Performance Optimization And Program Implementation Of Explicit Finite Element Program Parallel Automatic Global Contact Algorithm

Based on mechanics theory, the assessment of performance, safety and reliability of weaponry or industrial products, such as projectile penetration, birds crashing, car crash test, can be abstracted as contact problem. Using numerical simulation method to predict and evaluate and validate experiment result is one of the most common methods. EFEM (Explicit Finite Element Method) is proved to be effective and reliable. But the common problem was: contact-detect calculation takes lots of time (usually40%, sometimes even more). So it's necessary to optimize the contact-detect algorithm.Firstly, this paper introduces the theory, formulation and computing procedure of EFEM and then tests the program from 4 processors to 64 processors. Through contrasting and analyzing the ratio different parts takes, the bottleneck of the program was found. Global contact-detect algorithm is the most important factor, especially the contact domain decomposition algorithm in it, that affect the calculation efficiency heavily. With the increasing of processors, communication will become the bottleneck of this program. So it's necessary to analyze the comprehensive influence of communication and load-balance.Secondly, in order to support the optimization of domain decomposition, contact node cross-bucket detection which successfully decrease the contact domain decomposition times was added to take place the fixed interval decomposition strategy. Besides that, the whole period was separated into sparse decomposition and non-sparse decomposition, in order to decrease the times of contact node cross-bucket detection. The optimization proved to be right and efficient by actual computing test.Thirdly, analyzing the comprehensive influence of communication and load-balance which comes from dual domain decomposition. According to the test results, a qualitative conclusion was got which shows that:the increasing of contact-detect processor may lead to efficiency improve when it's small, but more may lead to efficiency decrease. So contact-detect processor number need a carefully set.Through those analysis, algorithm optimization and tests, global contact-detect algorithm optimization has been done with an effective improve. At the same time, a qualitative was got, which can works as guidance for large scale parallel computing.