Research For Sheet Metal Forming Algorithm Based On Heterogeneous Parallel Strategy

Auto-body panel stamping production is widely considered to be an important process in vehicle manufacturing. Numerical simulation of the Auto-body panel forming by using finite element method can improve the product design efficiency and reduce the production costs. Development of parallel computing provides a faster solving method for this field. Focusing on dynamic explicit algorithm for sheet metal forming, we have analyzed different parallel computing methods, and proposed a parallel co-algorithm based on task partition strategy on heterogeneous platform, which improves the computation performance significantly.KMAS is a self-owned intellectual property sheet metal forming software. we reconstruct the KMAS/Incremental module and make a parallel computing improvement. Compared to the original Fortran language used in the program, C language is of better universality, and could support parallel frameworks better. Firstly, by using combining FORTRAN and C programming, we translate the FORTRAN language version into a C language program. Focusing on data structure, algorithms, program structure, we reconstruct the program based on modularization, parallellelization technology with an object-oriented thinking, leading to the improvement of the readability, scalability and parallelism of the algorithm.Secondly, we study a variety of parallel frameworks, especially the OpenMP multi-core CPU parallel method and the GPU-based CUDA platform parallel method. After analyzing the dynamic explicit algorithm flow, we have determined the parallelizing process strategy of the serial program. Based on OpenMP, we achieve a sheet metal stamping algorithm working under multi-core CPU, Based on CUDA platform, we achieve a "CPU as a master and GPU as parallel coprocessors" collaborative parallel computing algorithm.Finally, we presents a parallel dynamic explicit sheet metal forming simulation algorithm on a heterogeneous platform based on CPU-GPU task partition collaborative strategy. It combines two parallel methods to maximize the use of the computing resources of the computer, and to further release the computing potential. Task partitioning algorithm and CPU-GPU load balancing strategy are also studied to optimize the "data synchronization" bottleneck, we design a better data storage structure to improve the computing performance. Based on VS2010, CUDA5.0platform, we complete all the algorithm above through C/C++programming. Examples are given to verify the accuracy and the efficiency, such as a box forming simulation and a fender panel forming simulation. We compare the results in different parallel methods and study a speedup trends under different computing scales. It is proved that the collaborative algorithm is a highly efficient and advanced method on heterogeneous platforms. It dramatically reduces the simulation time of sheet forming and it obtains about28X speedup comparing to the traditional serial algorithm in large-scale problems. We believe this study is of great value to the Parallel Optimization and reconstruction of finite element method algorithm.