Rolling Bearing Fault Diagnosis Method Based on an Improved 1DCNN-Transformer

To address the frequent occurrence of multiple fault types, the difficulty of feature extraction, and the susceptibility to noise interference in rolling bearings under complex operating conditions, this paper proposes a fault diagnosis method based on an improved one-dimensional convolutional neural network (1DCNN) integrated with a Transformer architecture. This approach leverages the 1DCNN to efficiently extract local impact and energy features from vibration signals, while the improved Transformer enables global modeling of long-range temporal dependencies, thereby significantly enhancing the recognition accuracy for multi-class fault signals and the generalization capability of the model. Experimental data are sourced from the Case Western Reserve University bearing fault dataset, with multi-channel vibration signals subjected to preprocessing and balanced sampling, and various types of simulated noise systematically introduced to comprehensively verify the noise robustness of the proposed model. Experimental results on the public dataset demonstrate that the improved 1DCNN-Transformer model achieves a classification accuracy of 99.43%, markedly outperforming traditional methods such as ANN, CNN, LeNet, and SVM. Further t-SNE visualizations and confusion matrix analyses reveal the method’s superior feature discrimination and high-precision performance across multiple fault categories. Tests under strong noise conditions further indicate that the model exhibits high robustness and excellent potential for engineering applications. In summary, the proposed method provides an efficient and reliable solution for intelligent fault diagnosis of rolling bearings in complex environments and lays a solid foundation for future model development and industrial deployment.