Fault Diagnosis Method for Planetary Gearboxes Based on SIFT-BoW and IResNext

Composite fault diagnosis is a hot spot and a difficult problem in the field of planetary gearbox fault diagnosis. This work applies image processing technology and deep learning to a new method of planetary gearbox composite fault diagnosis based on the model HFXZ-I planetary gearbox solid fault diagnosis platform to solve this issue. First, the original vibration acceleration signal is converted into an image using an image recognition algorithm. Second, the scale-invariant feature transform (SIFT) algorithm is applied to extract the feature points of the image and describe them vectorially. Then, the SIFT algorithm is applied to extract the feature points of the image and describe them vectorially, the bag-of-words model (BoW) is formed. Finally, the extracted features are input to an improved inverse residual network (IResNext) for diagnosing planetary gearboxes, which preserves the features while avoiding network degradation. The SIFT-BoW and IResNext-based approaches are shown to have superior stability and higher accuracy when comparing other fault identification techniques. The approach performs similarly when tested against various experimental datasets from simulation and physical experiments, demonstrating the method’s stability on different datasets.