Aerodynamic optimization for the vaned diffusers of an ultra-high-pressure-ratio centrifugal compressor based on an efficient machine learning algorithm

High-efficiency, wide-range high-pressure-ratio centrifugal compressors for gas turbines are in high demand in the aviation industry. Although the centrifugal compressor impeller efficiency could exceed 90%, the stage efficiency remains so low that approaches for more efficient diffusion are urgently needed. Optimizing the diffuser vane angle and thickness is the main effective method for improving the diffusion performance. In this paper, an ultra-high-pressure-ratio centrifugal compressor with vaned diffusers was numerically studied, and the machine learning algorithm, XGBoost, was employed to optimize the blade angle and blade thickness of the vaned diffuser. Three operating points at 100% rotational speed were chosen as the optimization objects to maintain the choke mass flow and surge margin. An active learning approach was introduced to the iterative process, which increased the possibility of finding the global optimal solution. The results showed that compared with other surrogate models, the adopted optimization method in this paper reached the global minimal with fewer iterations and comparable accuracy as the Computational Fluid Dynamics results. With the same diffuser inlet and outlet radius, the surge margin was almost unchanged, but the peak stage efficiency increased by 1.03%. The influence of aerodynamic parameters and design parameters of the optimized diffuser on the stage performance was discussed. This research could provide guidance and reference for the design and optimization of ultra-high-pressure-ratio centrifugal compressors with vaned diffusers.