Application Research of Machine Learning in Optimal Design of High Pressure-Ratio Centrifugal Impeller

With the advent of the era of big data and the development of artificial intelligence, many beneficial attempts and explorations have been made in the field of aerodynamic optimization. The traditional optimization method has the disadvantages of time-consuming, complicated parameterization of the blade, and high calculation cost. This paper uses the SVM (support vector machine) method based on the centrifugal compressor database to construct a nonlinear model between the parameters of blade and the efficiency, which replacs the CFD. After that, the genetic algorithm is used for global optimization to obtain the optimal parameters combination. Compared with the traditional optimization method-D3D, the time consumed by this method is greatly reduced, and the change trend of the optimized blade is close to that of D3D. The optimization results have been numerically verified by Numeca, and the peak efficiency is 0.2% higher than D3D, indicating that machine learning optimization has high reliability.