Digital twin-driven fault diagnosis for CNC machine tool

Traditional data-driven fault diagnosis methods require a massive amount of data to train diagnosis models. However, the complex and coupled structure of CNC machine tools makes it difficult to obtain enough usable data. Current data generation methods ignore actual operating conditions and have imbalance, which reduces the accuracy of fault diagnosis. To tackle these problems, this paper presents a digital twin-driven fault diagnosis method for CNC machine tools. Firstly, a digital twin model of a CNC machine tool is established and validated. Then, a twin model library is constructed to include multiple twin models under different fault status. A model data fusion method is presented, using the decision tree algorithm Classification and Regression Tree (CART) to train a model selector and actual sensor data as input to select the optimal model from the library and realize fault diagnosis with the model. Finally, taking the CNC machine tool spindle as an example, the stiffness deterioration of the spindle during operation is effectively diagnosed, which verifies the effectiveness and feasibility of the proposed method.