An incipient fault diagnosis and detection method for bearings based on feature fusion of acceleration and acoustic emission sensors

Abstract

Acoustic emission (AE) sensors demonstrate significant advantages in identifying metal fatigue cracks. However, their application in testing and studying the fatigue process of bearings remains insufficiently explored. This study validates the effectiveness of AE sensors in identifying bearing fatigue faults and proposes an integrated approach for incipient fault diagnosis and detection in bearings by fusing features from acceleration and AE sensors. First, multidomain feature information was extracted from sensor data. A comprehensive index was designed to select sensitive features for bearing faults, considering both feature correlation and standard deviation. Subsequently, a weight matrix for features was computed using the Softmax function, enabling the fusion of multisensor feature data based on the weight matrix. The fused features, comprising a small amount of labeled and a large amount of unlabeled data, were input into the semisupervised learning model S4VM for training and real-time incipient fault diagnosis. Fault location was then determined by analyzing the envelope spectrum signal characteristics at the fault occurrence moment. Finally, the proposed method was validated through an accelerated bearing fatigue life test. The results indicate that the method can accurately identify the time and location of incipient bearing faults, while addressing issues related to the redundancy of multisensor data and the allocation of fusion weights. This method offers significant guidance for the predictive maintenance of bearings.