A physics-guided unified damage index framework for explainable milling anomaly detection using interval-wise MTConnect multi-sensor fusion

Purpose Reliable anomaly detection in precision milling is essential for maintaining dimensional accuracy, tool life, surface integrity and component reliability. Conventional data-driven approaches often average sensor signals over entire machining records and rely on black-box machine learning models, reducing physical interpretability and limiting industrial deployment. This study proposes a physics-guided Unified Damage Index (UDI) framework integrating interval-wise MTConnect multi-sensor fusion with explainable machine learning for interpretable milling anomaly detection. Design/methodology/approach MTConnect sensor signals were extracted only from active cutting intervals using machining start and end timestamps to eliminate idle machine states. Process parameters, including spindle speed, feed rate, depth of cut and tool geometry, were combined with spindle vibration, axis vibration, spindle load and power consumption to formulate a physics-guided Unified Damage Index. An XGBoost classifier was developed for anomaly prediction, while SHAP (SHapley Additive exPlanations) interpreted feature contributions. Model robustness was evaluated using repeated stratified five-fold cross-validation and bootstrap confidence interval analysis. Findings The proposed framework achieved 93.75% accuracy, 91.7% precision, 100% recall and a 95.0% F1-score on the independent test set without missing anomaly cases. Repeated cross-validation yielded a mean accuracy of 88.85 ± 8.04%, demonstrating stable predictive performance. SHAP and feature importance analyses identified spindle vibration instability and X- and Y-axis vibration as the dominant indicators of machining anomalies. Originality/value The framework integrates interval-wise MTConnect feature extraction, a physics-guided Unified Damage Index, explainable XGBoost modelling and statistical robustness analysis into a unified predictive maintenance approach, providing an interpretable solution for intelligent manufacturing and digital twin applications.