Efficient and precise detection for surface flaws on large-aperture optics based on machine vision and machine learning

Abstract

Large-aperture optical components have a wide application in high-power laser facilities. Surface flaws such as damages and contaminants generated under high-power laser irradiation can greatly affect the optical or mechanical performance of optics, so the optical components must be carefully inspected to evaluate optics damage and cleanliness. However, it is of great challenge to achieve rapid and accurate positioning, classification, and measurement of micron-level flaws on the surface of large-aperture optics. This paper proposes a novel surface flaw detection technology based on machine vision and machine learning. To balance efficiency and accuracy, a dark-field imaging system based on progressive scanning is designed to obtain the image of optics surface. A set of surface flaw detection algorithms based on machine learning including object segmentation, flaw feature extraction, and classification and size calibration of flaw are proposed to accurately assess the state of surface flaws. The experiments show that the system can complete the detection of an optical component (430 mm × 430 mm) within 6 min. The minimum detectable size of the flaws can reach 20 μm, and the position accuracy of the flaws is better than 50 μm. The classification accuracy of flaw is 98.59 %, and the average relative errors of the size measurement of damages and contaminants are 3.23 % and 6.58 %, respectively. The experimental results demonstrate the effectiveness of the method in detecting surface flaws of large-aperture optics.