This paper proposes a rolling bearing fault diagnosis method based on Variational Mode Decomposition–Newton–Raphson (VMD-NR) modal decomposition and multi-feature fusion to address the issue of insufficient accuracy in rolling bearing fault diagnosis under complex noisy environments, which is caused by inadequate feature extraction and weak model representation capabilities. Firstly, based on the Variational Mode Decomposition (VMD) theory, the Newton–Raphson-Based Optimization (NRBO) algorithm is introduced to construct an improved VMD model, enhancing the accuracy and anti-interference capability of fault signal decomposition. Secondly, a Wide-Dilated Convolutional Neural Network (WDCNN)–Informer dual-branch fusion architecture is proposed to collaboratively extract local and global long-sequence features from vibration signals. Through feature-level concatenation and fusion, feature complementarity is achieved, improving the completeness and effectiveness of signal feature representation. Experimental results on the Case Western Reserve University (CWRU) public bearing dataset demonstrate that the proposed diagnostic method achieves a fault classification accuracy of 99.8%, an increase of 5% compared with conventional methods, enabling precise rolling bearing fault classification and providing a theoretical basis for rolling bearing fault diagnosis.
Li et al. (Thu,) studied this question.