基于机器视觉的在线高速检测与精确控制研究及应用

发布时间：2018-07-21 15:34

【摘要】：机器视觉检测是建立在计算机视觉和图像处理基础上的一门新兴检测技术，它通过图像处理获得被测工件对象的各种可描述参数，并对参数进行理解和判断，最终应用于实际检测、测量和控制，具有非接触、测量精度高、适用范围广和自动化程度高的特点。由于机器视觉在线检测设备安装在生产线上，其检测速度必须与高速生产线保持高度同步，实现其同步的关键技术是相机拍照的精确控制和图像的高速检测。因此研究机器视觉在线检测的相机高速拍照精确控制和图像的高速检测，开发研制自主知识产权的智能在线检测设备，对促进高速视觉检测的理论探索和创新及满足当前国内智能制造业市场的迫切需求皆具有重要的意义。本文围绕实现机器视觉在线检测的关键技术，对机器视觉中的高速准确控制和图像的高速检测进行了研究，并以皇冠瓶盖的在线检测为应用案例给出了可行的设计方案，主要工作概括如下：首先，提出了基于两层网络控制理论的机器视觉在线检测系统组成结构，为实现高速的机器视觉在线检测提供了新的研究思路和方向。将图像处理任务放在高层处理单元，将系统控制放在本地控制单元，各单元根据需要处理的任务分别采用相应的处理器，化解了信号集中处理时处理器负担过重的问题，同时采用模块化设计各子系统，便于安装调试、维护和扩展。其次，提出了基于迭代学习控制和卡尔曼滤波的高速准确控制方法，实现了相机位置受限条件下高速运动工件图像的精确抓拍，解决了视觉检测中复杂现场环境下的工件图像高质量精确采集的难题。同时建立了基于迭代学习控制和卡尔曼滤波相结合的相机控制模型，并理论推导和分析了模型的收敛性和收敛范围，给出了数值仿真和实际的实验结果。第三，提出了基于局部能量离散路径水平集方法的图像边缘搜索策略。将窄带水平集搜索减少为按照窄带中有限条数的线路搜索，极大降低了边缘搜索的数据量。同时考虑水平集内外的局部能量因素，克服了现场图像中出现的干扰以及光照不均匀引起的误差。第四，提出了圆区域投影直方图旋转不变特征，将2D匹配数据转化为1D数据，提高了匹配效率，为高速检测提供了前提。同时提出采用稀疏表示的方法进行图像的旋转匹配和瑕疵检测的策略，该策略在执行实时检测前通过对标准样本的学习，建立标准数据字典，减少了检测过程中的计算量，缩短了检测计算时间，是实现实时高速在线检测的关键。第五，，设计搭建了模拟生产线的视觉检测实验平台，并通过调整实验平台的相关参数，测试在线视觉检测的性能，实现了在实验室中对高速视觉检测的仿真和测试。最后，研制了用于实际生产的皇冠瓶盖在线检测系统，并安装到生产现场进行实际的测试，实现了2600个/分钟的皇冠瓶盖在线检测，文中给出了现场测试结果。通过10个月的试运行，该系统达到了皇冠瓶盖高速在线检测的要求，验证了本文理论研究成果的可行性与有效性。本文研究成果不仅仅局限于皇冠瓶盖的在线检测，还可以扩展到其他领域产品的在线检测，市场前景广阔。
[Abstract]:Machine vision detection is a new detection technology based on computer vision and image processing. Through image processing, it can obtain all kinds of description parameters of the object being measured and understand and judge the parameters. It is finally applied to actual detection, measurement and control. It has non contact, high precision, wide range of application and self. Because the machine vision on-line detection equipment is installed on the production line, the detection speed must keep high synchronization with the high speed production line. The key technology to realize its synchronization is the precise control of camera photography and the high speed detection of the image. The rapid detection of images and the development and development of intelligent online detection equipment for independent intellectual property are of great significance to the theoretical exploration and innovation of high speed vision detection and the urgent needs of the current domestic market of intelligent manufacturing industry.
This paper focuses on the key technology to realize the on-line inspection of machine vision, and studies the high-speed and accurate control of machine vision and the high speed detection of the image. The feasible design scheme is given with the online detection of the crown cap. The main work is summarized as follows:
First, the structure of the machine vision on-line detection system based on the two layer network control theory is proposed, which provides a new research idea and direction for the realization of the high-speed machine vision on-line detection. The image processing task is placed in the high-level processing unit, and the system control is placed in the local control unit. The tasks of each unit are respectively processed according to the needs. The corresponding processor is used to solve the problem that the processor is overloaded when the signal is centralized. At the same time, each subsystem is designed by modularization, which is easy to install, debug, maintain and expand.
Secondly, a high speed and accurate control method based on iterative learning control and Calman filter is proposed, which realizes the accurate capture of the high speed moving workpiece image under the limited position of the camera, and solves the difficult problem of the high quality and precision acquisition of the workpiece image in the complex scene environment. The camera control model is combined with the Kalman filter, and the convergence and convergence range of the model are theoretically deduced and analyzed. Numerical simulation and practical experimental results are given.
Third, the image edge search strategy based on the local energy discrete path level set method is proposed. The narrow band level set search is reduced to the line search of the limited number of narrow bands in the narrow band, which greatly reduces the amount of data in the edge search. Error caused by uneven illumination.
Fourth, the rotation invariant feature of the circular region projection histogram is proposed, and the 2D matching data is converted into 1D data. The matching efficiency is improved and the precondition for high-speed detection is provided. At the same time, the strategy of using sparse representation to carry out the rotation matching and defect detection of the image is put forward. The strategy adopts the study of the standard sample before executing the real-time detection. The establishment of standard data dictionary reduces the amount of computation in the detection process and shortens the detection time. It is the key to achieve real-time high-speed online detection.
Fifth, the visual inspection experiment platform of the simulated production line is designed and built, and the performance of the on-line visual inspection is tested by adjusting the related parameters of the experimental platform, and the simulation and test of the high speed vision detection in the laboratory are realized.
Finally, the online inspection system for the crown bottle cap for actual production is developed, and the actual test is installed on the production site to realize the on-line test of the crown bottle cap of 2600 / minute. The test results are given in the paper. The system has reached the requirement of the high speed on-line detection of the crown bottle cap through the trial operation of 10 months. The feasibility and effectiveness of the theoretical research results.
The research results in this paper are not only limited to the on-line detection of crown caps, but also can be extended to other products for on-line detection.
【学位授予单位】：上海大学
【学位级别】：博士
【学位授予年份】：2014
【分类号】：TP391.41

【参考文献】