Automation of multi-fault diagnosing of centrifugal pumps using multi-class support vector machine with vibration and motor current signals in frequency domain

Reliable detection and isolation of centrifugal pump (CP) faults is a challenging and important task in the modern industries. Hence, this paper proposes an artificial intelligence-based multi-fault detection of CPs driven by induction motor. The intelligent fault detection methodology is developed based on the multi-class support vector machine (MSVM). The mechanical and hydraulic faults in CPs are mutually dependent and therefore may exist concurrently. Hence, in the present research, an assortment of various flow instabilities like the suction re-circulation, discharge re-circulation, pseudo-re-circulation and dry runs are considered coexisting with mechanical faults, like the impeller cracks and pitted cover plate faults. The power spectrum of the CP vibration and the induction motor line-current data is used for monitoring the CP condition. The best statistical feature combination is selected based on a wrapper model. Gaussian radial basis function (RBF) is used for the kernel mapping. In addition, the RBF kernel parameter (width) and MSVM parameters are optimally selected using a fivefold cross-validation technique. Also, variation of operating speed of the CP drastically changes the system vibration level owing to the change in fault manifestations; hence, in the present work a methodology that is independent of CP operation is proposed and tested. Thereafter, it is observed that the proposed methodology is remarkably robust and successfully classifies multiple individual as well as coexisting CP faults at all the tested CP speeds.