A rolling bearing fault diagnosis method based on multi-source feature fusion

This paper presents a novel approach for fault diagnosis in rolling bearings using a multi-source feature fusion method. The proposed method aims to enhance the diagnostic accuracy by integrating both implicit and explicit features extracted from vibration data. Initially, one-dimensional Convolutional Neural Networks (1D CNNs) are employed to extract implicit features from the raw one-dimensional vibration signals. These features capture the complex patterns and correlations within the vibration data that are difficult to discern manually. Subsequently, explicit time-frequency domain features, such as mean and variance, are computed from the same vibration data. The implicit and explicit features are then fused to create a comprehensive feature representation that captures both the deeplevel patterns and the statistical properties of the vibration signals. The fused feature set is fed into a softmax classifier for fault classification. The performance of the proposed method is evaluated using the Case Western Reserve University (CWRU) bearing dataset, which contains various fault types and severity levels. Experimental results demonstrate that the proposed multi-source feature fusion approach achieves superior diagnostic accuracy compared to traditional methods that rely solely on either implicit or explicit features. The combination of 1D CNN-extracted implicit features and statistical explicit features provides a robust and effective solution for rolling bearing fault diagnosis.