Prediction of surface roughness in duplex stainless steel top milling using machine learning techniques

This study presents an innovative approach to predicting the quality of finished parts in top milling processes by integrating robust parameter design with artificial intelligence techniques. A central composite design was used to combine controllable variables (cutting speed, tooth advance, milled width, and cutting depth) with noise variables (tool flank wear, fluid flow, and cantilevered length). Duplex stainless steel was milled under each experimental setup, and roughness data were collected. These data were used to train three machine learning models: random forest, decision tree, and support vector machine. The models predicted surface roughness, and their predictions were validated through experimental tests. The root means square error values were 0.031 for the random forest, 0.038 for the decision tree, and 0.066 for the support vector machine, indicating that the random forest model performed the best. This innovative study highlights the importance of including noise variables along with controllable factors in machine learning models, significantly improving prediction accuracy and making them more reflective of real-world results. Including these variables is crucial, as neglecting them can lead to inaccurate predictions.