Machine Learning-Based Prediction of Machinability Responses in Meso-Scale Ultrasonic Vibration-Assisted End Milling (UVAEM) of Inconel 718 Superalloy: A Comparative Study of GPR, SVR, Random Forest, and Ridge Regression

Abstract

Meso-scale ultrasonic vibration-assisted end milling (UVAEM) is an advanced manufacturing technique. UVAEM of Inconel 718 superalloy presents significant modeling challenges due to complex thermo-mechanical interactions and meso-scale size effects. This study introduces a machine learning framework that systematically compares Gaussian Process Regression (GPR), Support Vector Regression (SVR), Random Forest (RF), and Ridge Regression for predicting cutting force, tool wear, and surface roughness, thereby contributing to sustainable manufacturing. Experiments followed a Taguchi L16 orthogonal array with two replicates (n = 32), varying cutting speed (10–40 m/min), feed rate (0.01–0.025 mm/tooth), depth of cut (0.10–0.25 mm), vibration amplitude (0–9 μm), and tool coating (TiAlN, TiSiN, nACo, Uncoated). Tool coating was one-hot encoded, and strict leave-one-out cross-validation (LOOCV) with within-fold standardization ensured unbiased generalization metrics. Following nested hyperparameter tuning, SVR achieved the highest accuracy (R2 = 0.9552, 0.9473, 0.9294), marginally outperforming GPR (R2 = 0.9543, 0.9420, 0.9290). Ridge regression was competitive for tool wear (R2 = 0.9235), while random forest ranked last due to limited ensemble diversity at n = 32. Pearson correlation identified depth of cut as the dominant driver of cutting force and surface roughness (r = 0.767, 0.759), cutting speed as the primary driver of tool wear (r = 0.663), and vibration amplitude as consistently beneficial across all three responses. SVR and GPR are recommended as reliable surrogate models for process optimization in UVAEM of Inconel 718.