Fault diagnosis of harmonic drive based on efficientnet-B0

Harmonic drive is a commonly used transmission device in the robotics and automation industries. In recent years, the rapid development of artificial intelligence technology has led to the application of deep learning in the field of machine fault diagnosis, achieving good results. This study proposes a fault diagnosis method based on the lightweight convolutional neural network EfficientNet-B0. First, the collected vibration signals are transformed into 128×128 time-frequency feature maps through continuous wavelet transform. The transfer learning strategy is then employed to freeze the first 50 layers of the pre-trained model. The network depth and width are optimized using a compound scaling strategy, and the network parameters are fine-tuned. Finally, the method is validated on a self-built harmonic drive dataset. The experimental results show that this method achieves an average diagnostic accuracy of 94.5% on the test set, with a 42% reduction in model parameters compared to traditional ResNet-50 and SVM methods. It provides a feasible technical solution for the intelligent diagnosis of industrial equipment, which is of great practical significance and application value.