Modeling of ball-end micromilled surface roughness and geometry in ultrafine-grained and dual-phase steels using interpretable machine learning

This study presents an integrated experimental and computational approach to analyze surface integrity in ball-end micromilling of two low-carbon steels with distinct microstructures: dual-phase (DPh) and ultrafine-grained (UFG). The influence of tool diameter, neck length, feed per tooth, and milling strategy (up- and down-milling) on surface roughness (Ra, Rz, skewness, kurtosis), burr formation, and profile accuracy was systematically investigated. Tool deflection effects, more critical in UFG due to its higher ductility, were quantified through geometrical deviation metrics. Predictive models using Random Forest (RF) and Multilayer Perceptron neural networks (MLP) were developed to estimate surface roughness based on machining parameters. The MLP model showed superior performance for UFG steel (R² = 0.71), indicating enhanced prediction capability for homogeneous microstructures. Feature importance analysis highlighted the dominant effect of tool diameter and feed per tooth. The results advance the understanding of process-material interaction in micromilling and demonstrate the potential of interpretable machine learning for surface quality prediction in ultrafine-grained steels.