Abstract:
With the rapid development of high-performance computing technologies, the complexity and scale of scientific computing problems have significantly increased. As a critical input for scientific computing, mesh generation has become a key research area in high-performance computing. To address the challenges of low computational efficiency and complex human-machine interaction in large-scale mesh generation, intelligent mesh generation methods have become a research hotspot. However, achieving a deep integration of high-performance computing and artificial intelligence in mesh generation remains an open research problem. To tackle these issues, this paper introduces, for the first time, an intelligent parallel mesh generation framework based on unsupervised learning, designed to efficiently generate large-scale multi-block structured meshes. The framework integrates a block partitioning method based on cross-field, a Fourier neural network model, and a mesh alignment technique. It employs an unsupervised learning approach to adaptively learn high-quality mesh partitioning rules, effectively overcoming the limitations of existing intelligent methods in handling complex geometries. Additionally, a parallel transfer learning mechanism is introduced for the first time, significantly enhancing the model's convergence efficiency by transferring similarity features of mesh generation tasks across different processes. Experimental results show that, for million-dimensional mesh generation tasks, this method achieves more than an order of magnitude improvement in efficiency compared to traditional mesh generation methods. At the same time, it outperforms existing intelligent mesh generation methods in terms of mesh quality, reaching the optimal level.