The application of support vector regression and mesh deformation technique in the optimization of transonic compressor design

Abstract

With the development of modern aero-engine, the compressor is required to achieve higher performance such as transonic operation, high stage pressure ratio and efficiency. At present, the traditional blade design methods have the disadvantage of highly relying on designer's experience and heavy computational burden. In order to reduce design variables used to define blade geometry, the present work introduces the Free-Form Deformation technique. It can achieve high degree-of-freedom deformation and update mesh and geometry simultaneously. To further alleviate computational cost, the Support Vector Regression surrogate model is adopted to replace the time-consuming numerical simulation. It performs well in small sample learning problems. On this basis, a surrogate-based optimization design framework is established by combining the Advanced Latin hypercube sample and NSGA-II multi-objective genetic algorithm. Then an aerodynamic optimization of the transonic NASA Rotor 37 is conducted as a validation. The results show that the pressure ratio and isentropic efficiency are increased by 4.2% and 2.5%, respectively. The optimized shape sweeps forward with a slight increase in twist angle, and shock loss is reduced effectively. Compared with traditional optimization method, the proposed framework can improve the optimization efficiency by decreasing design variables and training samples.