Study of Effective Feature Combination for Fault Diagnosis of Bearings in Motors

Condition monitoring for bearing faults is remaining more vital in the present scenario. This study presents a simple approach for diagnosing various bearing faults of induction motor, which uses effective statistical time domain features. The statistical time domain features like mean absolute value, zero crossing, waveform length, slope sign changes, simple sign integral and Willison amplitude are considered. The performance of features is studied with 6 feature ensembles (FEs) to identify the effective FE for fault diagnosis in the induction motor. This identification is based on the classifications that are performed by using linear discriminant analysis (LDA), naïve Bayes (NB), support vector machine (SVM) and logistic model tree (LMT) classifiers. Each feature ensemble is investigated for 5 datasets derived from a bench mark database for a maximum of 48 faulty conditions. The results indicate the FE constituting six features with SVM classifier outperforms other fault diagnosis techniques.