Prediction of quality class in face milling of S355J2 steel using meta model based on cutting temperature and surface roughness data

In this study, the effects of cutting parameters on surface roughness and cutting temperature in the face milling process of S355J2 steel were experimentally investigated under different cooling strategies (dry, wet, and air-assisted). The experimental data obtained were analyzed by establishing regression models for the prediction of both cutting temperature and surface quality, and the workpiece quality class was estimated by classification algorithms. Initially, various regression and classification algorithms were evaluated as benchmarks, and then the most successful models were subjected to hyperparameter optimization with the Optuna algorithm. According to the experimental findings, the lowest surface roughness was obtained under air cooling conditions, showing a significant improvement of 19.81 % and 20.21 % compared to dry and wet machining, respectively. Conversely, the lowest cutting temperature was observed in wet cutting, which was 78.84 % lower than dry cutting and 62.92 % lower than air-assisted machining. In modeling studies, NGBoost for cutting temperature and LightGBM for surface roughness algorithms showed the highest performance, achieving R2 values of 0.816 and 0.698, respectively, after hyperparameter optimization. The highest success in quality class prediction was achieved with the Extra Trees algorithm, which reached an accuracy of 73.85 % as a benchmark model. The combination of these algorithms created a metamodel structure that excelled in quality classification, achieving a final accuracy rate of 92.3 %. Additionally, the transparency of model decisions is increased with LIME analysis, and the effects of basic cutting parameters on the outputs are visualized. As a result, it is shown that the proposed approach can be used as an effective tool for the optimization of milling processes and the development of automatic quality control systems in the production line. © 2026 the author(s), published by De Gruyter, Berlin/Boston.