Surrogate model-based tool trajectory modification for ultra-precision tool servo diamond turning

Abstract

Tool servo diamond turning is extensively used for machining complex-shaped freeform surfaces due to its deterministic material removal capabilities. However, the inherent bandwidth limitations of the current slow slide servo technique lead to significant tracking errors, posing critical challenges to achieving high-efficiency and high-precision fabrication of these intricate surfaces. To address these challenges, this work proposes a novel data-driven surrogate model for tool trajectory modification that predicts servo axis tracking errors and adjusts the reference tool path prior to cutting operations, thereby enabling effective feedforward compensation. A two-dimensional convolutional neural network (2D-CNN) surrogate model is employed to capture the dynamic properties of tracking errors inherent in servo axes, with particular emphasis on the servo axis along the depth-of-cut direction. The predicted tracking errors serve as feedforward compensation terms for the initial reference trajectory, generating the modified diamond tool trajectory. Experimental validation on a commercial three-axis ultra-precision machine tool demonstrates the effectiveness and practical applicability of this trajectory modification method. Comparative results indicate that, with the assistance of the proposed modification method, the peak-to-valley (PV) error for segments of the tracked tool trajectory decreases from 1.29 μm to 0.59 μm, and the root-mean-square (RMS) error decreases from 373 nm to 138 nm; the PV error for the cross-sectional profiles of machined freeform surfaces decreases from 1.25 μm to 0.65 μm, and the RMS error decreases from 196 nm to 117 nm.