A hybrid WSST–DenseNet framework for accurate and robust bearing fault diagnosis

Abstract Bearing fault diagnosis is crucial for ensuring the safe and stable operation of industrial systems, as key components in rotating machinery. To address the problem that traditional methods rely on manual feature engineering and have limited accuracy, this study proposes a novel bearing fault diagnosis method based on a combination of time-frequency analysis, deep feature extraction, feature dimensionality reduction and machine learning. First, wavelet synchrosqueezed transform (WSST) is used to transform the one-dimensional signal into a time-frequency image to obtain rich and robust time-frequency representation. Then, the DenseNet-201 pre-trained deep learning model is used for deep feature extraction. Finally, the bearing fault diagnosis performance of nine cases consisting of three dimensionality reduction methods and three machine learning methods is systematically evaluated. Experimental results show that the WSST method performs best in feature representation, DenseNet-201 performs best in deep feature extraction, and Case achieves the best bearing fault diagnosis performance under ten-fold cross-validation, with an average accuracy of 98.56%, which is superior to existing state-of-the-art methods. The novel bearing fault diagnosis method proposed in this study not only has advantages in accuracy and stability, but also shows good potential for engineering applications.