SVM BASED FAULT DIAGNOSIS OF MONOBLOCK CENTRIFUGAL PUMP USING STATIONARY WAVELET FEATURES

Fault diagnosis of monoblock centrifugal pump essentially forms a pattern recognition problem. There are three important steps to be performed in pattern recognition namely feature extraction, feature selection and classification. In this study, stationary wavelet transform (SWT) is used for feature extraction and SMO algorithm (a WEKA implementation of Support Vector Machine (SVM) algorithm) is used for classification. The different fault conditions considered for the present study are cavitation (CAV), impeller fault (FI), bearing fault (BF) and both impeller and bearing fault (FBI). The representative signal is acquired for all faulty conditions, features are extracted, classified and the results are presented. The experimental set up and the procedure for conducting the experiments are discussed in detail.