Multi-objective optimization framework based on SACGAN-NSGA-II for film cooling performance of conical holes on the leading edge of turbine blades

Film cooling design of turbine blade is a typical knowledge-intensive engineering task, requiring the integration of multi-domain expertise. This paper proposes a novel knowledge-guided multi-objective optimization framework (SACGAN-NSGA-II) including implicit knowledge learning and explicit knowledge decision to search for optimal conical hole configurations that simultaneously maximize average film cooling effectiveness (AFCE) and film coverage ratio (FCR). Self-attention conditional generative adversarial network (SACGAN) is used to establish a high-dimensional nonlinear mapping between conical hole configurations and film cooling effectiveness on the leading edge of turbine blades from 1200 sets of cases calculated by computational fluid dynamics. Non-dominated sorting genetic algorithm II (NSGA-II) is used to explore the SACGAN-established design space and search for optimal conical hole configurations. Based on the mapping relationship established by SACGAN, sensitivity analysis is conducted to quantify the influence of conical hole design parameters on film cooling performance. Sensitivity analysis results are then used to guide the interpretation of optimization results and support the mechanism analysis behind the observed improvements in AFCE and FCR. A comparative analysis is conducted between the optimization results obtained using SACGAN and those derived from Kriging surrogate model and General Regression Neural Network (GRNN). The comparison results show that SACGAN-NSGA-II exhibits superior optimization performance, achieving significantly better optimization results. SACGAN-NSGA-II demonstrates significant engineering value through 53.30% improvement in AFCE and 13.03% improvement in FCR compared to baseline configuration.