Intelligent fault diagnosis method for rolling bearings based on VMD-CNN-Transformer
作者
Ao Li,Dongao Li,Ting Fang,Jiaming Han
出处
期刊: [Institute of Electrical and Electronics Engineers] 日期:2025-12-19卷期号:: 223-229被引量:1
标识
DOI:10.1109/mlcipr68329.2025.11407307
摘要
As the core component of rotating machinery, the vibration signal of rolling bearings usually presents nonlinearity and non-stationary characteristics and is easily overwhelmed by strong background noise, which makes it difficult to extract weak fault features in the early stage. To solve this problem, a hybrid fault diagnosis method combining Variational Mode Decomposition (VMD) and CNN-Transformer is proposed. Firstly, aiming at the modal aliasing problem of the original signal, the VMD algorithm is used to decompose the complex vibration signal into multiple intrinsic mode functions (IMFs) with specific center frequencies to achieve signal-noise separation and feature enhancement. Subsequently, a serial hybrid deep learning model is designed: the one-dimensional convolutional neural network (1D-CNN) is used to adaptively extract the local morphological features of modal components, and the Transformer encoder is introduced to capture the global dependency of the fault signal over a long time series, thus making up for the shortcomings of the receptive field limitation of the traditional convolutional network. The verification results based on the Western Reserve University (CWRU) dataset show that the recognition accuracy of the proposed method reaches 99.48% in 10 different health states. Crucially, quantitative comparative experiments demonstrate that under a strong noise environment (SNR = -8 dB), the proposed model maintains a high accuracy of 95.12%, and significantly reduces the Root Mean Square Error (RMSE) by 0.17 compared to Bi-LSTM and traditional CNN models, demonstrating superior diagnostic performance and generalization ability under complex working conditions.