Deep Learning based Fault Classification Algorithm for Roller Bearings using Time-Frequency Localized Features

Abstract

The paper proposes an algorithm to classify different conditions of a bearing based on vibration data using a deep convolutional neural network. Spectrograms of vibration data are generated by means of Short-time Fourier Transform and then provided as input to a convolutional neural network. The network is successful in predicting the health condition of the bearing from the spectrograms and achieves a classification accuracy of 97%. The trained model is then tested on a different dataset and the model is able to predict the classes with an accuracy of 96%. The proposed model is finally compared with pre-existing models to evaluate its performance and the results demonstrate the state of the art performance of our proposed algorithm.