Three-dimensional blade shape optimization for a transonic axial flow compressor through incorporating surrogate model and sequential sampling

High level aerodynamic performance has been always expected for the axial flow compressors, and it is the consistent goal for axial flow compressor research. To achieve such a goal, the incorporation of CFD with optimization algorithm and surrogate model in blade geometry optimization has become a common practice and been used extensively. But the conventional surrogate model based on merely initial sampling often deviates from the real optimization problem during optimization process and then brings the optimizer to search locally, leading to the compromised optimal results. There are yet much to do to improve such optimization design method. An optimization method of surrogate model being updated by sequential sampling strategy is developed to achieve global optimal design geometry and permit high-level of aerodynamic performance for axial flow compressor. Preliminary Kriging surrogate model is constructed with a small number of selected DOE samplings, where the multiple optimization objective functions are obtained based on CFD simulations. The optimization is performed on the surrogate model with NSGA-II optimization algorithm and Pareto fronts successively obtained. To improve the surrogate model, the MSE (Mean Squared Error) criteria is used to select the refinement point from the newest Pareto front, and it is used to update and improve the surrogate model gradually during the optimization. Such adaptive feature of the surrogate model has enabled the optimizer to search globally. The method is used to optimize transonic Rotor 37 at design flow rate, where the blade shape is varied simultaneously in terms of sweep and lean, and the geometry is optimized. In the converged Pareto front, abundant candidate designs with significant performance gains are produced. Three points over the Pareto front are selected and analyzed to take an insight into the optimization effectiveness. Overall performance curves of optimized geometries are predicted over the entire flow range and they are significantly improved compared with the original ones. Significant overall performance gains arising from the blade optimization are supported by the improved flow behavior. The overall pressure ratio or efficiency gains of the optimized blades are attributed to the significant improvement in the radial distribution of aerodynamic parameters. Further research shows that the shock structure is changed and separation zone is reduced with the optimized blades, which are the major reasons for the improvement of the aerodynamic performance of optimized blades.