An integrated framework for a multi-material surface roughness prediction model in CNC turning using theoretical and machine learning methods

Indirect monitoring and prediction of surface roughness for computer numerical control (CNC) machining enables manufacturers to ensure quality outcomes are achieved, increase process productivity, and decrease the risk of scrapped components. Several approaches have been examined for surface roughness prediction in CNC turning; however, few studies have explored the development of models for multiple materials. These studies typically use a purely empirical approach, limiting the range of prediction with a high reliance on process data. This study presents a framework for predicting surface roughness in three different materials. The framework combines a kinematics-based prediction model with an ensemble boosted regression tree machine learning algorithm. This combination allows for the accurate prediction of the machined surface roughness using the machining parameters and sensor data as input variables. In order to develop this approach, experimental data is collected for dry turning of CGI 450, AISI 4340, and AISI 316. The influence of machining parameters on measured surface roughness is analyzed using analysis of variance (ANOVA) and Pearson’s correlation coefficient (PCC). The prediction results show that the multi-material model achieved a root-mean-square error (RMSE) of 0.166 μm, and 70% of predictions were within the surface roughness limits defined using the ASME B46.1–2019 standard. This provides a similar result to the individual models for CGI 450 and AISI 4340 and outperforms the individual model for AISI 316. These results demonstrate the potential for a multi-material model to be applied in an optimization strategy as a decision-making tool for operators in the manufacturing industry.