Research on bearing fault detection based on convolution neural network

Bearings are widely used in industrial equipment, and their fault detections have tight relationship with the safety production and economic operation. Its running state can influence the performance of the whole machine directly, and the fault of bearing is one of the main factors that lead to the fault of machinery equipment during the running process. Instead of using traditional measures based on signal filtering, decomposition, transformation, feature extraction, and so on, we carried out the research on convolution neural networks(CNN) for fault training and classification. The model architecture and parameters are studied with experiments based on the vibration signal samples of bearing failure. It shows that, the fault detection can reach a high accuracy level based on CNN by simply inputting the original signal plots. Although, long period of model training is needed, the lateral online classification can be simple by taking the benefit of trained model, particularly, no more decomposition, transformation, and feature extraction works need to do.