Residual‑Based Monitoring with Incremental Model Update for CNC Milling Processes

Flexible CNC milling requires process monitoring systems that remain reliable under changing workpieces, materials, and machine conditions. This paper proposes a monitoring framework that combines signal-level process prediction, residual-based anomaly detection, and data-centric incremental model maintenance. A hybrid predictive model generates axis- and spindle-specific current references from kinematic and physically informed process features, while interpretable statistical detectors operate on the resulting residuals to identify process anomalies. Two experiments are conducted to first define the monitoring model and second to quantify the integration of an incremental learning with a density-based data selection approach, active-learning-based sampling, and physics-informed dataset optimization. First, established statistical monitoring methods are compared for deployment within the residual domain, where Statistical Process Control achieves the best overall trade-off between detection quality and stability. Second, the full monitoring pipeline is validated under progressive domain shift in an incremental setting across two machines. ... mehrThe results show that all adaptive variants significantly outperform the static baseline, confirming the necessity of maintaining models in flexible production. The adaptive variants differ more in their temporal progression, stability, and data efficiency than in their aggregated performance. Overall, the study demonstrates that residual-based monitoring combined with data-centric adaptation provides an interpretable and deployment-oriented solution for robust CNC process monitoring under evolving production conditions.