A Novel Bearing Fault Diagnosis Method Based on the DLM-CNN Framework

Bearings are critical components of rotating equipment with a high failure rate. Detecting faults in bearings is essential for improving the reliability and safety of rotating mechanical systems. In this paper, by combining the ability of deterministic learning mechanism (DLM) to accurately model the intrinsic dynamics of data and the automatic feature extraction capability of convolutional neural network (CNN), we propose a DLM-CNN framework for bearing fault diagnosis. To be specific, the inherent dynamics of raw vibration time series are first learned/modeled by the deterministic learning mechanism. The learned dynamics information is transformed into images and then fed into the CNN for automatic feature extraction. The discriminative features extracted by the CNN are used to efficiently classify health types as well as different bearing failure types. To verify the effectiveness of the proposed method, experiments are conducted on the Case Western Reserve University (CWRU) dataset with a classification accuracy of 98.89%.