Multi-objective optimization of centrifugal pumps based on extreme learning machine with Adaptive Non-dominated Sorting Genetic Algorithms III

In order to improve the hydraulic performance and operational stability of centrifugal pumps, this paper selects M192-400 centrifugal pumps as the research object to ensure that the centrifugal pump head, efficiency does not decrease and reduce the axial force as much as possible as the optimization goal, using the Plackett–Burman screening test will be centrifugal pump impeller balance hole diameter, radial position, circumferential position, and axial placement angle of four geometric parameters as optimization variables. The optimal Latin hypercubic sampling method was used to create a sample space of 90 groups, and the optimization variables were fitted to the optimization targets using the extreme learning machine. Multi-objective optimization is carried out by the Adaptive Non-dominated Sorting Genetic Algorithms III, and the optimal solutions of impeller balance hole diameter, radial position, circumferential position, and axial placement angle of 14 mm, 68.08 mm, −22°, and 6.62° are obtained by using the hierarchical analysis method and the validity of which is verified. Compared with the initial model, the optimized centrifugal pump increased head by 1.28%, increased efficiency by 0.17%, and reduced axial force by 96.47% to meet the optimization target. Before and after the optimization of the centrifugal pump wall pressure, cross-sectional pressure, velocity, and flow lines, turbulent kinetic energy distribution differences were analyzed. The optimization method of this paper makes the centrifugal pump hydraulic performance improved, the basic balance of the axial force, for the centrifugal pump optimization design to provide a reference basis.