基于点云拼接的车轮踏面在线三维测量技术研究

发布时间：2018-06-18 06:09

本文选题：车轮踏面轮廓 + 三维测量　；参考：《西南交通大学》2017年硕士论文

【摘要】：随着我国高速铁路的迅猛发展,大大地提升了我国铁路的客运运输能力。为保证行车安全及旅客生命财产安全,必须时刻关注轮对的尺寸、踏面的各项参数变化。基于光栅投影的三维测量技术广泛应用与自动化检测、质量控制、机器视觉、生物医学等领域,具有非接触式、高精度、高分辩率等特点。本文以CRH2A型动车组样板轮为研究对象,结合三维测量技术与点云拼接技术应用于车轮踏面外形轮廓360°测量中,从理论和实验两个方面进行了深入研究,具体研究内容如下:首先,在车轮踏面点云获取方面,研究了基于光栅投影的三维测量技术的原理。重点针对相位展开技术,在质量图引导算法的基础上,利用相邻像素点之间的相对可靠度作为展开路径,然后通过区域划分与合并进行区域设置后独立展开,并与枝切法、质量图引导算法以及最小二乘法进行了仿真及实验对比分析,其均方误差比质量图引导算法降低约35%,展开速度提升约10倍,且有效地抑制了路径相关算法中的噪声扩散现象。其次,介绍了点云拼接技术的基本流程及原理,了解传统ICP算法的数据配准机制。利用基于协方差矩阵法对点云数据进行粗拼接,并与基于点特征直方图方法仿真对比分析。研究一种以点到切平面为迭代目标函数,利用kdtree算法对点云进行加速匹配搜索,最后基于刚性变换一致性进行原则约束的改进ICP算法。仿真结果表明,基于协方差矩阵方法完成粗拼接的时间仅需0.1s,更适合车轮踏面点云数据粗拼接。改进的ICP算法的迭代精度比其他精拼接算法综合提升约10%,且精拼接时间不超过3s。最后,设计并搭建了列车车轮踏面三维轮廓检测试验平台,结合光学三维测量技术与三维点云拼接技术对车轮踏面外形轮廓实现360°在线测量,且利用改进的ICP算法完成整个车轮的拼接时间为148.48s,平均精拼接精度为0.0638mm,提升了迭代的精度与在线检测的效率。
[Abstract]:With the rapid development of high-speed railway in China, the passenger transport capacity of Chinese railway has been greatly improved. In order to ensure the traffic safety and the safety of passengers' life and property, we must always pay attention to the size of wheelset and the change of parameters of tread. The 3D measurement technology based on grating projection is widely used in the fields of automatic detection, quality control, machine vision, biomedicine and so on. It has the characteristics of non-contact, high precision, high resolution and so on. In this paper, the model wheel of CRH2A EMU is taken as the research object, and the 3D measurement technology and point cloud splicing technology are applied to the measurement of the profile 360 掳of the wheel tread. The theoretical and experimental studies are carried out in depth. The main contents are as follows: firstly, the principle of 3D measurement based on grating projection is studied. Focusing on the phase unwrapping technique, based on the quality map guidance algorithm, the relative reliability between adjacent pixels is used as the expansion path, and then the region is set up independently by region division and merging, and then it is expanded independently with the branch cutting method. The simulation and experimental results show that the mean square error is about 35 times lower and the expansion speed is about 10 times higher than that of the quality map guidance algorithm, and the noise diffusion phenomenon in the path correlation algorithm is effectively suppressed. Secondly, the basic flow and principle of point cloud splicing technology are introduced, and the data registration mechanism of traditional ICP algorithm is understood. Covariance matrix method is used to combine the point cloud data, and the results are compared with the point feature histogram method. In this paper, an improved kdtree algorithm based on the consistency of rigid transformation is proposed, in which the point to tangent plane is taken as the iterative objective function, and the point cloud is searched by kdtree algorithm. The simulation results show that the covariance matrix method takes only 0.1 s to complete rough stitching, which is more suitable for rough stitching of wheel tread point cloud data. The iterative accuracy of the improved ICP algorithm is about 10% higher than that of other fine stitching algorithms, and the time of fine stitching is less than 3 s. Finally, a testing platform for 3D profile detection of train wheel tread is designed and built. Combined with optical 3D measurement technology and 3D point cloud splicing technology, the profile of wheel tread profile is measured on line 360 掳. By using the improved ICP algorithm, the stitching time of the whole wheel is 148.48 s, and the average precision is 0.0638 mm, which improves the precision of iteration and the efficiency of on-line detection.
【学位授予单位】：西南交通大学
【学位级别】：硕士
【学位授予年份】：2017
【分类号】：U270.7;O439

【相似文献】