A Machine Learning Approach to Fault Diagnosis of Rolling Bearings

Abstract

This paper presents a method based on classification techniques for automatic fault diagnosis of rolling element bearings. Experimental results achieved on vibration signals collected by an accelerometer on an experimental test rig show that the method can automatically detect different types of faults. Furthermore, the method is able, once trained on an appropriate representative set of basic faults, to recognize more serious faults, provided they are of the same type. We also analyzed the trend of correct classification of bearing faults on variation of the signal-to-noise ratio achieving high levels of robustness.