Bearing Condition Monitoring Using Classifier Fusion

This paper presents a method which automatically detects and diagnoses defects of rolling element bearings. Four different defects are taken into account, namely, indentation on the inner raceway, indentation on the roll, sandblasting of the inner raceway and unbalanced cage, besides three severity levels of indentation on the roll are considered. The proposed method exploits soft computing techniques, in particular neural networks and multi classifier systems. We use the Fast Fourier Transform to represent the bearing vibration signals recorded in the time domain by four accelerometers. Whenever possible, simple classifiers, such as LDC and QDC, are used to perform both feature selection and classification. On the other hand, to solve particularly difficult classification problems we adopt Multi-Layer Perceptron neural networks and multi-classifier systems. In particular, three different types of methods of classifier fusion are analysed: maximum, minimum and average. The proposed method shows a good sensitivity as it provides accuracy higher than 99 % in all the experiments related to the recognition of different defects and different severity levels of the defects.