Hybrid mechanism-based and neural-network-assisted real-time prediction of surface anomalies in radial immersion micromilling

The transfer of quality control tasks to the manufacturing stage, referred to as in-process quality control, has become an important trend in modern production systems. In response, a neural network assisted system was developed for real-time evaluation of surface roughness during micro milling, including the prediction of surface anomalies, based solely on cutting force signals. The system is based on a mechanistic model used to identify the physical mechanism of anomaly formation and to define an analytical parameterization of the surface profile. The neural network significantly reduced computation time compared to the purely mechanistic approach, making real-time application feasible. Additionally, the proposed system can be used to predict the occurrence of surface anomalies, achieving prediction accuracy exceeding 90%. The highest prediction accuracy was achieved for parameters directly related to profile height, with a relative error below 4%. The presented approach enables real-time monitoring of surface structure and has the potential to support in-process quality control in micro milling.