基于双目线结构光的轮对测量技术研究

发布时间：2018-05-16 00:31

本文选题：视觉测量 + 双目线结构光　；参考：《长春工业大学》2017年硕士论文

【摘要】：随着我国轨道交通运行速度的不断提升,列车运行安全问题也变得越来越突出和重要。轮对作为轨道交通中重要的支撑运动部件,承受较为复杂荷及冲击,列车的高速运行要求轮对的各项参数满足列车高速运行时的技术要求,检测其关键尺寸对列车安全运行有着重要的意义。本文结合吉林省科技发展计划项目及某客车厂具体实际要求,根据线结构光检测原理,对双目线结构光轮对检测技术进行了研究,主要研究内容及成果如下:(1)为得到较为完整的车轮踏面轮廓数据,提出了利用两套线结构光组成的双目线轮对测量方法。对于两套线结构光获得的点云,提出在频域中进行配准,该方法通过将相同特征的点云还原到数字矩阵当中,接着进行极坐标转换,求其互功率谱并的傅里叶逆变换,求取突击函数的峰值位置,得到两点云的相对旋转角度,同理可求得两点云矩阵的平移向量。(2)对点云频域配准方法进行了精度验证。点云平移验证可以得出:平移配准不存在线性或者非线性的误差,同时确定放大系数对检测精度的影响。对点云旋转配准精度验证得出:配准误差是呈周期变化的奇函数,提出将旋转配准误差进行分段拟合,然后根据分段拟合后的直线与求得的初值进行迭代取优,最终补偿至实际角度值。(3)对传感器测量时本身存在的噪声干扰进行了分析,在同一点云添加噪声后再次检测配准参数,可以验证传感器本身存在噪声影响并不影响本文提出的配准方法。(4)针对配准后点云整体倾斜的问题,提出对配准后车轮点云内侧面数据进行主成分分析的方法来拟合基准平面,根据平面参数对点云进行整体旋转,将旋转后点云在直径方向上标定,确定了直径标定系数,给出了车轮三维重建模型。(5)本文搭建了双目线结构光的轮对测量实验平台,对标准轮对及其他被测轮对进行了多次测量实验。实验证明该测量系统的车轮直径重复测量精度小于等于0.07mm,端跳、径跳的重复测量精度小于等于0.04 mm;直径的测量误差小于0.05mm,端跳的测量误差小于0.06mm,径跳的测量误差小于0.04mm,满足轮对测量精度要求。
[Abstract]:With the increasing speed of rail transit in our country, the safety problem of train operation becomes more and more prominent and important. As an important supporting component in rail transit, wheelset is subjected to more complex loads and shocks. The parameters of wheelset are required to meet the technical requirements of high-speed train operation. It is of great significance to detect the key dimensions for the safe operation of the train. Based on the project of Jilin province science and technology development plan and the practical requirements of a bus factory, the detection technology of binocular structure light wheel pair is studied according to the principle of line structure light detection. The main research contents and results are as follows: (1) in order to obtain more complete profile data of wheel tread, a measurement method of binocular wheelset is proposed, which is composed of two sets of structured light. For the point clouds obtained by two sets of line structured light, a registration method in frequency domain is proposed. By reducing the point clouds with the same characteristics to the digital matrix, the polar coordinates are transformed to obtain the inverse Fourier transform of the cross-power spectrum. The peak position of the shock function is obtained and the relative rotation angle of the two-point cloud is obtained. The translation vector of the two-point cloud matrix is obtained by the same principle. The accuracy of the registration method in the frequency domain of the point cloud is verified. It can be concluded that there are no linear or nonlinear errors in the translation registration and the effect of magnification factor on the detection accuracy is determined. To verify the accuracy of point cloud rotation registration, it is concluded that the registration error is an odd function with periodic variation. It is proposed that the rotation registration error be fitted in sections, and then iteratively optimized according to the straight line and the obtained initial value after piecewise fitting. Finally compensated to the actual angle value. (3) the noise interference existing in the sensor measurement is analyzed, and the registration parameters are detected again after adding noise to the same point cloud. It can be verified that the sensor itself has the effect of noise and does not affect the registration method proposed in this paper. Aiming at the problem of the overall tilt of the point cloud after registration, a principal component analysis method is proposed to fit the datum plane for the data of the inside side of the point cloud of the wheel after registration. According to the plane parameters, the point cloud is rotated, the point cloud is calibrated in the diameter direction, the calibration coefficient of the diameter is determined, and the three-dimensional reconstruction model of the wheel is given. Several measurement experiments were carried out on standard wheelsets and other wheel pairs under test. The experimental results show that the precision of repeated measurement of wheel diameter is less than 0.07mm, and the end jump is obtained. The precision of repeated measurement is less than 0.04mm, the error of diameter is less than 0.05mm, the error of end jump is less than 0.06mm, and the error of diameter slip is less than 0.04mm.
【学位授予单位】：长春工业大学
【学位级别】：硕士
【学位授予年份】：2017
【分类号】：U270.7;TP391.41

【相似文献】