A multi-sensor tool wear monitoring method based on mechanism-data fusion for industrial scenario

Abstract

In industrial manufacturing scenarios, tool wear directly impacts product quality, production efficiency, and machine reliability. However, the deployment of traditional tool condition monitoring (TCM) solutions is hindered by their reliance on invasive, costly sensors and complex configurations. This study presents an innovative, industry-oriented multi-sensor TCM framework that integrates mechanism-based modeling with data-driven intelligence. A low-cost, non-invasive sensing system is developed for compact CNC lathes, combining spindle power and vibration signals. A cutting force coefficient k_c derived from power signals is introduced as a physically interpretable indicator, capable of identifying the onset of severe tool wear under varying machining conditions. This coefficient is fused with vibration-based statistical features through a gated recurrent unit (GRU) network enhanced by attention mechanisms, forming a hybrid model for real-time tool wear estimation. Extensive experiments in realistic industrial settings validate the method’s robustness and scalability. Compared to conventional approaches, the proposed mechanism-data fusion model improves interpretability, enhances prediction accuracy, and enables deployment in cost-sensitive production lines. This research provides a novel pathway toward intelligent, explainable, and deployable TCM systems for industrial scenarios.