CEEMDAN-based Single-Channel Blind Source Extraction for Bearing Fault Detection and Diagnosis

Abstract

Bearings are key components for power transmission in industrial applications. However, how to accurately detect bearing status and perform fault diagnosis in intense background noise remains a challenging task. This paper presents a single-channel blind extraction algorithm for bearing fault detection and diagnosis. First, the vibration signal is decomposed into several intrinsic model functions (IMFs) and a residual. Using a kurtosis index technique to filter the decomposed IMF components, thereby building a multi-channel feature set and overcoming the problem of mode mixing. Then, a blind extraction technology is used to extract fault features from the virtual multi-channel mixed signals. Furthermore, the maximum energy amplitude at the fault feature frequency is obtained to identify fault features by using envelope spectrum analysis. Finally, determining the type of fault in the bearing through the magnitude of singular value entropy to achieve fault diagnosis. Experimental results based on the public bearing fault signal datasets verify the feasibility and superiority. It can make the pulse component in the measurement signal more prominent and suppress the noise component entirely, achieving a powerful de-noising ability. The signal-to-noise ratio has been increased by a maximum of53.35%, and the root mean square error has been decreased by a maximum of 10.86%