Roughness Prediction in the End Milling of Additively Manufactured Maraging Steel with Machine Learning

Additive manufacturing (AM) strengthens mass-customized components manufacturing because of its high flexibility, even for metals like Laser Powder Bed Fusion (LPBF). Nevertheless, the need to improve the surface quality with other operations after LPBF and some predictability issues concerning the product's final properties parameters are still hindering this technology. In this way, machine learning applications are gaining even more attention in dealing with manufacturing management. So, this study investigated ways to predict the roughness of additively manufactured maraging steel upon milling using regression techniques. The linear regression traditionally used for prediction in the Design of Experiments was compared with the gradient boosting (GBR) and multilayer perceptron (MLPR) regressors. After tuning the model hyperparameters, the MLPR approach outperformed the other two techniques in modeling surface roughness. Also, it was possible to reach a coefficient of determination higher than 0.95 with acceptable error metrics during R
a
predictions. Hence, the obtained findings can be useful for developing data-driven control systems in the scope of Industry 4.0.