【作者】 黄杰贤；

【导师】 李迪；

【作者基本信息】 华南理工大学 ， 机械电子工程， 2012， 博士

【摘要】 挠性印制电路（FPC）是以聚亚胺或聚脂薄膜为基材制成的一种具有可弯曲的、可卷绕的印刷电路，是航空、航天军事汽车消费电子产品不可或缺的重要载体或关键连接手段。随着电子产品朝着轻、薄、小、巧的趋势发展，FPC的精密度以及制程的复杂度日益增加，对FPC质量控制亦提出更加严苛的要求。针对目前FPC制造过程中外观缺陷检测严重依赖人工的落后现状，本文研究了FPC的外观缺陷自动光学检测关键技术及实施方法，主要工作如下：1）提出了基于互信息熵的定位误差校正方法。针对挠性印制电路基材制成品与标准设计存在的差异导致的定位误差，本文采用图像灰度互信息熵作为衡量检测目标与设计标准之间差异的指标，提出灰度互信息熵与关键区域特征相结合的误差校正方法，实现了检测目标的精准匹配。实验结果表明，对于75微米制程该方法定位精度可达16.6微米，优于霍夫变换法与坐标转换法两种同类算法；当对存在噪声干扰、信息缺失、外形偏差的检测目标进行匹配时表现出良好的抗扰性。2）提出了量化目标轮廓边缘方向性的表达方法——纹理梯度方向性分布熵，基于该方法能够获取轮廓坐标与方向参数。对存在局部形变的焊盘进行匹配时，根据轮廓坐标与方向参数筛选出正确的轮廓并实现对焊盘的匹配。3）基于灰度分布及纹理梯度分布建立纹理粗糙度的熵统计量，该统计量能够充分描述焊盘纹理粗糙度波动的二维分布情况。针对焊盘金面的化金、脏污、机械损伤等主要缺陷进行提取、量化的实验表明，该统计量能够灵敏地反映焊盘金面正常或异常情况下的不同表面视觉特征。可准确检测出焊盘金面缺陷。4）提出基于Radon变换的纹理方向特征提取方法，解决了断路、开路、凸起、缺口等线路缺陷的检测问题。 Radon算子是一种直线积分的投影变换，主要应用于二维平面上的直线特征提取。针对线路缺陷，采用该变换建立纹理方向特征相似度函数，对具有相同方向的纹理图像进行分割，然后以Radon变换值的波动程度来衡量纹理的方向强度特征。结果表明：当线路存在缺陷时，线路图像的纹理方向强度减小，同时Radon变换值的波动显著降低，其Radon变换的波动值可有效表达线路纹理方向强度特征，可作为线路缺陷识别的重要判据。本文在提出上述检测算法的基础上，构建了FPC外观缺陷检测平台，并成功应用于实际FPC生产质量控制的过程中。取得良好的效果。相关研究工作具有较好的理论意义以及推广应用价值。更多还原

【Abstract】 Flexible printed circuit (FPC) is a kind of printed circuit board that is made of flexiblematerial such as polyurethane and polyester. It can be bent, folded, rolled stretched and“moved freely” within a three-dimension space. Due to the above characteristics, FPC canconnect PCB and parts, which makes the product to fit the requirement of small size.Therefore, FPC is widely used for aviation, aerospace, military, automobile, consumingelectronics and so on. So the quality control plays a crucial role in the FPC producing process.The paper presents inspecting work based on machine vision. It contains four aspect, there isrespectively inspecting target location, searching inspecting area, extraction of feature, patternrecognition. On the basis of previous research, considering realistic application, the inspectingwork is discussed in sequence and the corresponding solutions have been given, and theresearch content and innovation work are as in the following:1) The method of correcting position error based on mutual information entropy isproposed in this paper. The realistic product is always different from standard template sincethe FPC manufacturing material is easily affected by thermal expansion and contraction.Correcting matching error is crucial work during manufacturing process. The mutualinformation entropy (MIE) is applied on evaluating the correlation between the images. TheMIE increases and reach maximum when the referent and inspecting FPC solders are alignedwith the same place. Moreover, The MIE in some image’s region which contains more greyinformation can react to position error strongly. According to the above principle, the MIEand key region feature is applied to searching best optimal correction parameters to improvepositioning accuracy. The method is verified by a simulation and applied to the inspectingsystem. It demonstrates that the method can correct the position error more effectivelycompare with Hough transformation method and affine transformation method. The matchingaccuracy can reach16.6micron.2) The texture gradient directionality distribution entropy is proposed to describe thedistribution of inspecting objective contour’s directionality.When solder deformation accurs,the method is applicated in extracting solder’s contour feature and direcionaity feature. Finally,the contour similarity function is established. The valuable contour information is reversedand adopted to match inspecting area.3) Entropy statistic based on distribution of grey and texture gradient is proposed todescribe texture roughness feature.And it can effectively measure texture roughness.Then themethod is applied to the identification of defect on solder.The algorithm reaches satisfactory results and overcome the shortcoming of the traditional method.4) Directionality is an important visual feature of texture too. The thesis presents a robustimage texture directional measurement method. Circuit’s defect can be detected by thetextural direction feature. Radon transformation is mathematical expression based onprojection transformation. During inspecting process it is implemented on circuit image toestablish texture direction characteristic similarity function. Then, some region in imagewhich contains the same texture direction feature is segmented. Furthermore, thedirectionality estimation method is presented. As the circuit is damaged, the directionality isweakened correspondingly.And the radon transformation value’s fluctuation can response todirectionality.The detection of circuit’s defect is achieved through analysis of the fluctuationon radon transformation value. The inspecting experiment demonstrates the method canextract defective region correctly and get satisfactory result.5) The platform is constructed and evaluated on-line testing. The inspecting resultdemonstrates accurate defect detection with low false alarms, and the efficiency can satisfythe requirement of online and real time inspection. The research work has good theoreticalsignificance and application value.更多还原