Blockage and cavitation detection in centrifugal pumps from dynamic pressure signal using deep learning algorithm

Centrifugal pumps play an important role in many industrial applications even in harsh environment for prolong duration. High efficiency with very low power consumption makes them very popular in industry. However, during their operation, they may fail due to some operationally developed faults, which may subsequently lead to the interruption in the continuous operation of pumps. Therefore, monitoring the health status of the centrifugal pumps is essential to prevent unwanted stoppage, which may further lead to the breakdown of the whole system. The main focus of this study is to propose a methodology to identify the presence and severity of blockages, and cavitation in the centrifugal pump using fluid pressure, which is very vital for fluid related faults. To simulate the blockage in the pump, the flow area of the suction pipe is restricted by dividing into six equal intervals (i.e., 0%, 16.7%, 33.3%, 50%, 66.6% and 83.33%) using a mechanical modulating valve. Due to blockage and cavitation, the main parameter which directly gets affected is the fluid dynamic pressure. Hence, in the present study, pressure signatures were captured at different blockage levels and at different running speeds with the help of a pressure transducer, which was mounted on the circumference of the centrifugal pump casing. Deep learning based binary data classification methodology is used to classify the data acquired from the pressure transducer. To get better performance of the data classifier, statistical features are extracted from time domain pressure signals. In order to identify the severity of the faults, binary classification of the data is performed at different blockage levels and running speeds. Finally, based on the results obtained from the classifier, existence of the faults (i.e., blockage and the cavitation), their severity levels are presented.