Key points are not available for this paper at this time.
In this paper, the aerodynamic performance and flow field of the centrifugal compressor are deeply studied by a novel Fourier series Kolmogorov-Arnold network (FKAN) and non-dominated sorting genetic algorithm-II (NSGA-II). Compared to artificial neural networks (ANNs) and Kolmogorov-Arnold networks, the predictive performance (accuracy) of FKAN is not only significantly improved, but also the prediction error is significantly reduced. The optimization of a centrifugal compressor is conducted by integrating FKAN with the NSGA-II. The blades of the optimized impeller are offset toward the pressure surface at the root and bend toward the suction surface overall, which helps us to improve the aerodynamic performance by reducing flow separation and tip leakage flows. The aerodynamic performance is effectively improved across the entire range of operating conditions for optimized centrifugal compressors. Notably, the polytropic efficiency of the centrifugal compressor is increased by 1.47%, and the pressure ratio is improved by 3.05% at a mass flow rate of 1.65 kg/s. The polytropic efficiency of the centrifugal compressor effectively rises as much as 1.04%, and the pressure ratio effectively rises as much as 2.58% at the design condition of 1.77 kg/s. This study contributes to the development of a novel surrogate modeling technique and the understanding of the physical mechanism for performance improvement in fluid machinery optimization.
Zhang et al. (Thu,) studied this question.