Bearing Fault Diagnosis Method Fusing Digital Twin and Multiscale Feature Extraction

In order to improve the accuracy of bearing fault diagnosis and at the same time solve the problem of insufficient model training samples, this paper integrates the digital twin technology and proposes a multi-scale feature extraction bearing fault diagnosis method. Firstly, a four-dimensional framework of digital twin for bearing fault diagnosis is constructed, and based on the principle and dynamics analysis of bearing faults, a digital twin for bearing faults is constructed, and by comparing the eigenfrequencies with the actual signals, it is proved that the digital twin has a high fidelity, and the digital twin data with high credibility can be generated as the training samples of the model. At the same time, a multi-scale cross-domain feature extraction model is proposed, which combines one-dimensional Convolutional Neural Network (1DCNN), Squeeze and Excitation Networks (SENet), and 2D-SwinTransformer. This model can extract features from both time-domain and frequency-domain signals, and through global modeling, it enables the model to focus more on key features. The model achieves an accuracy of 98.9% and performs well in classification across various fault categories. In addition, a bearing fault diagnosis system is constructed by integrating digital twin and multi-scale cross-domain feature extraction methods, and through the accurate response to different faults, it is shown that the system can effectively identify the fault features, which verifies the effectiveness of the proposed method.