Rapid Inspection System for Weak Feature Micro-flaw on Large-aperture Optics Surface Based on Multi-Illumination Modes and Multi-Task Learning Algorithms

Zhaoyang Yin,Linjie Zhao,Jian Cheng,Henan Liu,Mingjun Chen
DOI: https://doi.org/10.1109/tim.2024.3472831
2024-01-01
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 the facility maintenance. Achieving rapid inspection of micron-level flaws (>20μm) on large-aperture optics surface (430mm×430mm) 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 multi-task deep learning 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. Experimental results show that the system can perform fast inspection of optics at a rate of less than 1.5 minutes per piece. The classification accuracy for four categories of flaws is 98.5% and the mean relative error of size measurement is 14.8%. In the range of 20μm-150μm, 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 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.
What problem does this paper attempt to address?