An explicit asynchronous step parallel computing method for finite element analysis on multi-core clusters

Abstract

The finite element analysis of complex structure often requires a refine mesh in some local domain. To reduce the computation time, an explicit asynchronous step parallel computing method is proposed. The domain decomposition with overlapping node method is used to divide the model into different subdomains. The multiple overlapping nodes between different subdomains constitute the coupling region. Each subdomain selects the time step based on the mesh characteristics. The subcycling method is adopted to tackle the matching of asynchronous step boundary. The subdomain model, boundary information and calculation results are stored in parallel files which mean the overall finite element analysis process is implemented in parallel. The validity and efficiency of the proposed method are verified through three simulation cases which conducted on Tianhe 2 multi-core supercomputers. The results of simulation cases show that the proposed method has a higher accuracy than classic subcycling method under the same time step. The total speedup of the algorithm relates to the step ratios between subdomains, the number of subdomain and the load balance. This approach offers an efficient way to solve large-scale and super-scale structural dynamics analysis with local refine mesh.