【作者】 李静；

【导师】 高晓蓉；

【作者基本信息】 西南交通大学 ， 光学工程， 2015， 硕士

【摘要】 机车零部件中,螺栓起着固定各个部件与车体的重要作用。在机车运行过程中,车体振动、外力冲击和疲劳锈蚀等因素的作用,螺栓可能出现丢失、松动等故障,危及行车安全。为了保障机车运行安全,除了利用定期检修的方式,还可在铁路轨道自动化设备进行列车实时图像状态监测,应用于列车日常检修。但由于车底部件结构复杂,对于关键螺栓故障的排查仍需人工判别,排查效率和可靠性极大地依赖于检修人员的工作状态,检修质量无法保持稳定的水平,且难以评估。而目前,图像识别技术已广泛运用于人脸识别、行人检测和车辆检测等领域,取得了一定成效,其对于机车运行故障图像检测的意义是可以预见的。本文基于结合数字图像处理技术和图像识别算法,针对HXD2C型货车车底关键螺栓进行了检测算法的研究,实现了螺栓丢失故障识别系统功能,对于提高机车运行故障自动检测水平具有指导参考的意义。本文首先在调研图像识别技术和螺栓故障检测技术的国内外研究现状基础上,对车底图像进行了图像预处理算法研究,对两类螺栓区域图像进行了定位算法的研究和实现,为实现螺栓检测算法制备了分类样本库。其次,本文重点研究了两种基于图像识别原理的螺栓检测算法,分别是基于梯度直方图(HOG)特征的支持向量机(SVM)螺栓检测算法和基于Haar特征的Adaboost分类器螺栓检测算法,并采用两种算法实现了对钩缓装置挡板螺栓和心盘螺栓的检测。最后,在对比分析两类算法性能的基础上,实现了螺栓丢失故障识别系统设计,并进行了故障识别系统测试,结果表明：采用HOG-SVM螺栓检测算法的螺栓丢失故障识别系统识别率在90%以上。本文的研究为实现关键螺栓故障自动检测提供了算法依据,并整合螺栓检测算法实现了故障识别系统设计,测试表明其针对螺栓丢失故障识别准确率达到了预期,具有良好的应用前景。更多还原

【Abstract】 As the most common parts of locomotive, bolt plays an important role to fasten other parts to the locomotive. There are some reasons such as vehicle vibration, external force and metal fatigue or corrosion that cause bolts to miss or loose from the car body, which may threaten the safety of running trains. In addition to maintenance the trains termly in the servicing depot, there is also an automation system to monitor the trains’condition through capturing images. However, the structure of the vehicle bottom is too complicated to inspect by machine, so the troubleshooting for the bolts still rely on the repairmen. In recent years, image recognition technique has been widely used in such fields as face recognition, pedestrian detection and vehicle detection, and it is effective. Combining the image recognition algorithm and digital image process technique, bolts fault for HXd2C freight train detection technique is investigated in the thesis to achieve the bolts miss fault detection system, which is the important reference for improving the detection efficiency of the train faults.Firstly, state of the art neural network recognition techniques and bolts fault detection techniques has been introduced in the thesis. And then the support vector machine and the Adaboost algorithms are chosen to detect bolts in the vehicle bottom image. The vehicle bottom image is preprocessed by Gamma and image enhancement algorithms. Bolts zone images are positioned in the bottom images based on the position of bolts and coupler, and bolts samples are made from these pictures. The samples are used to train the two classifiers, HOG-SVM classifier and Haar-Adaboost classifier. Then using the two classifiers to detect bolts in test images. Based on the above work, a bolts miss fault detection system is designed and tested. The experiments results show that the bolts miss fault detection system based on HOG-SVM detection algorithm has a detection rate as 90%.The thesis provides algorithm evidence for key bolts fault detection system, and also combines the detection algorithm with the system to implement the design of fault detection system. The experiments show that it achieve the expected effectiveness, and it has a good prospect in the future.更多还原