Fault Diagnosis of Rolling Element Bearing from Vibration Signal Using One Dimensional Convolutional Neural Networks

Failures of rolling element bearings can precipitate serious downtime. To ensure operational reliability in rotating machinery, it is necessary to diagnose the bearing failures accurately. The essential part of effective fault diagnosis lies in the detection of periodic impulses in the vibration signal. These impulses are characteristic of local faults and are directly related to the defect location in the bearing. Vibration analysis is widely employed to diagnose rolling bearing faults in condition monitoring. In this paper, a One- Dimensional Convolutional Neural Network (lD-CNN) was proposed for automatic feature extraction in the bearing fault diagnosis. The vibration dataset used here is from the Case Western Reserve University (CWRU) Bearing Data Center. The raw vibration data is augmented before input into the network. Both time and frequency domain signals were used as the inputs for feature learning. The results showed that the ID-CNN model can extract and classify bearing faults effectively. The average maximum accuracies of the results from the time domain input and frequency domain input were 98.97% and 99.75%, respectively. It can be concluded that the frequency domain input is superior in capturing anomalies in the bearing vibration signal.