Centrifugal pump analysis and prediction in cavitation using RapidMiner tool and machine learning algorithms.

Being a reliable and efficient pump, Centrifugal Pumps (CP) are widely used today in many mechanical and industrial applications. But cavitation is a serious problem that impacts pump performance, as it involves the vaporisation of the liquid at the suction end as a result of the pressure drop. The study is aimed at predicting and analysing cavitation in CP using a Cavitation Measuring Technique (CMT) based on Computational Fluid Dynamics (CFD) and Machine Learning (ML) approach. The CMT is based on the Net Positive Suction Head Available (NPSHa) and Net Positive Suction Head Required (NPSHr) ratio calculated from the results of the CFD simulation at various rotational speeds of 700 to 2900 RPM. The critical ratio is the main input feature for a machine learning approach to cavitation onset classification. CFD simulations were performed to simulate both functional and non-functional pump conditions and results were analyzed on RapidMiner platform with different machine learning algorithms. Experimental results show that the Fine Tree algorithm gave an accuracy of 96.1% and Naive Bayes 95.5% in predicting the cavitation status. In addition, the maximum vapour volume fraction of 0.9477 was observed at the water-moving part of the impeller during CFD analysis. The proposed framework illustrates the ability of CFD and ML approaches to accurately predict cavitation in centrifugal pumps and to assess centrifugal pump performance.