Efficient Method for Missing Component Detection and Glass Bubble Inspection
| Autor: | Cheng-Shian Guo, 郭丞諴 |
|---|---|
| Rok vydání: | 2013 |
| Druh dokumentu: | 學位論文 ; thesis |
| Popis: | 101 With increasing demand to drive down costs and maintain high quality, automation is the key success in today’s production environment. Automatic optical inspection (AOI) offers a range of solutions to meet the requirements of every production facility and has put within the research of all manufacturers. Nevertheless, the inspection must allow the GO/NG decision to be fast and reliable while also assuring that the training of the inspector is simple and not time consuming. In this thesis, we work for two tasks based on AOI mechanism, especially aim at the missing components in the mobile phone assembly and the bubble defects in the spherical glasses, where the glass surfaces require diffused lighting to intensify the hardly visible defects. The goal of the first task is to cope with the difficulties of tiny components mixed with the phone back color to validate the existence of the associated components on the dedicated positions. In order to handle these difficulties, the adaptive lighting compensation method combined with projection color space is presented for enhancing the distinction between with and without components. Experiments in real cases demonstrate the efficiency of our approach with 100% correction rate. For the second task, we devise two image capturing modes, far and near, based on the novel spherical back-lighting mechanism to robustly detect the bubbles in the glasses. Spherical glass may produce bubble defects with micrometers in size in the production line. However, the transparent bubbles can only be seen in front of the lighting source by constantly changing viewing angles for human eyes. Our detection is based on the techniques of energy analysis and gradient-based methods for the captured images, where the optical capturing system can locate bubble defects as small as 70~300μm by highlighting bubbles in reverse mode. The experiments show the feasibility of the proposed framework with accurate rates of 95% (far) and 93% (near), respectively. It can further get higher recognition rates of 100% (far) and 95% (near), respectively, if we could exclude the scratches by human and the particles introduced by dust in the experiments. |
| Databáze: | Networked Digital Library of Theses & Dissertations |
| Externí odkaz: |
|