Identification and optimization for low-quality CFD grids based on K-Means++ clustering method

Mesh generation and optimization represent a fundamental preprocessing stage in computational fluid dynamics (CFD), where their significance becomes particularly pronounced in hypersonic flow simulations. The complex physical phenomena impose exceptionally rigorous demands on mesh quality to ensure computational accuracy and stability. Although traditional mesh quality inspection can be performed automatically by software, the optimization of low-quality grids heavily relies on manual observation and adjustment, exhibiting low automation efficiency. This presents persistent challenges in CFD research. Addressing the local optimization of the intricate structured grid for CFD simulations, this study establishes an overall mesh quality evaluation model based on the maximum and minimum interior angle metric. Subsequently, it automatically clusters low-quality grids through the K-Means++ clustering method. Finally, it applies the Laplacian smoothing method to the local low-quality blocks partitioned by the K-Means++ clustering method to enhance the overall mesh quality. Taking a distorted wavy structured grid, a space shuttle structured grid, and a hypersonic blunt body structured grid as numerical examples, the results show that this integrated approach establishes a novel intelligent optimization pathway for CFD preprocessing workflows, effectively minimizing the need for manual intervention. Local optimization of low-quality regions demonstrates superior effectiveness compared to the entire mesh optimization. The proposed method demonstrates effective automatic partitioning of low-quality grid regions across different configurations, significantly enhancing mesh optimization efficiency while contributing to enhancing the accuracy of the flow simulation.