The Stopping Performance of a Centrifugal Pump with Splitter Blades at Small Discharge Valve Openings

To reveal the hydraulic characteristics of a centrifugal pump with splitter blades during shutdown, a low specific speed closed impeller centrifugal pump is subjected to shutdown experiments under eight non-rated operating conditions in this paper. The transient evolution characteristics of five performance parameters with time are obtained, including rotational speed, flow rate, inlet and outlet pressures, and head. Meanwhile, the shutdown fitting models based on three machine learning models are developed. The results show that the integrated neural network model can more accurately predict the hydraulic performance of the physical pump during shutdown than the decision tree regression and random forest regression models. During the pre-mid period of the shutdown, the integrated neural network model predicts a maximum error of about 3.21% for the instantaneous flow rate and about 3.58% for the instantaneous head. This study provides a reference for the performance control of centrifugal pumps during transient operation.