Development Of Explicit Finite Element Multi-GPU Parallel Simulation System And Its Application In Car Body Structure Design

The design of automobile body structure is a key part in the development process of automobile,and the use of CAE finite element simulation software for auxiliary calculation is an important means to reduce development cost.Most of the simulation software used in China’s automobile industry is foreign.The autonomous simulation software has some problems,such as low computational efficiency,limited computational scale by hardware and imperfect post-processing flow.Therefore,it is very important to develop an autonomous finite element simulation system with high efficiency,complete process and the ability to deal with large-scale problems.Based on this,this paper combined the GPU parallel computing method in the finite element simulation,and made use of the high concurrency of GPU to accelerate the finite element calculation.Moreover,the multi-GPU parallel computing technology is used to deal with the problem of super-large scale and further improve the computational efficiency of the simulation system.At the same time,we also developed the postprocessing module based on the finite element simulation system,which can efficiently analyze the post-processing data through the second extraction of the calculated data,which is very suitable for the enterprise to process a large number of computing models.Specific work contents and achievements are as follows:(1)The multi-GPU parallel computation of the whole process of the shell element is realized,and the multi-GPU parallel computation model for displaying the finite element analysis is established to realize the development of the multi-GPU solver of the system.Firstly,the Metis open source software was used to partition the data of the finite element model,which ensured the computing load balance of each GPU and minimized the boundary points between each partition.Then,a preprocessing algorithm of partition nodes based on counting sort is developed,and the boundary points of each partition are found with linear complexity,which successfully realizes the data basis of multi-GPU parallel.The location mapping array is established to link the units and nodes in different partitions.The boundary point partition numbering queue is introduced,and the boundary point transmission strategy is established to realize the corresponding transmission of the common boundary nodes in different partitions.(2)The boundary point processing strategy in the parallel algorithm of the multiGPU solver is improved,and the transmission time of different GPU boundary point data is superposed by the same transmission time,which further improves the data transmission efficiency in the parallel process of multi-GPU.Secondly,by adjusting the number of registers and local memory for many times,the efficiency of node coordinate displacement update function,acceleration update function and cell calculation function is optimized.In the final test example,the optimization effect of node displacement acceleration update function is the best,reaching more than 40%.(3)The post-processing framework of multi-GPU parallel simulation system is developed.Firstly,the total output management class is created object-oriented,which is used to manage five types of output objects,such as unit and node.And by adding three methods of memory management,data extraction and data output for each output object,the whole output framework realizes memory alignment during post-processing data extraction in GPU,which greatly improves the speed of post-processing data extraction from multi-GPU memory.At the same time,it is found through testing that the output speed of binary format is significantly higher than that of text format,so the output format of post-processing file module is designed based on binary,and a fast extraction method based on index file is designed.After testing,the specified postprocessing data can be extracted accurately and the post-processing analysis can be realized quickly.(4)The interface of the simulation system is developed based on the Qt environment,which improves the usability and integration of the system.It mainly includes initialization interface,material interface,solution interface and postprocessing interface.Finally,the system was calculated and verified through multiple body models,among which the largest example reached 12 million degrees of freedom.The results show that the accuracy of the solution results of the system can meet the engineering requirements,and the post-processing module can accurately extract data for analysis.Moreover,in the test examples,the absolute acceleration ratio of the dual GPU solver is more than 27 times,and the relative acceleration ratio of the single GPU solver is more than 1.9 times,which verifies the efficient solving performance of the multi-GPU simulation system developed in this paper,and can be used to deal with large-scale engineering problems.