Machine Learning Prediction of Transient Performance and Impeller Hydraulics in Centrifugal Pumps

ABSTRACT To explore new approaches for monitoring the operating status of centrifugal pumps, this study employs four machine learning algorithms—support vector regression (SVR), random forest regression (RFR), backpropagation neural network (BPNN) and convolutional neural network (CNN)—to predict the transient performance (head, efficiency and shaft power) and hydrodynamic forces (radial and axial forces) of a centrifugal pump under throttling control. Time‐domain pressure signals collected from discrete monitoring points near the volute tongue were used as the primary carriers of performance‐related information. The results show that both the RFR model and the CNN developed in this study achieve average relative errors below 5% when predicting pump performance and hydrodynamic forces. The CNN yields the highest correlation coefficients for head, efficiency, shaft power and radial force, all exceeding 0.9, indicating strong agreement between predicted and measured values. These findings demonstrate that machine learning, particularly CNN, can provide accurate predictions of instantaneous pump performance and impeller hydrodynamics within the investigated pump configuration and throttling range, highlighting its potential for centrifugal pump condition monitoring and soft‐sensing applications.