Bearing faults diagnosis using cepstral analysis and 1D Convolutional neural network

The diagnosis of faults in the rotating machines has become necessary recently, in the order to ensure their safety and efficiency. the rolling bearing is one of the most components prone to failure in the rotating machines. In this work, we propose a novel approach to detecting and classifying the rolling bearing faults by using the cepstral analysis and 1D-CNN. First, the real, complex and power cepstrum are calculated, which are later used as input to the classifier. Second, a 1D-CNN is used as a classifier to diagnose the bearing faults. The proposed method is tested on the CWRU dataset from bearings under variable working conditions. Results of the proposed method gave a testing accuracy of 97.5 % for the complex cepstrum method and it also gave a testing accuracy of 99.88% for the real and the power cepstrum