Channel-temporal reinforcement with global compensation network for tool wear monitoring

Abstract Tool wear monitoring (TWM) techniques are of great significance in intelligent manufacturing, benefiting both machining quality enhancement and production cost reduction. However, the degradation of cutting tools is complex and variable, posing challenges for accurate TWM. Although complex signal preprocessing and feature engineering can enhance TWM accuracy, they are time-consuming, limiting their suitability for real-time applications. To overcome these challenges, a novel TWM method based on channel-temporal recalibration network with global compensation network (CTRGCnet) is proposed. In this framework, a deep convolutional neural network (DCNN) is used as the baseline model for feature extraction. To improve time-series processing, the channel-temporal recalibration module (CTRM) is designed to recalibrate the input features before DCNN learning, while the information reinforcement with global compensation (IRGC) module enhances long-term dependencies. The effectiveness of CTRGCnet on TWM was validated using the prognostics and health management 2010 challenge dataset and the carbon fiber reinforced plastics milling dataset. The plug-and-play capability of the designed CTRM and IRGC modules was assessed through further discussion. The results demonstrated that incorporating the CTRM and IRGC modules significantly enhanced the TWM accuracy of the baseline model, specifically, the root mean square error was reduced by 27.74%–40.89%, and the mean absolute error was lowered by 29.92%–42.66%. Moreover, incorporating CTRM and IRGC effectively reduces both the time and space complexity of the model, offering new insights for lightweight real-time deployment in industrial applications.