【作者】 张瑞秋；

【导师】 张宪民；

【作者基本信息】 华南理工大学 ， 机械制造及其自动化， 2014， 博士

【摘要】 日新月异的电子消费产品(笔记本电脑、智能手机、PDA、MP4、GPS等)、工业控制电子产品、国防尖端电子信息装置等都需要各种规模电子集成度的电路板组装产品，它是各种智能设备与终端的核心。随着半导体技术摩尔规律的作用，集成电路的集成度越来越高，表面贴装技术（Surface Mounted Technology，SMT）这种新型电子装联技术，彻底改变了传统的通孔插装技术，使电子产品向微型化、轻量化方向发展。而面向SMT的微焦X射线精密检测设备是针对电路板组装结构性缺陷检测的关键设备。为了提高X射线检测系统自动化程度，本文深入研究了智能化的检测算法和重建算法。主要研究内容如下：（1）基于X射线的检测过程中，由于康普顿效应和X射线能量不稳定会发生散射等现象，降低图像质量。本文在对图像增强方法进行总结和分析的基础上，提出了基于鲁棒估计的Retinex图像增强方法。该方法将一些不符合假设的图像数据称为“出格点”，根据鲁棒估计可以求出不受出格点影响的估计结果，把边界像素作为出格点进行分区域处理。论文中的增强算法是利用Malik和Perona提出的各向异性扩散的两种边界终止函数的凸组合形式作为新的边界终止函数，从而增加了图像的对比度，克服边界附近被严重模糊从而产生伪影的问题。（2）为了能够准确、快速地从X射线图像中把BGA焊点分割出来，提出了基于模式识别理论中的Fisher准则函数的BGA图像分割算法，利用Fisher评价准则对目标与背景分离程度进行定量判断，以不同的灰度值将图像划分为目标和背景。当Fisher准则函数取最大值时，目标和背景达到最佳分离度，所取的灰度值为最佳阈值。实验证明这种分割结果，不仅能够得到图像边缘细节，还可以克服边缘存在断点的缺陷，适合BGA焊点图像的分割。（3）对BGA焊点的特征进行分析与提取，包括短路、偏移、少锡、空洞、翘曲和开路等现象。提出水平断层算法、垂直断层算法，根据图像特征进行连续性、偏移量、面积率和空洞率等的判定，然后利用决策规则判定是否存在某种缺陷，最终将良品与缺陷品分开。本文还提出了基于Fisher准则的BGA焊点检测算法，也通过实验进行了验证。（4）提出了基于FDK框架下的超短扫描的改进算法，以解决由于机械物理的限制，导致在某些投影角度下无法采集投影数据，由此产生的截断数据有限角度重构问题。通过对BGA芯片偏置放置使得其成为重构的ROI区域，进而扩展Parker权函数以处理三维投影下的冗余数据。引入该算法后,可缩短扫描及图像重构时间，在保证重构图像质量的同时提高检测效率，使CT更加适合在工业中应用。（5）探讨了基于迭代法的少角度投影下，适用于BGA电路板检测的同步型重构算法和TV算法，提出以投影值大小来决定反投影速度的惩罚型算法－WP型算法，同时也提出了一种适合于平面型物体的S-TV型算法，它抛弃在平面型物体中较弱的厚度方向，可准确地在少角度投影下重构出平板型物体。最后，在总结全文内容的基础上，本文还对基于X射线检测的研究重点和研究方向进行了展望。更多还原

【Abstract】 The consumer electronics products (such as laptop, Smartphone, PDA, MP4, GPS and soon), industrial control electronics and defense high-end electronic information devices whichchanged with each passing day, all need PCB assembly products of all sizes electronicintegration. The PCB assembly products are the core of the various intelligent device andterminal equipments. As the Moore law of semiconductor technology took effect, integratedlevel of the integrated circuit has become increasingly higher. Surface Mounted Technology, anew technology of electronic assemblies has completely changed the traditional holeinstrumentation technology and made the electronics products become more miniaturizationand lightweight. However, Micro confocal X-ray precision testing equipment for SMT is thecritical equipment to detect the defects of PCB assembly structure. In order to improve thelevel of automation for the X-ray inspection systems, This paper deeply studied the intelligentdetection algorithm and reconstruction algorithm. The main contents are as follows:(1)Based on the detection of X rays, due to the Compton effect and X ray energyinstability occurs scattering phenomena etc. the image quality can degradate. On foundationof the summary and analysis of current image enhancement method, we propose the Retineximage enhancement method based on robust estimation. The method calls some date which isnot consistent with the hypothetical image data set "very little", according to the robustestimation that can be obtained to estimate the result that is not affected by the special effects,and process a special point by subarea domain with a robust estimation of the boundary pixelsas. Enhancement algorithms proposed in this paper utilize that Malik and Perona proposedanisotropic diffusion termination function with two kinds of boundary convex combination asa new form of boundary termination function, so as to increase the image contrast, overcomethe seriously blurred boundary and the artifact problem.(2) To separate the BGA solder joint image accurately and quickly from the X-rayimages, the project is about the BGA image segmentation algorithm based on patternrecognition theory, to have a quantitative judgment to the degree of separation between thetarget and background by the evaluation criteria of Fisher,and to classify image into twosorts--target and background in different gray value. When Fisher criterion function takes themaximum value, it comes out the best resolution of the target and background, and the grayvalue achieves the best threshold. Experiments show that this division is not only able to getthe image edge detail, but to overcome a defect of the edge breakpoints, a good method for the segmentation of BGA solder joints X-ray image.(3) According to the analysis and extract of the characteristics of solder joints, includingshort circuit,offset,less tin,cavity,warping,open circuit and so on, horizontal fault algorithmand vertical fault algorithm were proposed.Based on graphics characteristics, we candetermine continuity, offset, the area ratio, voidage and so on. Finally, we can judgewhether there are some defects with the decision rules. Furthermore, we distinguish the goodand the defect according to the Fisher Criterion Function,and put forward the BGA SolderJoints Detection Algorithm which is on the basis of Fisher Criterion,and verified it by meansof experiment.(4) This article propose one improved FDK-type algorithm of the ultra-short scan toresolve the limited-views problems which happen frenquently because of some mechanicaland physical limitations. It try to process the BGA reconstruction as a ROI reconstruction problem with an offset approach, and then modify the Parker’s weight function to let it adopt to3D situation. Through introduction of the algorithm, the time of scan and image reconstructionis reduced, the image quality of3D reconstruction can be guarantied while improving the detection efficiency, which will make CT more suitable in industrial applications.(5) This article has explored the synchronization type reconstruction algorithm andTV algorithm for BGA circuit board examination, which based on less angle projections. It proposed to use the information of projection value to control the back-projection speed whichcalled punishment-based algorithm: WP algorithm. At the same time, presented one S-TV algorithm which is quite suitable for flat-type objects as it give up a weak thickness direction of the object but can accurately reconstruct object with less angle projections.Finally, the main results of the dissertation are concluded, this article has also discussedthe future of the research focus and research directions based on X-ray test.更多还原