Integrated optimization of a turbine stage at a low Reynolds number via NURBS surface and machine learning

Abstract

As the turbine load increases, both the blade profile losses and secondary flow losses in the endwall region cannot be ignored for turbines operating at low Reynolds numbers. To fully explore the flow control effects of the blade and endwall shapes, a viable integrated parameterization method and optimization method are formulated for turbine stages. NURBS surfaces are utilized for parameterizing the endwall and blade geometries. Combined with machine learning methods and improved chaos particle swarm optimization (ICPSO), an efficient surrogate model optimization strategy for high-dimensional small-sample problems is discussed. The above methods are applied to the integrated optimization of the endwalls and suction surfaces of a low-speed turbine stage, which is operated at a low Reynolds number. Under the constraint of nearly unchanged output power, the isentropic efficiency of the turbine stage is improved by 1.33%. The flow control effects of blade reshaping and endwall contouring are separately compared. The results show that the flow structures near the suction surfaces of the blades can be adjusted by blade reshaping. Moreover, the flow near the endwalls can be reorganized by using contoured endwalls and optimized blades, and the secondary flow in the endwall region can thus be significantly weakened.