The pursuit of high-fidelity aerodynamic optimization has long been hindered by three interdependent bottlenecks: (1) exponential computational costs in parametric mesh generation, (2) geometric fidelity loss during automated shape discretization, and (3) inflexible batch processing in commercial computational fluid dynamics suites. Here, a topology-aware deep learning framework that synergizes physics-guided O-grid generation with sparse autoencoder-based parameter optimization was presented. The methodology integrates topology-specific strategies for closed/open trailing edges based on aerodynamic characteristics, with generalizability validated through wind tunnel domain and surface mesh generation tests. Optimal grid parameters (surface cell size yb = 3, radial node count jx = 120, wind tunnel diameter d = 6 × 340C (C: chord length) are determined via sparse autoencoder deep neural networks. Comparative validation against experimental datasets confirms the method's efficacy, demonstrating seamless integration with novel airfoil generation algorithms. This paradigm shift enables real-time batch optimization previously deemed computationally intractable, as evidenced by second-level grid generation (1.74 s vs 20 min manually), and with the optimal grid parameters determined based on computational parameters that yield lift-to-drag ratio curves closest to experimental data, thereby significantly enhancing computational throughput in aerodynamic optimization workflows.
Wang et al. (Fri,) studied this question.