Key points are not available for this paper at this time.
Abstract To address the challenges encountered during a multioperating-point optimization of centrifugal compressor blades, including a large design space, redundant search, limited flexibility, and low optimization efficiency, this paper presents an aerodynamic optimization approach for centrifugal compressor impeller blades based on parameterized free-form deformation (FFD) and deep learning. To realize the parametric control of a compressor blade, a three-dimensional FFD control body is arranged around a blade according to the geometric characteristics. The inverse distance weighting (IDW) mesh deformation algorithm is used to extrapolate the deformation of surface mesh points to volume mesh points. Using this method, the analysis mesh can be directly updated with high quality instead of regenerating the geometry model when design variables are changed during optimization. Based on this and convolutional neural networks, a design optimization framework is established by combining the optimal Latin hypercube sample method and the NSGA-II multi-objective genetic algorithm. An optimization of the centrifugal compressor blade of a micro gas turbine was carried out, and the results showed that the total pressure ratio and isentropic efficiency at the design point was increased by 4.03% and 1.21 percentage points, respectively. The optimization results demonstrate that the developed optimization framework can effectively improve the performance of centrifugal compressors and provide support for compressor design.
Xu et al. (Mon,) studied this question.