Surface Form Error Prediction in High-Speed Micromilling of Thin-Walled TC4 Alloys: A Stacking-Based Ensemble Approach

Abstract Surface form error (SFE) prediction in high-speed micromilling of thin-walled structures is challenging due to dynamic factors such as tool deflection, workpiece flexibility, and machining vibrations. The existing mechanistic model lacks the adaptability needed for diverse machining environments, as they are material and cutting conditions dependent, which limits its application. This study presents a novel validated framework that combines adaptive signal processing with stacking-type ensemble machine learning models to address these challenges. Premachining and postmachining laser scans capture surface deformation, and any temporal discrepancies between them are resolved using the dynamic time warping (DTW) technique to generate the machined surface envelope. This machined surface envelope has been used to obtain the SFE. Furthermore, machined surface analysis has been performed to compare the generated SFE envelop with experimentally obtained machined surface SFE, revealing a maximum prediction difference of less than 12%. A stacking-based ensemble approach has been proposed and validated to predict SFE by training the proposed models on the estimated SFE from different machining conditions. Experimental validation has also been carried out by performing low-immersion radial micromilling on thin-walled TC4 specimens, and the proposed framework achieved an R2 of 0.973 with less than 10% prediction error. The ensemble model with the highest R2 value has further been applied to develop a graphical user interface (GUI) that enables real-time SFE prediction for machining optimization.