Optimization of centrifugal pump performance and excitation force based on machine learning and enhanced non-dominated sorting genetic algorithm III

• Published in: Engineering applications of artificial intelligence Vol. 143; p. 110036
• Main Authors: Haoqing, Jiang, Wei, Dong, Sucheng, Li, Fan, He
• Format: Journal Article
• Language: English
• Published: Elsevier Ltd 01.03.2025
• Subjects: Axial force; Centrifugal pump; Genetic algorithm; Machine learning; Optimization design; Centrifugal pump; Genetic algorithm; Optimization design; Axial force; Machine learning
• ISSN: 0952-1976
• DOI: 10.1016/j.engappai.2025.110036

This study addresses the issue of excessive vibration in centrifugal pumps without compromising their performance by employing six machine learning methods. These include the Backpropagation (BP) Neural Network, Feedforward Neural Network (FNN), Support Vector Machine (SVM), Extreme Learning Machine (ELM), Gaussian Process Regression (GPR), and Tree Ensemble (TE). These methods are used to construct a mapping model that correlates optimization objectives and parameters. The model served as the fitness function in the Adaptive Non-dominated Sorting Genetic Algorithm III (A-NSGA-III). A total of 1000 optimization results generated by the A-NSGA-III were evaluated to determine the final optimization scheme, using a reverse design approach to identify optimal parameters.A comparison of the hydraulic performance before and after optimization revealed minor improvements in the head, efficiency, and shaft power. The radial force decreased by 17.81%, and the axial force decreased by a substantial 98.12%, meeting the set objectives. The significant reduction in axial force was predominantly observed at the balance chamber. Entropy production theory analysis indicated a considerable decrease in the entropy production rate on the impeller's inner wall.Cross-wavelet transform and wavelet coherence analysis showed a significant correlation between the entropy production rate and the energy variation of the axial force. Before optimization, there was a pronounced coherence between the entropy production rate and the resonance period, which was notably reduced after optimization. This method significantly addresses the issue of excessive axial force in the operation of centrifugal pumps without compromising the basic hydraulic performance. This study provides a reference for the structural optimization of centrifugal pumps.