Data-Driven Tool Condition Monitoring Platform for Grooving Operations with Chip Control Issues

Accurate Tool Condition Monitoring (TCM) is key to maximizing tool life and minimizing downtime in machining. However, real-time tool wear assessment remains challenging, particularly in turning, due to non-periodic signals and chip entanglement issues. This study frames tool wear prediction in grooving operations as a supervised regression problem using machining monitoring data as inputs. Multi-source sensor data were acquired and processed using an edge platform during groove machining to train six machine learning regression models. Also, a feature and sensor importance analysis was performed to identify opportunities to simplify instrumentation and reduce industrial implementation. The results demonstrate that Random Forests and XGBoost consistently achieve high accuracy (R2 > 0.9), effectively handling the non-linear, noisy nature of turning data. Based on these results, a final ensemble model combining both approaches was defined and deployed at the industrial edge as part of a final condition tool monitoring system. To this end, the ensemble model was encapsulated within a Docker-based container and was executed on an edge computing platform, enabling reliable, portable, and scalable real-time deployment for TCM.