Feature Extraction for Bearing Fault Diagnosis in Noisy Environment: A Study

Operating induction machine throughout the year is vital, thus maintenance and diagnosing within stipulated time is very essential. Else, this leads to complete shutdown of the systems and huge production losses. Thus a robust diagnosing scheme is essential for mechanical fault detection at incipient stages, as the probability of mechanical fault occurrences is high. In this work normal bearing, natural cage defective bearing and natural outer race defective bearings are considered. Prior to classification, the presence of fault is validated using spectra of current and vibration signals. This work demonstrates a simple fault diagnosing scheme using 3 different classifiers. 12 statistical time domain features are extracted from vibration and current signals. Feature plots of all conditions are investigated in detail. Experimental results showed that features slope sign changes, variance and standard deviation are performing good irrespective of the signal recorded and NB classifier performs well irrespective of the signal recorded and the feature employed.