Fault Diagnosis Method of Planetary Gearbox Based on Transfer Learning and EfficientNet

In the field of gear fault diagnosis, the traditional convolutional neural networks model has a single precision improvement method and has limitations. This paper proposes a planetary gearbox fault diagnosis model based on Gramian Angular Fields ( GAF ) image coding technology and TL-EfficientNet network model. The collected one-dimensional vibration signal is preprocessed and converted into GAF diagram. The EfficientNet is pre-trained on the ImageNet dataset, the trained feature weights are migrated to the EfficientNet network, and the samples are trained to complete the fault identification of the planetary gearbox. The experimental results show that the proposed method can effectively improve the fault diagnosis accuracy of planetary gearboxes, reaching 99.49%, and has faster convergence speed and better stability than a single method. At the same time, the TL+EfficientNet-B2 model has strong generalization and can be applied to other fault diagnosis with similar feature distribution data.