Summary: | 碩士 === 朝陽科技大學 === 工業工程與管理系碩士班 === 90 === Ripple-texture is a common defect because of the steam left on Surface barrier layer (SBL) chip surface. Ripple-texture defects influence not only the appearance of SBL, but also the electronical properties of the products. The reasons why the inspection of ripple-texture defect cannot be done automatically are: (1) ripple-texture defect is semiopaque; (2) ripple-texture is un-structural texture; (3) the edge of ripple-texture defect can be changed gradually. It is not only hard to find the ripple-texture by artificial inspection, but also get the error by human subjectivity and fatigue.
This research proposes new computer vision methods of the texture inspection procedures. WCMP (Wavelet Characteristic Multivariate Processing) model, GCMP (Gray Characteristic Multivariate Processing) model, and WCMT (Wavelet Characteristic Multivariate Transform) model are proposed in this study. After multivariate processing, the defects will have high multivariate power. Last, new threshold techniques include of Weight-Iterative, Mode-Double, and Mode-Triple are used to locate the defects for different demands.
After experiments of this research, WCMP and WCMT models with Mode-Double or Mode-Triple threshold techniques have good results about 90% of image accuracy. It is sure to solve the ripple-texture problem well. For other regularity texture image, WCMT model has over 95% of image inspection rate. And, this research results can be extended to be used in related image inspections about medical image or oxidization defects on leadframe.
|