双目标定研究及其在风机叶片振动模态测量中的应用

发布时间：2018-05-19 20:35

本文选题：风力发电机叶片 + 摄影测量　；参考：《湖南科技大学》2017年硕士论文

【摘要】：摄影测量技术由于其高精度、易操作、实时性及非接触等优点,逐步应用于大型、复杂、柔性结构的位移和振动测量。相机标定指在测量之前通过一定的方法建立物点像素坐标与其三维坐标之间的映射关系,所以标定直接决定标志点三维坐标的提取。风力发电机叶片是一种具有复杂曲面结构的柔性体,其振动模态能反映整机运行状态及潜在故障。本文结合风力发电机叶片振动测量的实际工程背景,针对双目标定问题开展基于神经网络的标定方法研究,并且针对标定中角点的识别和定位精度问题开展了基于SV算子的亚像素级定位方法的研究,具体工作如下:第一:提出基于神经网络的虚拟靶标双目标定方法。相机的标定精度很大程度上取决于标定板在相机视野中的覆盖率,然而大面积的标定靶标制造加工困难而且不易操作。基于此,本文提出了利用单角点靶标构建虚拟立体靶标,并结合BP神经网络的非线性映射特性对双目相机进行标定。实验证明,该方法的标定精度的相对误差为0.0445%,明显较低于MATLAB标定工具箱0.1329%的标定误差。第二:提出双目标定过程中SV算子角点识别定位精度的改进方法。标定过程中,角点识别定位精度是相机标定精度的另一个重要影响因素。SV算子是针对棋盘格角点识别的一种主要方法,较其他特征点识别方法有其特殊性。本文在现有的SV算子角点识别定位的基础上,增加了双线性差值及质心提取的方法,进而将定位精度提高到亚像素级。实验证明,该角点识别定位的改进方法使相机标定的相对误差由原来的0.0445%下降到0.0211%,整体平均误差由原来的0.34727mm下降到0.16458mm,说明SV算子的改进方法有效地提高了标定精度。第三:针对本文所提出的标定方法及其改进方法的精度问题,搭建风力发电机叶片振动模态的双目视觉摄影测量实验台,并分别采取本文方法与MATLAB标定工具箱对实验结果进行分析,并通过叶片模态参数来评价本文标定方法的有效性。
[Abstract]:Photogrammetry has been gradually applied to the displacement and vibration measurement of large, complex and flexible structures because of its advantages of high precision, easy operation, real-time and non-contact. Camera calibration refers to the mapping relationship between pixel coordinates of object points and 3D coordinates before measurement, so calibration directly determines the extraction of 3D coordinates of marker points. The blade of wind turbine is a kind of flexible body with complex curved surface structure. Its vibration mode can reflect the running state and potential fault of the whole machine. Based on the practical engineering background of wind turbine blade vibration measurement, the calibration method based on neural network is studied in this paper. Aiming at the problem of corner recognition and location accuracy in calibration, the sub-pixel level localization method based on SV operator is studied. The main work is as follows: first, a method of virtual target double target location based on neural network is proposed. The calibration accuracy of the camera depends to a great extent on the coverage of the calibration board in the camera field of vision. However, the large area calibration target is difficult to manufacture and operate. Based on this, a virtual stereo target is constructed by using a single corner target, and the binocular camera is calibrated in combination with the nonlinear mapping characteristics of BP neural network. The experimental results show that the relative error of the calibration accuracy is 0.0445, which is obviously lower than the calibration error of 0.1329% in the MATLAB calibration toolbox. Second, an improved method for location accuracy of SV operator corner recognition in the process of double target determination is proposed. In the process of calibration, corner recognition positioning accuracy is another important factor affecting camera calibration accuracy. SV operator is a main method for chessboard grid corner recognition, which has its own particularity compared with other feature point recognition methods. Based on the existing SV operator corner recognition and location, the bilinear difference and centroid extraction methods are added in this paper, and the accuracy of location is improved to sub-pixel level. Experiments show that the improved method of corner recognition and positioning can reduce the relative error of camera calibration from 0.0445% to 0.02111.The overall average error is reduced from the original 0.34727mm to 0.16458mm, which shows that the improved method of SV operator can effectively improve the calibration accuracy. Thirdly, aiming at the accuracy of the calibration method and its improvement method, a binocular visual photogrammetry experiment bench is built for the vibration mode of wind turbine blade. The experimental results are analyzed by using the proposed method and the MATLAB calibration toolbox, and the effectiveness of the proposed calibration method is evaluated by the blade modal parameters.
【学位授予单位】：湖南科技大学
【学位级别】：硕士
【学位授予年份】：2017
【分类号】：TM315

【参考文献】