Application of different surrogate models on the optimization of centrifugal pump

An optimization process for impellers was carried out based on numerical simulation, Latin hypercube sampling (LHS), surrogate model and Genetic algorithm (GA) to improve the efficiency of residual heat removal pump. The commercial software ANSYS CFX 14.5 was utilized to solve the Reynolds-averaged Navier-Stokes equations by using the Shear stress transport turbulence model. The impeller blade parameters, which contain the blade inlet incidence angle Δβ, blade wrap angle φ, and blade outlet angle β2, were designed by random sample points according to the LHS method. The efficiency predicted under the design flow rate was selected as the objective function. The best combination of parameters was obtained by calculating the surrogate model with the GA. Meanwhile, the prediction accuracies of three surrogate models, namely, Response surface model (RSM), Kriging model, and Radial basis neural network (RBNN), were compared. Results showed that the calculated findings agree with the experimental performance results of the original pump. The RSF model predicted the highest efficiency, while the RBNN had the highest prediction accuracy. Compared with the simulated efficiency of the original pump, the optimization increased efficiency by 8.34% under the design point. Finally, the internal flow fields were analyzed to understand the mechanism of efficiency improvement. The optimization process, including the comparison of the surrogate models, can provide reference for the optimization design of other pumps.