A Rapid Design Method for Centrifugal Pump Impellers Based on Machine Learning

Centrifugal pumps are widely used across various industries, and the design of high-efficiency centrifugal pumps is essential for energy savings and emission reductions. The development of centrifugal pump models primarily uses an iterative design approach combining direct and inverse problem-solving based on one-dimensional flow theory. However, this semi-empirical, semi-theoretical design process is time-consuming and costly. To reduce development time and costs, this paper proposes a rapid impeller design method focused on hydraulic performance, integrating traditional similarity design theory with machine learning. The proposed model uses neural networks to predict empirical coefficients, determine key dimensions such as the impeller’s inlet diameter, outlet diameter, outlet width, and axial distance. Once these parameters are defined, the main dimensions of the impeller can be calculated. The blade profile is defined using a 5-point B´ezier curve. Variations in the cross-sectional area of the flow passage influence the internal flow state of the centrifugal pump, ultimately impacting its hydraulic efficiency. A genetic algorithm, guided by variations in the cross-sectional area of the flow passage, optimizes the blade profile, achieving an improved impeller flow path and completing the rapid design of the centrifuge. This method significantly shortens the development cycle and lowers design costs, making it a promising technique for future impeller designs.