APPLICATION OF MACHINE LEARNING METHODS FOR MODELING THE QUALITY AND STABILITY OF THE TURNING PROCESS

Purpose. Development and validation of a predictive modeling methodology that enables multi-objective analysis of the turning process by simultaneously predicting two key parameters: average surface quality and process stability (variability). Research methods. A comparative analysis of models based on Multiple Linear Regression (empirical formula) and the Random Forest algorithm was conducted. The models were trained on an open experimental dataset for 42CrMo4+QT steel. Accuracy was evaluated using R2 and MAE metrics on a test set. Validation was performed on an independent dataset. Results. The Random Forest model demonstrated slightly higher predictive capability for average roughness (R²=0.59 vs. 0.53) and significantly higher capability for process stability (R²=0.139 vs. negative values for the formula). The dominant influence of feed rate and tool nose radius on quality was established, as well as the key role of cutting speed and tool geometry on stability. Scientific тovelty. An approach for the simultaneous modeling of the quality and stability of the turning process is proposed. The substantial advantage of flexible ML models over classical regression for analyzing stochastic aspects of the process, such as process variability, has been quantitatively proven. Practical Value. The developed methodology serves as a tool for the multi-objective optimization of cutting parameters. Recommendations for improving process reliability have been formulated: increasing the depth of cut to enhance productivity, avoiding low-speed regimes (v<95m/min), and using a tool with a nose radius of r=0.8mm, which reduces the maximum expected process variability by more than 16%.