Detection and repair of cross-scale micro-nano surface defects for KDP optics based on integrated atomic force microscope and machine vision

Potassium Dihydrogen Phosphate (KH₂PO₄, KDP) crystals are widely used in laser systems owing to their outstanding nonlinear optical properties. However, cross-scale micro-nano surface defects inevitably introduced during ultra-precision machining severely induce laser-induced damage and propagation, especially under increasing laser energy. Thus, rapid defect detection and precise repair are critical to improve the laser damage resistance of KDP optics. This study developed a dual-station detection and repair system combining the EfficientNet-based machine vision for wide-field imaging and atomic force microscopy (AFM) for nanoscale-resolution imaging. The EfficientNet model was adopted to achieve efficient coarse detection and accurate localization, with defect coordinates transmitted to AFM for ultra-precise characterization. Furthermore, high-precision in-situ repair was realized using the condensed water bridge at the AFM tip-KDP interface. The system performance was optimized through mechanical design and finite element simulations. Full-aperture (50 × 50 mm) scanning was accomplished in 41 min, capturing defects from tens of micrometers to ∼ 500 nm. After repair at 65%, 75%, and 80% relative humidity, the scratch depths were reduced by 92.5%, 98.7%, and 95.8%, respectively. The overall surface roughness decreased from 18.50 nm to 4.21 nm, from 14.70 nm to 1.62 nm, and from 15.60 nm to 5.61 nm, respectively. A positive correlation between the initial scratch-wall slope and the initial filling rate was further observed, with a Pearson correlation coefficient of 0.871. These results demonstrate the feasibility of a dual-station machine-vision/AFM workflow for wide-field defect localization, nanoscale morphology detection, and localized water-bridge-assisted repair of KDP surface defects.