焊缝缺陷检测算法研究

发布时间：2018-07-13 13:23

【摘要】：随着图像处理技术和计算机技术的不断发展,对X射线焊缝图像的数字化起到了极大的推动作用。而目前X射线探伤检测仍以人工评定方式为主,且在评定过程中易受个人主观因素影响,工作量大,容易引起误判、漏判,因此实现焊缝缺陷自动检测十分必要。本文以DR系统采集的焊缝图像为研究对象,主要围绕焊缝缺陷检测算法进行研究。针对焊缝图像中纹理复杂、对比度差、背景起伏大等问题,首先对图像进行了相应的预处理算法研究;然后针对缺陷特点的不同,进行了相应的缺陷检测方法研究;最后利用不同算法间的互补性,将结果融合避免误检和漏检。首先,针对焊缝图像中信噪比低、对比度差等问题,利用图像增强和降噪改善图像质量、减少图像中噪声的干扰;针对动态视频检测,采用灰度归一化和每M帧连续检测N帧的方案分别解决了不同规格间灰度分布不一致和采集检测实时显示的问题;同时为了减少干扰和提高检测效率,提出了一类针对DR成像的焊缝边界自动提取方法,具有较好的适应性和实用性。其次,针对焊缝缺陷的提取问题,根据不同焊缝缺陷的特点,本文设计了相应的基于Canny、Lapalace、帧差法、ButterWorth滤波的缺陷检测算法,但同一方法只针对特定的缺陷类型效果明显,为了防止缺陷漏报,利用不同检测方法间的互补性,将检测结果融合,并针对动态视频和静态图片,设计了相应的检测方案,具有较好的通用性。本文通过对焊缝图像缺陷区域与非缺陷区域的特性分析,利用空间特性构建模式矢量,并利用SVM对样本进行训练,之后进行缺陷检测和相应的结果分析。最后,本文设计并实现了基于DR成像及缺陷检测系统。利用计算机多线程技术实现了图像的采集、抓拍、缺陷检测等功能。
[Abstract]:With the development of image processing technology and computer technology, the digitization of X-ray weld image has been greatly promoted. But at present, the main method of X-ray flaw detection is manual assessment, and it is easy to be affected by individual subjective factors in the process of evaluation, so it is easy to cause misjudgment and miss judgment, so it is very necessary to realize automatic detection of weld defects. In this paper, the weld image collected by Dr system is taken as the research object, mainly focusing on the weld defect detection algorithm. Aiming at the problems of complex texture, poor contrast and large background fluctuation in the weld image, the corresponding pre-processing algorithm is studied firstly, and then the corresponding defect detection method is studied according to the different characteristics of the defect. Finally, by using the complementarities of different algorithms, the results are fused to avoid false detection and miss detection. Firstly, aiming at the problems of low signal-to-noise ratio and poor contrast in weld image, image enhancement and noise reduction are used to improve image quality and reduce noise interference. In order to reduce interference and improve detection efficiency, gray level normalization and continuous detection of N frames per M frame are adopted to solve the problems of inconsistent gray distribution and real-time display of acquisition and detection among different specifications, respectively. A method for automatic extraction of weld boundary for Dr imaging is presented, which has good adaptability and practicability. Secondly, aiming at the problem of weld defect extraction, according to the characteristics of different weld defects, this paper designs the corresponding defect detection algorithm based on Canny Lapalace, frame difference filter and ButterWorth filter, but the same method is effective only for specific defect types. In order to prevent defects from underreporting and to make use of the complementarity between different detection methods, the detection results are fused, and the corresponding detection scheme is designed for dynamic video and static pictures, which has good generality. By analyzing the characteristics of defect region and non-defect region of weld image, the pattern vector is constructed by using spatial characteristics, and the samples are trained by SVM, then defect detection and corresponding result analysis are carried out. Finally, this paper designs and implements the imaging and defect detection system based on Dr. The functions of image acquisition, capture, defect detection and so on are realized by computer multi-thread technology.
【学位授予单位】：西安理工大学
【学位级别】：硕士
【学位授予年份】：2017
【分类号】：TG441.7;TP391.41

【参考文献】