【作者】 蒋利容；

【导师】 王黎；

【作者基本信息】 西南交通大学 ， 光学工程， 2013， 硕士

【摘要】 在机车故障诊断的各项检修中,主要的方式有：人工肉眼判别、基于机器视觉的图像自动识别。机车底部部件结构复杂度高,细小部件多,仅靠人工检查,很难准确记住所有部件的正常状态和形式,易由视觉疲劳造成漏检,尤其在高速运行状态下的列车任何一个细小、细微的故障,都可能引发重大事故。机器视觉图像技术作为智能自动识别已经广泛应用于各个行业,对提高铁路车辆安全检测技术水平,具有重要意义。本论文旨在应用图像分割、特征描绘等图像处理技术实现列车底部心盘故障图像的自动识别,减少人为影响误差,提高检测效率。论文在研究目前铁路故障轨边图像检测系统和图像处理技术的基础上,分析了图像特征提取的各种方法和性质。然后结合机车底部心盘图像特有的结构特性,首先定义五个分割特征,利用线性加窗灰度投影进行粗分割,定位出待检测螺栓的目标区域,然后根据特征定义间关系分割得到各个螺栓图。对分割得到的螺栓图经过滤波、灰度直方均衡增强、面积阈值法二值图像、提取面积大小与故障特征疑似的区域判别、建立封闭矩形描绘子等图像处理技术,最终实现机车底部心盘螺栓丢失故障图像识别算法模型的设计。最后,设计了机车底部图像采集的方案,实现方案搭建和图像采集实验,并用实际图像对建立的算法模型进行了验证。该算法能够准确自动识别螺栓丢失故障,满足代替人作业的要求,提高列车检修效率。本课题对类似故障识别的案例或者自动检测系统的研发也具指导作用。更多还原

【Abstract】 In recent years, with the rapid development of world’s high-speed railway, especially for the running of high-speed motor train units, rail safety inspection accept new challenge, more rigorous and sophisticated maintenance and monitoring of trains are put forward to implement. Because of complexity structures and so many small-parts, it is too hard and tired for works to accurately handle all problems. Any subtle fault could cause a serious accident in high speed. As intelligent automatic recognition technology, machine vision image processing technology which been widely used has great significance to realize railway vehicle safety inspection.Aiming at designing locomotive-bottom-image acquisition system and realizing faults automatic recognition to reduce man-made influence and improve detection efficiency, it is guided for the research of similar cases or automatic detection system.In the paper, on the premise of investigating and summarizing railway image detection technologies and devices and image processing technologies, locomotive-bottom image acquisition system is designed and set up. Combining with analysis of properties of image feature extraction methods and structure characteristics of train, five structure-segmentation rules are defined. Take the integration of rectangular function and gray projection to segment image, and then get the object-bolts image date based on the relationship between each structure-segmentation rule. A set of defects recognition algorithms are proposed based on filtering, histogram equalization, area thresh, connected region extraction, closed rectangular areas descriptors. Faults of bolts missing can be positioned accurately through using the proposed method in the testing image. Over all, it will give some guidelines for the work afterward.更多还原