Centrifugal Pump Optimization via Integration of Machine Learning and Computational Fluid Dynamics

Shahid Rabbani1,2

Kürşad Melih Güleren3

Imran Afgan1,4,Email

1Department of Mechanical & Nuclear Engineering, College of Engineering, Khalif University, Abu Dhabi, P.O.Box 127788, UAE.
2Center for Interacting Urban Networks (CITIES), New York University, Abu Dhabi, UAE.
3Eskisehir Osmangazi University, Faculty of Engineering and Architecture, Department of Aeronautical Engineering, 26040, Turkey.
4Department of Fluids and Environment, School of Engineering, University of Manchester, Manchester, M13 9PL, United Kingdom.


This paper presents a novel approach towards centrifugal pump design optimization, employing an integration of deep learning and Computational Fluid Dynamics (CFD). Centrifugal pumps’ complex operation environments require innovative strategies for optimal performance. By harnessing machine learning, this study innovatively shifts the traditional design focus towards maximizing pressure rise and eliminating reverse flows in the impeller. The approach uses a deep learning model to predict pump performance parameters which are evaluated via comparative analysis with predefined outputs. Moreover, correlations between pressure head and relative velocity angle are investigated. Key manipulated design parameters include relative diffuser vane angle, number of diffuser vanes, number of impeller blades and the impeller wrap angle. The results demonstrate the efficacy of machine learning in delivering accurate predictions and valuable insights into pump performance, thereby paving the way for more efficient and reliable centrifugal pump designs.