Rapid Inspection System for Weak Feature Micro-Flaw on Large-Aperture Optics Surface Based on Multi-Illumination Modes and Multitask Learning Algorithms

Abstract

The occurrence of surface damage and contaminants can shorten the service life of large-aperture optics in high-power laser facilities, and efficient and accurate inspection of these flaws is an integral part of facility maintenance. Achieving rapid inspection of micrometer-level flaws ( $\gt 20~\mu $ m) on large-aperture optics surface ( $430\times 430$ mm) is a great challenge for both imaging systems and detection algorithms. To solve this problem, a rapid inspection system based on multi-illumination system and dark-field imaging principle is designed, which can complete full-aperture inspection of front and back surface of optics at one time through single-frame imaging and image processing. The application of multi-illumination modes increases the differentiation between different categories of weak feature flaws and provides sufficient information for flaw classification and measurement. A multitask deep learning (DL) framework that joints flaw classification and size regression is also designed to improve the classification capability and size measurement precision of the system for micro-flaws. The experimental results show that the system can perform a fast inspection of optics at a rate of less than 1.5 min per piece. The classification accuracy for four categories of flaws is 98.5% and the mean relative error (MRE) of size measurement is 14.8%. In the range of 20– $150~\mu $ m, exceeding the resolution limit of the imaging system, the detection system still maintains a high accuracy (98.9%) and low measurement error (15.6%). The proposed inspection system and the method for large-aperture optics achieve rapid detection of weak feature micro-flaws, which provides a highly promising solution for rapid, accurate, and low-cost inspection of precision optics.