Online Fault Diagnosis of CNC Machine Tools Based on Deep Learning

Bearings and cutting tools are the core components of CNC machine tools, due to their complex working environment, high-speed requirements, and variable stress during daily operation, they are prone to malfunctions. Traditional fault diagnosis methods require human construction of features, and the diagnostic process is cumbersome and lacks intelligence. In response to this issue, the author has developed an online fault diagnosis system for CNC machine tools based on deep learning technology, focusing on the cutting tools and bearings on the machine tool spindle, in response to the characteristics of small differences and strong dynamic timing in monitoring data of CNC machine tools, the author proposes a one-dimensional convolutional sequence fault diagnosis model by combining a one-dimensional CNN network with strong feature extraction ability and an LSTM network with long and short time memory function, and optimizes the structure of the proposed model. The experimental results show that the classification performance of the one-dimensional convolutional sequence model proposed by the author is significantly better than that of one-dimensional CNN and LSTM networks, and the optimized model improves the accuracy of fault classification by 8.42%. Aiming at the problem of unbalanced experimental data samples, based on the one-dimensional convolutional sequence model, the Loss function of the model is optimized through cost sensitive learning to achieve adaptive updating of the weight of categories and improve the sensitivity of the model to small class samples. Experiments show that the classification performance of the model is improved by 10.11% after Loss function optimization. In addition, to verify the normal operation of the online fault system designed in the paper, the system data transmission and model diagnosis accuracy of the diagnostic system were tested and verified in a certain manufacturing workshop. The research results found that the classification accuracy reached 95.11%, which can monitor the current wear status of spindle bearings and tools in real time, avoid damage to equipment caused by CNC machine tools working in abnormal conditions, and ensure equipment safety.