Optimization and Knowledge Discovery of a Three-Dimensional Parameterized Vane with Nonaxisymmetric Endwall

Abstract

The optimization and knowledge discovery of a low-aspect-ratio cascade was conducted. The techniques of nonaxisymmetric endwall contouring, section profiling, and compound lean were combined for cascade parameterization. Their effects on cascade performance were investigated through analysis of flow details and well-designed scatter plot. It is found that nonaxisymmetric endwall contouring is helpful to reduce the profile loss while also inhibiting the secondary flow. Section profiling is an effective technique to reduce the profile loss. And the lean structure is responsible for the multimodality of the loss function because it makes the reduction of profile loss and endwall loss in severe conflicting relation. The combination of the techniques is shown to effectively improve the cascade performance. In addition, through the analysis of self-organizing map and Pearson’s correlation, the mass flow rate and outlet flow angle are found to be linearly correlated, whereas the relation between the loss and mass flow rate is complex. Finally, because the optimal solution is sensitive to the objective function as shown by the two optimization processes, the data mining of interactions among performance functions and design variables becomes very important, which can help to achieve the desirable design performance in the prescribed design space.