Anomaly Detection based on Broad Leaning System for Rolling Element Bearing Fault Diagnosis

Le Yang, Zelin Yang, Ziwei Zheng, Lijun He, Fan Li, C. L. Philip Chen

Abstract: The latest proposed Broad Learning System (BLS) demonstrates an efficient and effective learning capability in many machine learning problems. In this paper, we develop a BLS based anomaly detection method (AD-BLS) for rolling element bearing fault diagnosis. Our method applies wavelet packet transformation (WPT) to extract the energy features, which will be further fed to the proposed BLS-based model to learn the intrinsic representations for the normal states of rolling bearings to conduct anomaly detection. Benefiting from the WPT pre-process, AD-BLS can be effectively trained with a small size of data. Moreover, due to the online updating mechanism of BLS, the designed AD-BLS can be easily updated with new extra data without re-design or retraining from scratch, which largely improves the flexibility of our method in real-world applications. Experimental results on the rolling element bearing fault diagnosis dataset further demonstrate the effectiveness of our method.