节点文献
货车走行部弹簧缺损图像检测技术研究
Research on Defects Image Detection Techniques of Freight Car’s Running Gear Springs
【作者】 张晓林;
【导师】 高晓蓉;
【作者基本信息】 西南交通大学 , 光学工程, 2012, 硕士
【摘要】 铁路货车走行部弹簧作为转向架的关键部件,在使用过程中会因原材料缺陷及异常的冲击等因素而产生缺损,失去缓冲和减振的作用,进而影响到行车安全。一直以来,弹簧缺损故障的检测主要是由人工完成,不仅效率低、可靠性差,而且工人劳动强度大,存在很大的安全隐患。国内的货车运行故障动态图像检测系统能通过人机结合的方式,检测出弹簧缺损,在一定程度上减小了工人的劳动强度,提高了故障发现率。但实质上,该系统本身仅是完成了图像的采集、传输及适当的预处理,弹簧故障的判断仍是通过人工肉眼浏览采集到的图片来完成的,依然存在很多问题,需要进一步改进和升级。基于此,本文研究了如何利用图像处理技术来实现弹簧缺损故障的自动识别,对于提高故障检出率,保障货车安全运行具有十分重要的意义。本文首先在调研国内外走行部弹簧缺损图像检测技术的研究现状的前提下,研究了基于特征扫描的弹簧图像定位算法。思路是:先对采集到的走行部图像进行光照不均校正、二值化处理,然后扫描分析二值图像,根据图像特征找到弹簧区域边界位置,实现定位。通过对120幅图像进行实验及处理,结果表明:该算法能适用于不同车型不同光照环境下的弹簧图像定位,并且准确率达到95%以上其次,本文根据弹簧本身的宽度特征,提出了弹簧丢失自动识别算法:在对定位后的弹簧组合进行单个弹簧的分割后,通过比较相邻弹簧的宽度,可判断是否存在弹簧丢失故障。最后研究了基于模式识别理论的弹簧断裂自动识别算法:利用灰度共生矩阵提取图像纹理特征;基于Relief算法进行特征选择;利用最小距离分类器对未知样本进行分类判断。利用模拟的断裂弹簧图片进行样本训练和测试,结果表明:弹簧断裂识别率可达到90%以上。本文的研究为实现弹簧缺损自动检测提供了算法依据,准确率高、通用性好,具有良好的应用前景。
【Abstract】 As the key component of the freight car’s bogie, running gear springs are easily fractured and lost because of the materials defects and abnormal impact, which will lose the role of buffering and damping vibration and threaten freight car’s security.Since always, detection of the springs defects was mainly accomplished by manual work. Not only the efficiency was low, the reliability was poor, but also the labor intensity was big, and there’s a security risk. The trouble of moving freight car detection system could detect the springs defects by the way of man-machine integration, which, to some extent, reduced the labor intensity and improved the defects detection rate. But in fact, the system itself only finished image collection, transmission and some pretreatment, defects recognition was still completed by worker’s naked eye. So there were still many problems and needed to be further improved and upgraded. Consequently, this thesis studied how to automatically identify the springs defects through using image processing techniques in order to improve the defects detection rate and guarantee the freight car’s security.Firstly, on the premise of investigating the research status of automatic recognition of springs defects based on image processing techniques at home and abroad, springs location algorithm was studied in this thesis. It was first to correct the uneven illumination and take binaryzation on image. Then the springs’boundary was found by binary image scanning and features analysis. Through experiment on120images, the results showed that the algorithm was appropriate for springs location of different models and different illumination environment, and the accuracy rate was above95%.Secondly, according to the structure features of spring itself, a spring loss recognition algorithm was proposed. After springs location, division of single spring was operated, then one could judge whether there was spring loss by comparison of the adjacent spring width.Finally, spring fracture recognition algorithm based on pattern recognition theory was studied in this thesis. Gray level co-occurrence matrix was used to extract the image texture feature; Relief algorithm was used to choose the optimal feature subset of image; minimum distance of template matching was used in image recognition. Through experiment on fracture springs images which were simulated, the judgment accuracy rate was above90%.This study provides algorithm reference for the realization of springs defects automatic recognition with high accuracy and good commonality and has a good application prospect.