Multi-objective optimization design of a centrifugal impeller by positioning splitters using GMDH, NSGA-III and entropy weight-TOPSIS

A centrifugal impeller with splitters was designed by three-dimensional (3D) inverse design method, and its efficiency, velocity non-uniformity at impeller exit and maximum equivalent stress of blades were optimized by providing the suitable blade stacking angle, work ratio and circumferential location of the splitter. First, 80 samples were generated by optimal Latin hypercube technique and the corresponding impellers were designed by 3D inverse design. Using computational fluid dynamics (CFD) and fluid structure interaction (FSI), the optimization objectives were obtained. Then, the group method of data handling (GMDH) artificial neural networks was established to link the design parameters and objectives by the specific formulas. The reference-point-based non-dominated sorting genetic algorithm (NSGA- III) was applied to search the Pareto front. Finally, the preferred impeller was selected by adopting entropy weight and the method of technique for order preference by similarity to an ideal solution (TOPSIS). The results showed that the splitter of the preferred impeller had a circumferential location of 0.58, blade stacking angle of 28 degrees and work ratio of 0.48. The nonuniformity at impeller exit and maximum equivalent stress of blades of preferred impeller obtained by NSGA-III were decreased, while the efficiency was improved.