Research on Gas Turbine Rotor Fault Diagnosis Method Based on VIT Model and Transfer Learning

Aiming at the problems of difficulty in fault data collection and scarcity of fault samples in gas turbine rotor fault diagnosis, a gas turbine rotor fault diagnosis model based on VIT model and transfer learning is proposed. Firstly, the acquired one-dimensional vibration signal is converted into a two-dimensional time–frequency image by continuous wavelet transformation as the input of the VIT model, and the fault diagnosis is realized through its recognition of time–frequency image and feature extraction. Using CRWU bearing dataset to pre-train the model combined with transfer learning, only the weights in the MLP Head structure are migrated by freezing the part before the MLP Head structure and using the gas turbine rotor dataset fault classification, the results show that the fault recognition accuracy reaches 99.5% after transfer learning, and the superiority of the proposed model is proved by comparative experiments.