Multi-Objective Optimization of Multistage Pump Balance Drum System Based on BP Neural Network and Genetic Algorithm

Abstract

The axial force balancing capacity of a balance drum is a key factor affecting the life of multi-stage centrifugal pumps. In this paper, a double shell segmental multistage pump is taken as the research object. The hydraulic performance and axial force performance are set as the optimization objectives, and the performance data are obtained by numerical simulation with FLUENT software. The BP neural network is used to establish the prediction model of structural parameters of the balance system, hydraulic performance and residual axial force performance, and it is used as the adaptive value evaluation model of genetic algorithm to solve the optimal value in the sample space. The results show that the radial clearance of the balance drum and the balance tube orifice flowmeter, the axial width of the balance cavity are the significant factors affecting the hydraulic performance and axial force performance of the multistage pump. When the radial clearance of the balance drum is 0.1mm, the clearance of the orifice flowmeter is 1.95mm, and the axial width of the balance cavity is 55mm, the multi-stage pump has the best hydraulic performance and the smallest residual axial force. The vortex band in the balance cavity can increase the amount of the fluid spin and enhance the axial force balancing capacity of the balance drum. The greater the area occupied by the negative high-rotation fluid in the balance cavity, the stronger the ability of the balance drum to balance the axial force. The test results show that compared with the prototype multistage pump, at nominal flow rate, the head and efficiency of the optimized model are increased by 0.71% and 1.63% respectively, and the bearing temperature and vibration speed of the multi-stage pump are significantly reduced.