenMP is a parallel processing tool based on the Shared memory parallel system,and adopts the fork-join execution mode,including the compilation guidance,API function and environment variables.In order to find the performance bottleneck of serial program,Intel VTune is used to analyse the serial topology optimization modules for testing examples.Attention is paid to the analysis of the proposed topology optimization of the overall design idea,and then to design variable management module,structural response sensitivity calculation module,updating and normalized modules.Special parallelization is performed for the static and dynamic topology optimization computation including updating modules and the corresponding response sensitivity calculation.The methods of parallelization include direct parallelization based on dataparallel,task-parallel and nested-parallel.For the loop scheduling strategy in parallel process,dynamic scheduling is adopted to improve load balancing.In the parallel process,in order to achieve consistent parallel results,it is necessary to pay special attention to the problems of data competition and thread synchronization.Numerical examples shows that the results of parallel topology optimization are exactly the same as those by serial topology optimization,and can achieve a good acceleration ratio.