Intelligent Fault Diagnosis of Rolling Element Bearings Based on HHT and CNN

Automatic and accurate identification of rolling bearings fault categories and fault severities is still a major challenge in fault diagnosis. In this paper, a deep learning based approach is presented to translate traditional diagnostic methods based on one-dimensional time-series analysis into graphical images for fault type and severity identification, with rolling bearing as a representative example. Specifically, time sequences of vibration signals are first converted by Hilbert-Huang transform (HHT) to time-frequency images. Next, a convolutional neural network (CNN) learns fault-sensitive features in the time-frequency domain from these images and performs fault classification. Experiments on bearing data demonstrates effectiveness and efficiency of the developed approach with a classification accuracy 95%.