Feature extraction using wavelets and classification through decision tree algorithm for fault diagnosis of mono-block centrifugal pump

Abstract

Monoblock centrifugal pumps are employed in variety of critical engineering applications. Continuous monitoring of such machines becomes essential in order to reduce the unnecessary break downs. Vibration based approaches are widely used to carry out the condition monitoring tasks. Decision tree, fuzzy logic, support vector machine and artificial neural networks are some of the classification algorithms employed for condition monitoring and fault diagnosis. In the present study, fault discriminating capability of wavelets in its continuous form with the application of J48 algorithm is analyzed. Vibration signals are extracted from the experimental setup. The continuous wavelet transform (CWT) is calculated for different families and at different levels which form the feature set. The features are then fed as an input to the classifier (J48 algorithm, a WEKA implementation) and the classification accuracies are calculated. Then, the results are validated to find classification capability of CWT features for monoblock centrifugal pump. The different faults considered for this study are cavitation (CAV), impeller fault, bearing fault (FB) and both bearing and impeller fault.