Tool Wear State Identification Technology Based on Spindle Power Signal

With the increasing demand for state monitoring of the processing process in intelligent manufacturing, online detection technology for tool wear has become a key link to ensure processing quality. In view of the current situation that traditional tool wear monitoring and life prediction methods often rely on experienced operators for visual inspection or simple mechanical measurement, which are not only time-consuming and labor-intensive but also difficult to achieve real-time monitoring and accurate prediction, this study proposes a tool wear identification technology based on spindle power signals. By using the variational mode decomposition (VMD) algorithm to perform adaptive mode decomposition on the original power signals, and considering the mapping relationship between the spindle power signals and the tool wear state during the tool wear process, data-driven technology is introduced. An LSTM neural network model is constructed to achieve tool wear state identification technology. Experiments collected the spindle power data of three stages including initial wear, normal wear and severe wear of a four-edge milling cutter with a diameter of 8 during the cutting of 304 stainless steel. A wear amount true value database was established through multi-sensor synchronous calibration. The results show that compared with the single LSTM model (77.78% accuracy), the accuracy of this model in identifying 0.2mm wear amount reaches 96.3%. The research confirms that there is a strong correlation between the features of spindle power signals and the tool wear state. The proposed VMD-LSTM hybrid model has significant advantages in feature decoupling and time series modeling, providing a new idea for non-intrusive tool monitoring.