Tool Wear Prediction Using Informer Architecture for Multi-Source Sensor Data Fusion

In the field of machining, precise monitoring of tool wear status can fully utilize tool life. On the one hand, it prevents tool replacement before the tool reaches its service life, which increases costs. On the other hand, it is necessary to monitor the tool wear status during the cutting process in order to avoid tool replacement due to severe tool wear, which affects the machining accuracy of the workpiece. In addition, based on monitoring the current tool wear status, predicting future wear values can enable machine tools to issue corresponding warnings and replace tools in a timely manner before tool wear and damage occurs, while ensuring machining quality and improving production efficiency. Therefore, this article has conducted research on the monitoring and prediction algorithms of tool wear status, and the main research contents are summarized as follows: （1）Based on the theoretical foundation of tool wear related technologies, the variation trends of various sensor signal values and wear values in experimental data were analyzed. Preprocessing of sensor signal data was carried out, including removing invalid values, modifying outliers, etc., and optimizing the signal through wavelet threshold denoising, laying the data foundation for the establishment of subsequent tool wear models. （2）Propose a comprehensive model for multi-step prediction of tool wear. Establish a monitoring model based on Informer to monitor the wear value at the current time point through multi-sensor information. Considering that in actual machining, the tools and machining states at two closer time points are closer, that is, the short-term information at the time point closer to the predicted point has a greater impact on the predicted point than the long-term information at the time point farther away, an Informer prediction model with Attention mechanism is introduced. The encoding decoding structure can achieve multi-step prediction of tool wear values to obtain more sufficient time for tool control. Verify the effectiveness and advantages of the comprehensive model through experiments.