Application of machine learning techniques for performance prediction and optimization of a double-suction centrifugal pump

Nowadays, improving the efficiency of energy systems, and in particular of industrial pumps, has become a crucial aspect of the industrial sector. This thesis presents an artificial intelligence-based optimization approach applied to a double-suction centrifugal pump, integrating CFD simulations, machine learning (ML) models, and a multi-objective optimization algorithm. Several ML models were trained on a dataset consisting of 1,257 geometric pump configurations to predict efficiency and head. The model with the best predictive performance was subsequently coupled with the NSGA-II algorithm to identify an optimized configuration, aiming to increase efficiency while keeping the head nearly constant. The optimized geometry shows an efficiency improvement of 1.68% compared to the baseline design, with a negligible variation in head. CFD validation highlighted a more uniform velocity distribution, smoother pressure gradients, and a reduction in energy losses, further confirmed by entropy generation analysis. Moreover, the optimized pump maintains improved efficiency over a practical range of off-design operating conditions.