Intelligent real-time tool life prediction for a digital twin framework

Abstract A key challenge in the machining manufacturing industry is real-time tool wear prediction, as conventional methods rely on conservative tool changes, causing premature replacement or excessive wear that risks failure, part damage, or poor surface quality. Monitoring and predicting the wear condition of a cutting tool is key to guarantee the cutting quality and saving costs. This study presents an AI-driven digital twin framework for real-time tool life prediction to address these limitations by integrating multiple modules. These modules include an on-machine direct inspection system, a seamless connectivity integration module for real-time data management, and a deep learning module for tool wear prediction. Long Short-Term Memory networks were trained, optimised and tested on a milling dataset to then deploy onto a real-time implementation of the digital twin framework. A comprehensive design of experiments (DOE) was used to validate the real-time tool life prediction framework of a dynamic milling toolpath strategy of a Ti-6Al-4 V alloy. The models were able to predict tool maximum flank wear based on sensor data from the machining tests DOE with RMSE of 33.17 µm, whilst the real-time implementation yielded a minimum of RMSE of 119.36 µm. These results motivate further research for enabling real-time closed-loop control for a future digital twin system implementation.