【作者】 赵金才；

【导师】 刘书桂；

【作者基本信息】 天津大学 ， 测试计量技术及仪器， 2005， 博士

【摘要】 随着现代制造技术的迅速发展,特别是计算机集成制造系统日益备受重视并得以大规模实施,使得三坐标测量机正逐渐成为机械制造业特别是自动化生产中实现质量控制的主导检测设备,但传统三坐标测量机的自学习测量方式和人工编程测量方式已经制约了其功能的进一步发挥,使其不能够很好得融入到自动化生产环境中去。研究适合现代化生产的自动化、智能化三坐标测量机技术已经成为制造业发展的关键技术之一。在智能三坐标测量机的若干关键技术中,CAD与坐标测量机的接口技术(即CAD数据的识别与提取)和被测零件位姿的自动识别技术是整个三坐标测量机智能体系中的重中之重,也是三坐标测量机系统其它智能技术研究的前提和基础。前者根据零件的CAD设计数据文件自动完成对零件特征和检测项目等信息的识别与提取,从而确定被测零件所需的检测项目以及零件的特征尺寸、定位尺寸等内容;后者则是利用视觉系统通过对被测零件进行成像,确定零件在三坐标测量机工作台上的放置姿态、放置位置和放置方向,从而建立起零件坐标系,为后续检测路径的规划等工作奠定基础。本论文隶属于“三坐标测量机智能技术”课题,同时也是“零件轮廓测量机的研制”课题中的重要组成部分。论文的主要目标是对上述两个关键技术进行研究,从而使三坐标测量机的测量过程尽可能地减少人为的干预和管理,进一步实现三坐标测量机测量过程的自动化、智能化。围绕如何实现零件信息的自动提取和零件位姿的自动识别,论文对下列主要工作和创新之处进行了深入研究:一、全面回顾并深入分析了智能三坐标测量机系统的组成,进一步明确了实现三坐标测量机自动化、智能化的关键技术。二、针对采用Pro/ENGINEER软件所设计零件的信息提取,基于Pro/ENGINEER及其二次开发工具Pro/TOOLKIT进行了零件信息自动提取应用程序的开发。1.深入剖析了利用Pro/TOOLKIT所提供的资源进行Pro/ENGINEER应用程序二次开发的机制和方法,并对应用程序的工作模式及开发环境的设置等内容做了全面分析。2.开发了用于Pro/ENGINEER拉伸特征信息提取的应用程序,深入研究了这类特征几何信息、公差信息的提取问题。3.按照旋转截面图形形状的不同,将采用Pro/ENGINEER软件设计的旋转特征分为实心旋转特征和空心旋转特征两大类,并分别进行了几何信息、公差信息自动提取的研究与开发。三、进行了零件位姿自动识别系统的深入研究1.将CCD摄像机固定在三坐标测量机的Z轴上,通过测头坐标系建立起了摄像机坐标系与机器坐标系之间的联系,同时充分利用三坐标测量机的精确移动与白色陶瓷标准球的良好光学特性,巧妙地设计出了摄像机坐标系与测头坐标系的标定方法,利用该方法进行非共面标定不仅能够标定出摄像机的内部参数,同时还能够确定摄像机坐标系与测头坐标系的旋转与平移关系。2.为实现图像处理的自动化,设计了图像二值化阈值的自动选取算法。将贪婪思想引入到基本遗传算法中构成了贪婪遗传算法,并将所设计的贪婪遗传算法与最大类间方差法相结合开发出了基于贪婪遗传算法的阈值自动选取算法。该算法在零件位姿自动识别系统的图像处理方案中能够充分满足图像阈值化的需要,并且具有良好的实时性。3.全面深入地分析了图像矩及矩不变量理论,根据矩不变量所具有的旋转、平移以及比例不变性,将其创造性地运用到零件位姿自动识别系统的图像匹配技术中,这种匹配方法不仅具有受约束少的特点,而且具有运算量小,匹配速度快的优势,利于保证自动测量系统对实时性的要求。4.基于BP神经网络良好的模式识别能力,以及其所具有的自适应学习、联想记忆、非线性变换和很强的容错能力等特点,提出了利用零件虚拟图像的矩不变量对BP神经网络进行训练,将训练好的神经网络作为分类器,根据被测零件实际图像的边缘矩不变量进行模式识别的方法,该方法不仅能够实现零件姿态的自动识别,而且具有对被测零件进行判别的功能。5.根据零件图像的边缘矩,实现了零件在图像坐标系中位置和方向的自动识别。利用各坐标系之间的转换关系,最终实现了零件在三坐标测量机机器坐标系中位置和方向的自动识别功能。更多还原

【Abstract】 With the rapid development of manufacturing industry, especially the highly valued and being put in practice cosmically of Computer Integrated Manufacturing System, coordinate measuring machine (CMM) is becoming the major inspection equipment in quality control system of automatic production. Though CMM has two modes of automatic measurement: the programming mode and the self-learning mode,it does lie behind the requirement of automatic manufacturing industry and does limit the function applications of CMM in quality control. Therefore, the study on intelligent CMM is one of the key techniques for the development of manufacturing industry. Among the key techniques in intelligent CMM, the CAD/CMM interface technique and the automatic recognition for the position and orientation of parts are the most important key techniques. They are the precondition and foundation of other techniques in intelligent CMM system. The former can realize the automatic extraction of part features and inspection items based on the CAD design data files, which provide the necessary information such as inspection item, feature size and orientation dimension. The latter can recognize the posture, position and orientation by capturing the image of measured parts using CCD camera, which established the foundation for setting up part coordinate frame and programming the inspection path etc.This dissertation is subject to the project of“The Intelligent Techniques of CMM”. At the same time, it is a component of“the development of measuring machine for parts outline”, too. The goal of this thesis is to farther realize the intelligence of CMM for reducing operator’s participation to the least extent through the study of the above two pivotal techniques. Surrounding how to realize the automatic extraction of parts information and the automatic recognition for the position and orientation, the major research and creative work of this thesis includes the following aspects:Ⅰ. The composing of intelligence CMM system is roundly reviewed and thoroughly analyzed. And the key techniques to realize the intelligence of CMM are elucidated deeply.Ⅱ. For the information extraction of part designed by Pro/Engineer, the automatic extraction application is developed based on the software Pro/Engineer and the development tools Pro/TOOLKIT.1. The Development mechanism and methods of Pro/TOOLKIT application areanalyzed thoroughly. The work modes of application and the setting of development environments are discussed detailed.2. The application to extract the relative information of extrude feature is researched with emphasis on the extraction work of geometry information and tolerance information.3. Based on the section shape, the revolution feature is classified into two kinds: the solid revolution feature and the empty revolution feature. And their information extraction applications are developed.Ⅲ. The automatic recognition system for the position and orientation of parts is researched deeply.1. The CCD camera is fixed on the Z axis of CMM, and the relation between CCD coordinate frame and CMM coordinate frame is established by using probe coordinate frame as“bridge”. At the same time, the calibration method between CCD coordinate frame and probe coordinate frame is designed dexterously through taking full advantage of the precision movement of CMM and the nicer optics characteristic of white ceramic standard ball. By this calibration method, both the interior parameters of CCD and the coordinate frame relation of CCD and probe can be made certain in one calibration process.2. To realize intelligence, the automatic choosing arithmetic of image threshold is designed based on the improved Genetic Algorithm. In this method, the Greedy Genetic Algorithm and Otsu means are integrated. And the designed arithmetic can satisfy the real-time image processing system perfectly.3. The theory of moment invariants for the edge images is studied. And it is applied into the image matching of automatic recognition system for the position and orientation in a creative way. This image matching method is favorable to guarantee the real-time requirement because of the characteristics such as less restrict conditions, less calculation, matching rapidly and so on.4. Based on the ability of patter recognition and the characteristics such as self-adaptable learning, non-linear, strong fault tolerance capability and so on, the BP (Error Back Propagation) Neural Network is used as classifier. The moment invariants of virtual images of part are used to train the Neural Network. And the pose of measured part can be recognized by using the moment invariants of captured image as the input of trained Neural Network because of the matched virtual image representing the pose of part on the worktable.5. The position and orientation of parts in the image is recognized by calculating the edge moment of part image. And the automatic recognition function for the position and orientation of part in machine coordinate frame is realized through the coordinate transform between CCD coordinate frame and CMM coordinate frame.更多还原