Running Gear Global Composite Fault Diagnosis Based on Large Model

The running gear is a key component of rail vehicles, and timely diagnosis of faults in the running gear is crucial. It is very difficult to use local sensor data for global single fault and composite fault diagnosis. Therefore, this article proposes a multipart composite fault diagnosis method based on large model (CFD-LM). First, the time domain feature sample selection method based on the Pearson correlation coefficient proposed in this article is used to remove signals with less information about other parts. At the same time, the filtered data are reconstructed to make it suitable for the pretrained large model. Then, the processed data are input into a pretrained large model, which is used to extract weak fault information of other components contained in the data for accurate classification. During the training process, the cosine annealing algorithm is used to dynamically adjust the learning rate, preventing the training model from falling into local optima. Finally, public datasets are used for validation. The results indicate that CFD-LM has higher accuracy in global composite fault classification than the latest deep learning algorithms. This proves that CFD-LM can effectively use local data for global composite fault diagnosis.