Pump cavitation detection through fusion of support vector machine classifier data associated with vibration and motor current signature
Abstract
Centrifugal pump is one of the most widely used machines in various industries and cavitation is a common fault in this type of pump. This paper presents a diagnostic method based on data fusion in order to precisely and effectively detect the cavitation fault of a pump at an early stage. An experimental test system is set up to simulate electro-pump cavitation and vibration signal and the current signature of the motor-driven pump is measured for cavitation detection. In order to extract probabilistic features, each signal is analyzed separately. The data are then classified using Support Vector Machines (SVMs). Finally, a decision fusion strategy based on Bayesian theory is used to fuse all the classifiers resulting from the vibration signal and electrical current signal. Comparison of the fused data and the results obtained from each of the classifiers shows an improvement in cavitation diagnosis accuracy.
