Prediction of Removal Function in Ion Beam Polishing of Potassium Dihydrogen Phosphate Crystals Using a Back-Propagation Neural Network

To overcome the challenges of processing soft-brittle potassium dihydrogen phosphate (KDP) crystals, this study proposes a back-propagation (BP) neural network model for the rapid prediction of the ion beam removal function using Faraday cup scanning data (a method that measures the spatial distribution of ion beam current density). By correlating current density measurements with point etching experiment results, the model accurately maps both the linear relationship (R2 = 0.98) between peak removal rate and peak current density, and the non-linear relationship between the full width at half maximum (FWHM) of the beam and the removal function. The predicted removal function demonstrates high accuracy, with a volume removal rate error of just 2.56% compared to experimental results. Furthermore, this method drastically reduces calculation time from approximately 2 h (required by the conventional point-etching experiment, which involves iterative vacuum cycling, etching, and ex situ interferometry) to just 2 min, significantly improving efficiency. Applied to the ion beam polishing of a 50 mm × 50 mm × 10 mm KDP crystal, the model proved highly effective. The surface figure error was corrected from an initial 0.298λ peak-to-valley (PV) and 0.0496λ root-mean-square (RMS) to 0.167λ PV and 0.036λ RMS, where λ (632.8 nm) is the wavelength of the He-Ne laser used for interferometric surface measurement, achieving a convergence ratio (defined as the ratio of initial PV to final PV) of 1.78. This research provides a high-efficiency, high-precision technical solution for manufacturing KDP components for inertial confinement fusion (ICF) applications.