AI-Based Prediction of Depth-Volume (DV) Tool Influence Function (TIF) for Space Optical Materials in Freeform Optical Fabrication for a Wide-Field Camera

Precision polishing processes for freeform mirrors are required to meet the performance requirements of freeform three-mirror optical systems. Effective quantitative control of this process depends on the accurate prediction of the tool influence function (TIF), which defines the local material removal distribution under given process conditions and dwell times. We applied Gaussian process regression (GPR) to predict the TIF of space optical materials—cordierite, silicon carbide (SiC), and Aluminum-Silicon (Al–Si)—and compared the prediction accuracy between two target quantities: TIF peak depth and removal volume. To achieve this, 81 TIF samples were acquired per material by independently combining three levels of force, wheel speed, and dwell time (3 × 3 × 3 = 27 conditions, three replicates per condition). For each material, 72 samples were used for training, and one of the three replicates from each of nine selected process conditions was randomly held out for testing. As a result, the volume-based model achieved a lower mean absolute error (normalized MAE = 0.02) than the depth-based model (normalized MAE = 0.03), corresponding to a 33% reduction. The results suggest that TIF volume is a more robust target quantity for AI-based TIF prediction in the current dataset and evaluation setting.