Classification of Bearing Faults by Approximation of Peak-To-Peak Amplitudes Distribution

To diagnosis the bearings faults in electric machines, the common approach is to acquire vibration signals from the machine-mounted accelerometers and process them by machine learning (ML) algorithms. Although many features can be extracted from signals to perform the classification, methods that require a minimum number of features are of interest to facilitate real-time processing. As a promising technique, we present an approach involving approximation of the vibration signals peak-to-peak amplitudes distribution with a Weibull model, where two parameters of the model are considered as features. The resulting accuracy of the proposed classifier reached 99.6%, which is superior to other considered features pairs from time and frequency domain.