Multi-Conditional Optimization of a High-Specific-Speed Axial Flow Pump Impeller Based on Machine Learning

Abstract

In order to widen the range of high-efficiency area of a high-specific-speed axial flow pump and to improve the operating efficiency under non-design conditions, the parameters of the axial flow pump blades were optimized. An optimization system based on computational fluid dynamics (CFD), optimized Latin hypercube sampling (OLHS), machine learning (ML), and multi-island genetic algorithm (MIGA) was established. The prediction effects of three machine learning models based on Bayesian optimization, support vector machine regression (SVR), Gaussian process regression (GPR), and fully connected neural network (FNN) on the performance of the axial flow pump were compared. The results show that the GPR model has the highest prediction accuracy for the impeller head and weighted efficiency. Compared to the original impeller, the optimized impeller is forward skewed and backward swept, and the weighted efficiency of the impeller increases by 1.31 percentage points. The efficiency of the pump section at 0.8Qd, 1.0Qd, and 1.2Qd increases by about 1.1, 1.4, and 1.6 percentage points, respectively, which meets the optimization requirements. After optimization, the internal flow field of the impeller is more stable; the entropy production in the impeller reduces; the spanwise distribution of the total pressure coefficient and the axial velocity coefficient at the impeller outlet are more uniform; and the flow separation near the hub at the blade trailing edge is restrained. This research can provide a reference for the efficient operation of pumping stations and the optimal design of axial flow pumps under multiple working conditions.