A Mixed Cell Compressed Sparse Row for Time Domain Boundary Element Method in Elastodynamics

Weiyu Zhou,Yongqiang Chen
DOI: https://doi.org/10.1016/j.advengsoft.2024.103633
IF: 4.255
2024-01-01
Advances in Engineering Software
Abstract:In this paper, a novel mixed cell compressed sparse row (mCCSR) scheme is proposed to address the sparsity, block structure, and partial sub-matrix symmetry inherent in the coefficient matrix of the time domain boundary element method (TDBEM) in elastodynamics. The impulse function used in the fundamental solution of TDBEM introduces a time-lag phenomenon in the responses originating from the excitation point, resulting in sparsity within the coefficient matrix involved in convolutional integration at each time step. Furthermore, the fundamental solution of the displacement in the boundary integral equation exhibits symmetry between the component indices, leading to symmetry in the displacement coefficient sub-matrix. In contrast, the traction coefficient sub-matrix lacks such symmetry. The mCCSR scheme is proposed to efficiently store these two types of coefficient matrices in TDBEM, along with a detailed explanation of the corresponding matrix operations and the GMRES iteration solution method. The innovative scheme significantly reduces the number of indicators compared to the general compressed sparse row (CSR) scheme and is suitable for loop unrolling to accelerate computation. Examples are presented to analyze the factors affecting the average proportion of non-zero entries. Additionally, this paper also suggests a scale indicator for assessing computational capability, facilitating a performance comparison with the conventional method, adaptive cross approximation method (ACA), CSR scheme, and mCCSR scheme. The results demonstrate that mCCSR scheme achieves the computational capacity scale 5–10 times and computing speed 15–20 times faster than the conventional method, which significantly outperforms both the ACA scheme and the CSR scheme.
What problem does this paper attempt to address?