Abstract: Geometric degeneracies and topological ambiguities in complex CAD models often distort parametric domain mappings, leading to poor mesh quality, feature loss, or even generation failure. To improve mesh reliability and geometric consistency, this study proposes a robust high-quality surface mesh generation framework that combines both analytic surfaces and discrete STL meshes as input. First, the method generates adaptive curve-aligned meshes from B-rep surfaces while ensuring multi-surface coordination. It then classifies parametric domain distortions. For surfaces with boundary distortions, the algorithm repairs and reparameterizes input STL meshes, reconstructing their parametric domains under curve-aligned mesh constraints. Finally, it regenerates the mesh in the corrected parametric domain and interpolates it back to the physical space. Experimental results on practical engineering models show that the proposed method exhibits significant advantages in element quality and reliability. The average proportion of elements with a minimum angle less than 45° reduces by 43.27%; the proportion of elements with a radius ratio between 0.9 and 1.0 reaches 66.46%, which outperforms the 56.10% and 59.63% of comparative commercial software. For complex CAD models containing thin-walled surfaces, degraded surfaces and periodic surfaces, the method can generate geometrically consistent surface meshes.