A small-sample fault diagnosis approach for RV reducer with RP-SST parallel CNN-SVM model

For addressing the problem of low accuracy in the available diagnostic methods due to the insufficient fault-sample, this article proposes a small-sample fault diagnosis method with Recurrence plot (RP), Synchrosqueezed transform (SST) and parallel convolutional neural network (CNN) and support vector machine (SVM). This method first utilizes RP and SST techniques to extract the multimodal features into image processing. Then, the parallel CNN is used for multimodal feature fusion, enabling it to utilize more fault information within a limited number of samples. Finally, SVM is used for fault classification. The fault experiments for RV reducer were performed to verify the effectiveness of the proposed fault diagnosis method. The complementarity of RP-SST in multimodal feature extraction was verified for solving the problem of insufficient feature extraction due to sample size limitations effectively. Further, the SOTA fault diagnosis model showed that the proposed method achieved a fault diagnosis accuracy of 96.26% in small sample situations, significantly better than CNN-BIGRU, CNN-LSTM, and Resnet model.