Bearing Fault Diagnosis Based on Multi-domain Feature Fusion
Kaifeng Liang, Sen Zhang, Wendong Xiao · 2025
In order to address the challenge of distinguishing fault states solely through raw time-domain signals in rolling bearings, this paper proposes a multi-domain feature fusion-based fault diagnosis method. This method integrates features from the time domain, frequency domain, and time-frequency domain to enhance diagnostic accuracy. The methodology is initiated with the construction of a multi-domain feature extractor, which is designed to capture the salient features of vibration signals across diverse transform domains. Secondly, the extracted multi-domain features are input into an ML-BiLSTM-Cross Attention model for feature fusion. Finally, the proposed method is employed to achieve fault diagnosis through a Softmax layer. The experimental results demonstrate that the proposed method exhibits excellent generalisation performance, enabling rapid and accurate diagnosis of fault categories, with an average fault diagnosis accuracy of 99.1%.