Bearing Fault Diagnosis Based on Convolutional Neural Networks with Kurtogram Representation of Acoustic Emission Signals

Early detection of rolling-element bearings faults is essential, and acoustic emission (AE) signals are actively utilized for monitoring bearing health condition. Most existing methods for fault diagnosis comprise two steps: feature extraction and fault classification. The convolutional neural network (CNN) is a powerful deep learning technique that can perform both feature extraction and classification procedures without the need to separate these tasks into different algorithms. However, most of the known CNN architectures are used for image recognition and require a 2-D image as an input parameter. To employ CNN to resolve the problem of rolling-element bearings fault diagnosis, in the present work, the raw 1-D AE signal is transformed into a 2-D kurtogram representation. Experimental results using eight types of various bearing conditions indicate that the proposed fault diagnosis approach utilizing the kurtogram representation of the original AE signal and CNN extracts discriminative features and achieve high classification accuracy.