列车轮对几何参数在线检测关键技术研究

发布时间：2018-01-13 18:10

本文关键词：列车轮对几何参数在线检测关键技术研究　出处：《吉林大学》2017年博士论文　论文类型：学位论文

【摘要】：本文结合列车轮对制造企业生产实际问题和吉林省科技发展计划重点科技攻关项目《列车轮对制造质量在线检测关键技术研究》(项目编号:20160204005GX),以列车轮对为具体应用对象,研究了列车轮对车轮型面圆弧曲线半径、车轮滚动圆直径、车轮径向跳动等几何参数和车轴端面刻印字符的在线检测技术。根据列车轮对制造质量在线检测的设计要求,基于机器视觉和激光检测技术,研究开发了列车轮对几何参数和轴端刻印字符在线检测系统,设计了利用轮对车轴中心孔和轴承表面为基准的二种定位支撑机构、万向节旋转驱动机构、高精度水平/垂直移动机构、底座龙门支撑机构等机械结构;选择了由AT相机和线激光器构成的结构光传感器、数据采集卡、伺服电机、伺服放大器、运动控制卡等数据采集与电气控制元件;研发的列车轮对制造质量在线检测工程样机经生产现场实验测试,对列车轮对车轮型面圆弧曲线半径、车轮滚动圆直径、车轮径向跳动等几何参数和车轴端面刻印字符的在线检测结果能够满足设计要求。论文分析了视觉检测摄像机模型中摄像机坐标系、世界坐标系、图像像素坐标系和图像物理坐标系等四个坐标系之间的变换关系,研究了利用线结构光线传感器进行测量的数学模型。针对列车轮对车轮型面检测中存在的因轮缘遮挡导致测量盲区的工程实际问题,提出了采用二个线结构光传感器完成列车轮对车轮轮缘检测的技术方案,并利用Halcon软件中的标定工具箱完成了对两台摄像机的标定。研究了构成列车轮对轮缘曲线上各段圆弧半径的检测方法。针对轮缘曲线上BC段和EF段圆弧较短,且测量数据较少,按最小二乘法计算圆弧半径存在较大误差的问题,提出了通过基于圆弧设计端点和理论半径为约束条件的被测圆弧半径计算方法,有效提高了圆弧半径的测量精度;提出了应用人工蜂群算法搜索多圆边缘的检测算法,利用人工蜂群算法的优势,提高了算法的搜索效率;针对由多段圆弧组成的轮缘曲线,利用人工蜂群算法,优化了列车轮对型面曲线的检测数据;并通过实验验证了所提出的检测算法能够有效的提高列车轮对轮缘曲线上各段圆弧半径的检测精度。提出了在频域内列车轮对车轮型面三维点云数据进行配准的方法,利用编写的MATLAB程序,完成了二个线结构光采集的列车轮对车轮型面三维点云的准确配准,解决了单个线结构光传感器因轮缘遮挡只能采集到部分车轮型面数据而无法计算整个车轮型面参数进行检测的工程实际问题。针对列车轮对轴端刻印字符由于深浅不同及光照角度而产生的轮廓边缘模糊的问题,本文提出两种刻印字符识别方法。其一是根据字符图像边缘灰度梯度突变的原理并采用小生境人工蜂群算法,模拟蜂群寻找蜜源的群体行为,利用蜂群迭代搜索方法来搜索字符图像边缘点,并根据字符图像的特点,在人工蜂群算法中引入小生境技术,将蜂群分组及设定小生境半径引导算法收敛到每个对应的字符上,达到全局优化求解的能力及求解精度的平衡。其二是针对列车轮对轴端刻印字符与背景无色差、具有深度信息的特点,提出了采用线结构光传感器采集轮对轴端字符点云信息,并结合总体最小二乘平面拟合的方法,在字符点云深度数据中提取字符的轮廓信息。通过实验,验证了所提二种方法的有效性。实验表明,本文研究开发的列车轮对制造质量在线检测系统具有较高的检测精度,同时能够满足列车轮对生产在线检测速度要求。本文研究所取得的成果,对于应用智能检测技术完成列车轮对制造质量在线检测具有理论意义和实用价值。
[Abstract]:In this paper the train wheel production problems and Jilin province science and technology development plan of key scientific and technological project "the train wheel manufacturing quality on-line detection technology" (project number: 20160204005GX), in order to train for the specific application object, the circular train wheel wheel profile curve radius, wheel diameter, wheel radial runout the geometry parameters and the character of the online imprint axle end face detection technology. According to the design requirement of the train wheel manufacturing quality on-line detection, machine vision and laser detection technology based on the research and development of train wheelset geometric parameters and axial seal character online detection system, design the supporting mechanism by using two kinds of positioning of wheelset axle center hole and the bearing surface the benchmark, gimbal rotation driving mechanism, a vertical movement mechanism of high precision level / Longmen base supporting mechanism and other machinery Structure; select the structure light sensor by a AT camera and a laser line, data acquisition card, servo motor, servo amplifier, motion control card, data acquisition and electrical control components; the quality of online train wheel detection prototype manufacturing engineering research and development by production testing, the arc train wheel profile curve radius. Wheel diameter, wheel radial runout online detection results of geometric parameters and the axle end face engraved characters can meet the design requirements. This paper analyzes the visual detection camera coordinate camera model, the world coordinate system, the transformation between the four coordinate image pixel coordinates and image coordinates of the physical and mathematical model of the measured by line structure light sensor. According to the existing train wheel profile detection due to occlusion rim measurement blind area project The actual problem is proposed by using two line structured light sensor complete technical scheme of train wheel rim detection, and completed the calibration of two cameras using the calibration toolbox of Halcon software. The research of detection method of each arc radius of the train wheel set flange curve. The curve on the rim BC section and EF section of arc short, and there are few data, according to the least square method is used to calculate the radius of the large error problem, puts forward the calculation method of the radius was measured and the theory based on the radius of the circular arc design endpoint as the constraint conditions, effectively improve the measurement accuracy of radius; detection algorithm is proposed using artificial bee colony algorithm multi circle edge. The use of artificial bee colony algorithm has the advantage of improving the search efficiency of the algorithm; for the rim consists of multi circular curve, using artificial bee colony algorithm, optimization The test data of train wheel curve; and the experimental results show that the algorithm proposed can improve the detection accuracy of each radius curve on the train wheel rim is proposed. The method of registration in the frequency domain in wheel on the wheel surface three-dimensional point cloud data, using the MATLAB program, complete the accurate registration of 3D point cloud surface of the train wheel wheel type two line structure light collection, solve the single line structured light sensor for occlusion only capture part of the wheel rim surface data to calculate the wheel surface engineering detection parameters. According to the train wheel shaft end seal characters due to different depth contour and the angle of illumination caused by fuzzy problems, this paper proposed two kinds of graphic character recognition method. It is based on the principle of character image edge gray gradient mutation And the niche of artificial bee colony algorithm, simulation of group behavior for the nectar of the hive, the use of colony iterative search method to search the character image edge points, according to the character of the image, the niching technique in artificial bee colony algorithm, the swarm grouping and set niche radius of convergence to guide each algorithm corresponding to the character, ability and the solution accuracy of the global optimal solution of the balance. The second is for the train wheel shaft end seal character and background without color, has the characteristics of depth information, the line structured light sensor acquisition wheel shaft end character of point cloud information, combined with the method of total least squares plane fitting, extracting the contour information of the character in the character. The cloud depth data. Through the experiment, verify the effectiveness of the proposed two methods. Experimental results show that the online manufacturing quality of train wheels in this research and development The detection system has high detection accuracy and can meet the online detection speed requirement of train wheel production. The research results in this paper are of theoretical significance and practical value for the application of intelligent detection technology to accomplish online detection of train wheel manufacturing quality.

【学位授予单位】：吉林大学
【学位级别】：博士
【学位授予年份】：2017
【分类号】：U279.32;TP391.41

【相似文献】