In the present work, an optimization method coupled with computational fluid dynamics (CFD), backpropagation (BP) neural networks, and genetic algorithms (GA) is proposed to suppress internal flow and improve the performance of centrifugal compressor. The blade profiles near the impeller hub and shroud are both parameterized by using the Bezier curve. In addition, the coordinates of control points are identified as optimization variables, and the efficiency and pressure ratio of centrifugal compressor are conducted as optimization objectives. The internal flow of the original model and the optimization samples are conducted based on the CFD method. Simultaneously, the efficiency and pressure ratio are trained by using the BP neural network, and the prediction accuracy of the trained model exceeds 90%, which is satisfied to establish the regression model between the optimization variables and optimization objectives. The multi-objective optimization problem is tackled by employing the GA. The multi-objective optimization problem is displayed by Pareto front solution and sensitivity between variables and objectives is solved by Pearson correlation analysis. The aerodynamic performance of the optimized model is significantly improved. In particular, the polytropic efficiency and pressure ratio of the optimized compressor under the design condition are improved by 0.65% and 1.99%, respectively. The optimization is very beneficial to expand the high-efficiency range of the centrifugal compressor. The polytropic efficiency and pressure ratio of centrifugal compressor rise as much as 9.15% and 3.84% at high flow rate, respectively. The efficacy of the optimized compressor’s performance is effectively verified through some aerodynamic analyses, including comparison of streamline distribution, turbulent kinetic energy, and static pressure in the centrifugal compressor. The comparisons of aerodynamic characteristic and pressure fluctuation are implemented between the original and optimized centrifugal compressors. The fluid structure in the impeller is effectively improved and the work of impeller blade is enhanced, while the unstable phenomenon is effectively suppressed.
Li et al. (Fri,) studied this question.