基于机器视觉的端塞表面缺陷检测算法研究及实现

发布时间：2018-01-07 05:24

本文关键词：基于机器视觉的端塞表面缺陷检测算法研究及实现　出处：《陕西科技大学》2017年硕士论文　论文类型：学位论文

【摘要】：随着人类对新能源的需求量不断增加,核能作为一种新型能源,以其较高的能源集度和丰富的能源储存量越来越多地受到了人们的关注。其中核电是核能使用的一种重要形式,核电具有环保、经济性高等优点,但是核电的使用在安全上有着较高的要求。核电中核反应堆的安全最为重要,而燃料棒是核反应堆中重要的组成部分,燃料棒端塞(以下简称端塞)作为燃料棒的一种核心零件,起着防止反应物溢出的作用,其表面的缺陷直接影响到核反应堆的安全。目前针对端塞表面缺陷的检测基本还停留在人工检测阶段,人工检测存在劳动强度大,检测效率低等问题。针对这些问题,利用机器视觉技术和图像处理算法完成对产品缺陷检测成为了一种更好的方法。本文针对端塞表面的两大类缺陷——侧表面缺陷和中心孔缺陷,在研究缺陷特征的基础上,分别设计了两种表面缺陷检测算法,并进行了算法的仿真研究;同时设计了端塞缺陷检测系统的硬件测试平台,并且开发了相应的检测软件。本文的主要工作包括:(1)端塞表面缺陷检测硬件测试平台的设计本文设计并实现了“旋转平台+LED前向光源+线阵相机”和“LED背光源+面阵相机”的硬件测试平台设计方案,分别对端塞侧表面图像和中心孔图像进行采集,并且利用计算机完成对采集图像的处理。(2)端塞表面缺陷检测算法的研究和设计本文通过对端塞侧表面和中心孔缺陷特征的分析,采用了“图像预处理+缺陷提取+缺陷分类”的检测流程。并在两个流程中,根据端塞侧表面缺陷图像和中心孔缺陷图像的特性,设计了相应的检测算法。(1)端塞侧表面缺陷检测算法端塞侧表面图像是在前向光源照射下由线阵相机拍摄的图像,本文根据端塞侧表面缺陷图像的特征,利用中值滤波和基于多项式的背景补偿方法对侧表面缺陷图像进行预处理;采用阈值分割提取缺陷,并通过形态学分析的方法去除“非缺陷点”,提高缺陷检测准确性;最后通过支持向量机分类完成对侧表面缺陷的分类。(2)端塞中心孔缺陷检测算法端塞中心孔图像是在背光源照射下由面阵相机拍摄的图像,本文通过对中心孔边缘信息的极坐标空间处理来检测中心孔缺陷。具体过程:首先利用基于灰度梯度的滤波方法对中心孔缺陷图像进行预处理;其次采用改进的双峰法进行阈值分割,并通过边缘检测提取圆边缘信息,再利用极坐标变换获得缺陷信息;最后通过二叉树分类器完成对中心孔缺陷的分类。(3)端塞表面缺陷检测软件的设计和实现在VS2010软件开发环境下,以C++语言为基础,利用MFC框架、多线程技术、Open CV视觉开发库等编程技术设计开发端塞表面缺陷检测软件。该软件通过对界面的设计、缺陷检测算法的移植、和数据库的调用,实现了端塞表面缺陷的检测、检测过程的监控和缺陷信息的存储等功能。本文通过对端塞表面缺陷检测硬件平台的设计,算法的研究以及相应的软件实现,得到了一种基于机器视觉的端塞表面缺陷检测算法及软件实现。实验结果表明,本文所设计的端塞表面缺陷检测算法和软件,对侧表面缺陷的检测正确率可达到86%以上,对中心孔缺陷的检测正确率可以达到90%,基本可以达到端塞表面缺陷检测的要求。
[Abstract]:With the demand for new energy increasing, nuclear power as a new energy for its high energy intensity and abundant energy storage capacity by more and more people's attention. The nuclear power is an important form of the use of nuclear power, nuclear power has the advantages of environmental protection, economic advantages, but nuclear power use a higher requirements on safety of nuclear power. Nuclear reactor safety is most important, which is an important part of the fuel rods in nuclear reactors, fuel rods (hereinafter referred to as the end plug) as a core part of fuel rods, can prevent the overflow of reactants, the surface defects directly affects the safety of nuclear reactors. For the detection of surface defects of the end plug remains in the manual detection stage, artificial detection in the presence of high labor intensity, low efficiency of detection problems. To solve these problems, the use of machine vision McGregor technology and image processing algorithm of product defect detection has become a better method. According to the two kinds of defects in the end plug surface - hole side surface defects and defect center, based on defect characteristics, respectively the two kinds of surface defects detection algorithm design, and simulation algorithm; at the same time the design of hardware testing platform end plug defect detection system, and developed the corresponding software. The main work of this paper includes: (1) design the design end plug surface defect detection hardware test platform and implement the "test platform design scheme of +LED rotating platform to source + line camera" and "LED back light + camera" hardware, were collected on the end plug side surface image and center hole image, and use the computer to complete the processing of image acquisition. (2) the end plug surface defect detection algorithm The research and design based on the analysis of the end plug side surface and center hole defect characteristics, the detection process of image pretreatment + defect extraction + defect classification. And in the two process, according to the characteristics of the end plug side surface defect image and center hole defect image, the corresponding detection algorithm design. (1) surface defect detection algorithm end plug side end plug side surface image is in the forward light source by linear array camera, according to the characteristics of the end plug side surface defect image, and background compensation method based on polynomial on the side surface defect image is preprocessed by using median filter; threshold segmentation the extraction method of defects, and through morphological analysis to remove "non defect", to improve the accuracy of defect detection; finally, the support vector machine classification complete classification of side surface defects. (2) the end plug hole defect detection center The end plug hole in the center of the image algorithm is image captured by the camera in the back light source, the detection center hole by polar coordinate space processing of center hole edge information. The specific process: first using the filtering method based on gray gradient of center hole defect image preprocessing; secondly the improved Shuangfeng method threshold segmentation, edge detection and edge information through extraction, then the defect information using polar coordinate transform; finally completes the classification of center hole defects by two binary tree classifier. (3) the design of end plug surface defect detection software and the realization of VS2010 in the software development environment, based on the C++ language, using MFC the framework, multi threading technology, design and development of Open CV visual programming technology development base end plug surface defect detection software. The software through the design of the interface, defect detection algorithm and the number of transplantation. According to the data base, realize the detection of the end plug surface defect detection process, monitoring and defect information storage and other functions. In this paper, through the design of the terminal hardware platform plug surface defect detection algorithm, and the corresponding software, get a machine vision end plug surface defect detection algorithm and software implementation based on the experimental results show that the end plug surface defect detection algorithm and the software designed in this paper, detection of the side surface defect accuracy can reach more than 86%, detection of the central hole in the correct rate can reach 90%, can reach the end surface defect inspection plug.

【学位授予单位】：陕西科技大学
【学位级别】：硕士
【学位授予年份】：2017
【分类号】：TP391.41;TM623

【参考文献】