This invention discloses a bearing fault diagnosis method, a
hybrid machine, and a monitoring method. The bearing fault diagnosis method includes the following steps: constructing a
signal generator based on a rolling bearing fault dynamics model to generate synthetic vibration signals labeled with fault type and severity; constructing a bearing fault diagnosis model with
label preservation constraints, adapting the statistical characteristics of the synthetic vibration signals to the target field domain, while ensuring that the fault semantic labels of the synthetic vibration signals are fully preserved during the conversion process, achieving style transfer of unpaired synthetic vibration signals; mixing the style-transferred high-fidelity
synthetic data with the field-
labeled data to
train the bearing fault diagnosis model, and employing a hierarchical progressive fine-tuning strategy to achieve
domain adaptation, obtaining a well-trained bearing fault diagnosis model, and finally realizing bearing fault diagnosis. This invention has been successfully applied in industrial fields, yielding significant
economic benefits and providing a complete solution for the intelligent operation and maintenance of large and complex rotating machinery.