A Parallel Algorithm for the Natural Frequency of Chain Systems on CPU-GPU Heterogeneous Computing Systems

The multibody system transfer matrix method (MSTMM) is a novel dynamic approach which avoids global dynamics equations with a system inertia matrix, utilizes low‐order matrices independent of system degree of freedom, and is highly stylized. With the increase of the number of elements, the time to compute the natural frequency of chain system with the multibody system transfer matrix method is increasing. In order to improve the computational efficiency of the algorithm, an effective way is to parallelize the algorithm. The existing parallel algorithms of using the MSTMM to compute the natural frequency of multibody systems are based on CPU parallel computation, which cannot utilize the GPU of computer, resulting in a waste of computing resources. Therefore, this paper proposes a parallel algorithm for computing the natural frequency of chain systems in CPU-GPU heterogeneous environment. The CUDA parallel computing model is utilized to leverage the computational power of GPUs, while the MPI parallel computing model enables efficient communication and coordination between CPUs and GPUs. The results show that the proposed algorithm is effective, achieving an acceleration ratio 4 times higher than the existing parallel algorithms under the same hardware condition.