<![CDATA[Surface roughness is a critical indicator of machining quality that directly affects product performance and service life. However, most existing prediction studies focus on single-material machining and rely on a single predictive model, limiting their effectiveness in real industrial environments where multiple materials are commonly processed. To address this gap, this study proposes an intelligent multimodel system for surface roughness prediction in CNC turning of multiple materials. The experimental investigation was carried out using two commonly applied steels, ST41 and S45C, with 81 machining trials performed for each material. Vibration signals were recorded using a three-axis accelerometer and combined with machining parameters consisting of feed rate, spindle speed, and depth of cut. The acquired signals were analyzed in both time and frequency domains through Fourier transformation, resulting in the extraction of eighteen vibration-related features that were normalized and used as model inputs. Three prediction techniques, namely Multiple Linear Regression, Support Vector Regression, and Artificial Neural Networks, were developed and integrated within the proposed system. System performance was evaluated using Mean Absolute Percentage Error (MAPE) and statistically analyzed through one-way ANOVA and Tukey post-hoc tests. The results demonstrate that the ANN model consistently achieved the highest prediction accuracy, with MAPE values of 2.81% for S45C, 4.72% for ST41, and 4.42% for the combined-material dataset, outperforming the Regression and SVR models. These results confirm that the proposed intelligent multimodel system provides a robust, accurate, and practical solution for vibration-based surface roughness prediction in CNC turning of multiple materials.]]>
