基于磁痕图像的无损探伤关键技术研究

发布时间：2018-02-07 16:16

本文关键词： 无损检测图像处理磁粉探伤表面裂纹检测　出处：《西南科技大学》2017年硕士论文　论文类型：学位论文

【摘要】：磁粉探伤作为一种效果明显、技术成熟和成本低廉的无损检测方法,广泛应用于工业中铁铸件的质量检测环节,基于机器视觉的表面检测技术已经成为无损检测技术中的最常用的技术之一。本文将图像处理和模式识别技术与传统的磁粉探伤技术相结合,利用视觉检测技术实现轴承零件表面裂纹的自动检测。本文首先介绍了磁粉探伤技术的最新研究进展,阐述了磁痕成像的基本原理,分析了影响磁痕成像质量的因素。其次提出了一套基于机器视觉的磁粉探伤实施方案,其中硬件平台用于磁痕图像采集,软件算法主要包括磁痕特征提取与描述和裂纹识别。最后就目前磁痕识别方法中存在的磁痕提取与描述问题,本文提出了效果较好的解决方法。在磁痕提取上,首先介绍了磁痕图像的特点与预处理算法;其次针对磁痕存在提取困难的问题,本文利用二维Otsu、Laws和K-means等三种算法实现了磁痕的提取,与基于颜色和阈值的磁痕分割方法相比,本文三种算法的磁痕分割效果更好。在磁痕特征描述上,本文提出了不变矩、颜色矩和颜色聚合向量三种磁痕特征描述方法。通过实验对比与分析,与基于区域形状的几何特征描述方法相比,本文算法不需要预先进行磁痕分割,具有更强的通用性。本文最后针对现场采集的大量轴承零件磁痕图像,通过提取不变矩、颜色矩和颜色聚合向量等三种共计81维特征向量,利用SVM实现了轴承零件表面的裂纹识别。实验结果表明,本文提出的检测方法能有效地区分表面污染、裂纹和良品三大类工件,整体识别率高达85.62%,具有一定的工程应用价值。
[Abstract]:As an effective, mature and low cost nondestructive testing method, magnetic particle inspection is widely used in the quality testing of iron castings in industry. Surface detection based on machine vision has become one of the most commonly used techniques in nondestructive testing. This paper combines image processing and pattern recognition with traditional magnetic particle flaw detection technology. The automatic detection of surface cracks in bearing parts is realized by using visual inspection technology. In this paper, the latest research progress of magnetic particle flaw detection technology is introduced, and the basic principle of magnetic trace imaging is described. The factors that affect the quality of magnetic mark imaging are analyzed. Secondly, a set of magnetic particle flaw detection scheme based on machine vision is proposed, in which the hardware platform is used for the magnetic mark image acquisition. The software algorithm mainly includes magnetic mark feature extraction and description and crack recognition. Finally, a better method is proposed to solve the problem of magnetic mark extraction and description existing in the current magnetic mark recognition method. Firstly, the characteristics and preprocessing algorithm of magnetic mark image are introduced. Secondly, aiming at the problem of magnetic mark extraction, this paper uses two dimensional OtsuLaws and K-means algorithms to realize the magnetic mark extraction, which is compared with the magnetic mark segmentation method based on color and threshold. In this paper, we propose three methods to describe the feature of magnetic mark: invariant moment, color moment and color aggregation vector. Compared with the geometric feature description method based on the shape of the region, the algorithm in this paper does not need to carry out the magnetic mark segmentation in advance, so it is more universal. Finally, aiming at a large number of magnetic trace images of bearing parts collected in the field, the invariant moments are extracted. Three kinds of 81 dimensional Eigenvectors, such as color moment and color aggregation vector, are used to identify the cracks on the surface of bearing parts by using SVM. The experimental results show that the proposed detection method can effectively distinguish the surface contamination. The overall identification rate of crack and good products is up to 85.62, which has certain engineering application value.
【学位授予单位】：西南科技大学
【学位级别】：硕士
【学位授予年份】：2017
【分类号】：TG115.284;TP391.41

【相似文献】