【作者】 赵萍；

【导师】 赵文珍；

【作者基本信息】 沈阳工业大学 ， 机械设计及理论， 2013， 博士

【摘要】 数控磨削是提高复杂曲面加工精度和表面质量的有效方法，但砂轮廓形误差对加工精度影响较大，因此砂轮的修整一直是复杂曲面磨削工艺的瓶颈问题之一。普通磨料砂轮修形容易，但磨损较快，需要频繁修形，使加工效率降低。CBN等高硬度磨粒砂轮不需要频繁修形，但修形困难。基于砂轮实际廓形的曲面包络磨削是解决上述问题的有效方法之一，其必要条件是及时对砂轮廓形进行高精度测量。为此，本文提出基于机器视觉的砂轮廓形样板测量方法，并对相关关键技术进行了研究。机器视觉测量系统集光学、传感器、图像处理和模式识别等多领域多学科的关键技术于一体，可实现对物体尺寸或相对位置的快速测量。但由于系统的复杂性，在测量过程中存在多种随机噪声和系统误差，在较大视野内实现微米级的高精度测量至今仍是一个难题。本文将砂轮廓形复映样板为媒介实现对砂轮廓形的高精度测量，通过对视觉测量理论及关键技术的分析和研究，提出一套完整的系统设计方案，使视觉测量系统的测量精度达到±5μm，可以满足砂轮廓形的测量要求。论文主要工作有：（1）建立了机器视觉测量系统，为砂轮廓形样板的高精度测量奠定了硬件基础。该系统不仅可以实现砂轮廓形复映样板的精密测量，而且可以代替三坐标测量机对各种具有复杂二维曲线轮廓的小尺寸板类零件廓形进行精密测量。（2）根据边缘的结构特征，采用各向异性双边滤波算法（ABF），同时满足了平滑图像和保持边缘梯度的滤波要求。在空域内采用各向异性高斯核，对双边滤波的空域权因子进行了定义，使其在边缘的切向进行大尺度的滤波，最大限度地减小噪声的影响；在边缘的法向采取小尺度的滤波，尽可能地保持边缘梯度不变。在值域内采用灰度高斯核定义权因子，进一步减小滤波对边缘梯度的影响。该算法较好地解决了一般滤波器存在的图像平滑和边缘保持不能兼顾的矛盾。（3）在Facet曲面模型亚像素边缘检测方法的基础上，提出并实现了一种三级逼近的亚像素边缘检测方法。先用改进的Canny算法进行粗定位，提取单像素精度边缘；然后用以像素边缘为中心的5′5邻域数据集拟合Facet曲面模型，根据阶跃边缘的特征在像素边缘内确定亚像素边缘位置，实现边缘的精定位；最后为了消除滤波不完全和计算误差对边缘提取结果的影响，对提取的亚像素边缘点进行分段曲线拟合。该方法解决了Facet模型提取亚像素边缘存在的效率低、漏检率高和边缘点异常波动等问题。（4）针对机器视觉系统存在复杂系统误差的特点，提出一种基于直线成像特征的系统综合标定方法，有效地保证了测量精度。该方法通过对1等量块的平行直线边缘提取，建立量块边缘空间物点位置和像点位置的原始对应关系，利用量块边缘的理想直线特征，通过统计计算对测量系统进行综合标定，建立描述空间物点位置和像点位置的相互对应关系的二元三次标定函数。该标定方法采用工程通用量块作为标定工具，操作简便，容易实现，通用性好。可以综合修正光学畸变误差、透视误差、传感器位置误差、边缘检测算法的定位误差等各项系统性误差。以该标定方法为基础，对在系统视野内不同方位的量块进行多次测量，结果证明基于该标定方法的测量精度能够达到±5μm。更多还原

【Abstract】 CNC grinding is an effective method to improve machining accuracy and surface quality for the complex surface, but the profile error of grinding wheel will greatly influence machining accuracy, so surface dressing of grinding wheel has been a bottleneck problem in the field. Ordinary grinding wheel is easy to dress, but it wears quickly. It is necessary to frequently dress the grinding wheel, which reduces the grinding efficiency. The grinding wheel with high hardness abrasive, such as CBN grinding wheel, does not need frequent modification, but the modification is difficult for this kind of grinding wheel. Surface envelope grinding method that grinding trajectory is calculated with actual profile is one of effective method to solve the problem. The premise of the method is the prompt measurement for the grinding wheel profile. Therefore, the grinding wheel profile templet measurement method based on machine vision is put forward and the related key technologies are researched in the paper.Machine vision measurement system involves an integration of key techniques such as optical, sensor, image processing, pattern recognition, and so no. It realizes the rapid measurement of object size or relative position. But because of the complexity of the system, there are many random noises and system errors in the measuring process. So far it is a difficult problem to achieve micron level measurement in a larger field of vision. This paper, taking the wheel profile reflection model as a medium to realize the high precision measurement of the wheel profile, through the analysis and research on its detection theory and key technology, puts forward a complete set of system design to make the measurement accuracy of the vision measurement system to±5μmand to meet the requirement of grinding wheel profile measuring. The main contributions of the paper are listed as following.(1)Grinding wheel profile measuring system is built based on machine vision measurement technology, which establishes the hardware foundation for the high precision measurement of wheel profile model. The system not only realizes the precise measurement of wheel profile, but also can replace CMM to realize precision measurement of all kinds of small size plate parts with complex two-dimensional curved contour.(2)According to the structural characteristics of the edge, the anisotropic bilateral filtering algorithm (ABF) is adopted, which can meet the filtering requirements of the image smoothing and edge preserving at the same time. In the spatial domain, the anisotropic Gauss nuclear function is used to redefine the bilateral filter weight, filtering with large scale in the tangential direction in order that the noises are reduced as much as possible and filtering with small scale in the normal direction in order to maintain the edge gradient as far as possible. In the range domain, the gray Gauss nuclear function is used to further reduce the effect of filtering on edge gradient. It solves the problem that the common filter can’t take into account the image smoothing and edge preserving at the same time.(3)Based on Facet surface model subpixel edge detection algorithm, a three-level-approximation subpixel edge detection algorithm is proposed and realized. This method is to use the improved Canny algorithm for coarse positioning to extract the single-pixel-precise edges, and then use the5′5neighborhood data of the pixel edge points to fit Facet surface model and calculate subpixel location in the pixel edge based on the characteristics of the step edge, and finally piecewise fitting edge points as curves to remove the effect of incomplete filtering and calculation error on edge extraction. As a result, the problem of low efficiency, low localization and abnormal fluctuation of edge points for the Facet model to extract edge is fundamentally solved.(4)Because there are various system errors in machine vision system, a system comprehensive calibration method based on straight line imaging characteristics is proposed to effectively ensure the measurement accuracy. According to the extraction of parallel straight line edges of1grade gauge blocks, the original correspondence between the space point positions and the image point positions of the edges is built. Using the ideal linear feature of gauge block edges, comprehensive calibration of measurement system is done through statistical calculation, and a two-parameter second-order polynomial which describes the corresponding relation between the space point positions and the image point positions is built. Because of using the general engineering gauge block as calibration tool, the calibration method is simple to operate and is easy to realize and has good versatility. It can comprehensively correct all kinds of system errors such as the optical distortion, the perspective error, the sensor position error and location error of edge detection algorithm. Based on the calibration method, multiple measurement experiments of different azimuth gauge blocks in the field show that the precision of the measurement system can reach±5μm.更多还原