Deep learning enhances digital twin of machine tools: a case study on improving grinding accuracy for large shaft

Improving the machining accuracy of machine tools is a challenge in practice, and digital twins provide a feasible approach. However, a very difficult issue is that some important data cannot be directly measured during the machining process, which significantly affects the implementation of digital twins. In this paper, by taking large shaft grinding as an example, an approach using deep learning to enhance the digital twin of machine tools is proposed. An enhanced digital twin framework of a grinder is built, where the twin model of grinding accuracy was established by considering the system stiffness and grinding force pattern. Then, the CNN-LSTM neural networks with acceleration and power signals as input were used to deal with the time-varying characteristics of the core parameters to dynamically correct the twin model during the machining process. The corrected twin model was further used to generate the compensation instruction to improve the grinding accuracy. The experimental results show that the grinding accuracies of large shafts with three different profiles are improved by more than 80%, which proves the effectiveness of the proposed approach. Moreover, related discussion and outlook have been provided to help understand that deep learning enhances the digital twin of machine tools.