Domain decomposition strategies for high-performance smoothed particle Galerkin (SPG) modeling of SiCf/SiC scribing: balancing scaling efficiency and numerical stability

Abstract This study investigates domain decomposition strategies for high-performance smoothed particle Galerkin (SPG) modeling of single-grit diamond scribing of silicon carbide fiber-reinforced silicon carbide (SiC f /SiC) composites. Domain decomposition enables massively parallel processing (MPP) by partitioning the SPG model into subdomains, each assigned to a processor for parallel execution. The SPG model without domain decomposition is first validated against experimentally measured scribing forces and is used as the baseline. The computational time and force prediction error associated with three domain decomposition strategies are evaluated to identify the trade-off between scaling efficiency and numerical stability. Three principles for optimal domain decomposition are identified. First, subdomains should contain similar numbers of particles to ensure balanced workload across processors and maximize computational efficiency. Second, the number of subdomains should be selected to balance computational speedup and numerical stability. Too few subdomains underutilize available computational resources, while too many increase inter-subdomain communication and may amplify numerical discrepancies and force fluctuations. Third, subdomain boundaries should avoid high-deformation and high-contact regions. Aligning subdomains parallel to the cutting direction confines active deformation within fewer subdomains, reduces data exchange, and improves both accuracy and computational efficiency. Based on these findings, we envision that GPU-accelerated high-performance computing (HPC) architectures with thousands of processing cores can enable large-scale multi-grit grinding simulations using SPG modeling.