Multi-Objective Optimization of Deep-Sea Mining Pump Based on CFD, GABP Neural Network and NSGA-III Algorithm
Qiong Hu, Xiaoyu Zhai, Zhenfu Li
Abstract
In order to improve the hydraulic performance of a deep-sea mining pump, this research proposed a multi-objective optimization strategy based on the computational fluid dynamics (CFD) numerical simulation, genetic algorithm back propagation (GABP) neural network, and non-dominated sorting genetic algorithm-III (NSGA-III). Significance analysis of the impeller and diffuser parameters was conducted using the Plackett–Burman experiment to filter out the design variables. The optimum Latin hypercube sampling method was used to produce sixty sample cases. The GABP neural network was then utilized to establish an approximate model between the pump’s hydraulic performance and design variables. Finally, the NSGA-III was utilized to solve the approximation model to determine the optimum parameters for the impeller and diffuser. The results demonstrate that the GABP neural network can accurately forecast the deep-sea mining pump’s hydraulic performance, and the NSGA-III global optimization is effective. On the rated clear water conditions, the optimized pump has a 14.65% decrease in shaft power and a 6.04% increase in efficiency while still meeting the design requirements for the head. Under rated solid-liquid two-phase flow conditions, the head still meets the design requirements, the shaft power is decreased by 15.64%, and the efficiency is increased by 6.00%.