【作者】 曹文明；

【导师】 王耀南；

【作者基本信息】 湖南大学 ， 控制科学与工程， 2014， 博士

【摘要】 输电线路覆冰和积雪常会引起线路的跳闸、断线、倒杆、绝缘子闪络和通信中断等事故。世界各国都曾因输电线路覆冰引发安全事故,给各国带来了巨大的经济损失。而传统除冰方法效率低下而且安全性不高,因此研究新型的除冰方法替代人工除冰就变得十分迫切。除冰机器人是一种实现自动在线除冰的新装备,得到了研究人员和电力公司的广泛关注。但是它的运行环境非常复杂,需要解决许多关键技术难题,尤其在机器人的自主越障机构、传感器与控制等方面,是制约除冰机器人研究进展的主要因素。本文以除冰机器人的在线行走与越障为应用背景,研究利用视觉传感器为主要传感器的视觉控制方法。基于视觉的机器人控制是通过对视觉信息的分析与处理来感知环境,并利用视觉信息引导和控制机器人完成给定的任务。因此本文对除冰机器人的视觉控制研究主要包括两个方面：(1)除冰机器人通过对在线拍摄图像的分析处理,实现对工作环境的感知和识别；(2)利用相机反馈图像信息引导和控制机器人完成在线行走和越障动作。内容涉及机器人技术、图像处理技术、目标识别与空间定位技术、图像视觉伺服技术等。除冰机器人利用视觉信息可实现对环境的识别和运动的伺服控制,为了实现这一目标,本文在以下几个方面进行了研究：1、在借鉴国内外巡线机器人研究经验的基础上,提出了两臂式和三臂式除冰机器人本体设计方案。考虑到除冰机器人多手臂爬行机构的复杂性,利用旋量理论简化运动学分析,成功建立了机器人手臂的正、逆向运动学模型,为机器人在线行走与越障动作的控制提供了基础。2、除冰机器人工作环境复杂,其中安装在输电线路上的防震锤、悬垂线夹、耐张线夹等线路附件将是机器人在线行走时的障碍,而冰机器人要实现在线自主行走和越障,就必须能识别与定位前方线路上的各种障碍。在对大量实际图像观察后,提出利用障碍物图像局部特征进行障碍物目标识别与定位。首先,收集机器人在线行走时拍摄的各种障碍物样本图像,然后提取障碍物图像区域的SURF特征构造障碍物SURF (Speeded-Up Robust Features)特征模板库。在实际应用中,将在线拍摄实时图像的SURF特征与模板图像特征匹配,若达到匹配条件则认为匹配成功,即认为当前图像中存在与模板图像同类的障碍物。初步匹配成功后,选取4对以上的匹配点计算模板图像与实时图像平面间的单应性矩阵,再用单应性矩阵将模板图像中离相机最近的点(事先设置)映射到当前实时图像中,把该点坐标代入单目测距计算式得出机器人与障碍物之间的距离。机器人在了解前方障碍的类型和距离信息后就可实现在线行走的导航控制。3、基于障碍物外部形状特征的识别、定位方法。由于不同障碍物的外形和轮廓差别很大,可利用障碍物图像外形轮廓特征来识别它们。首先,对机器人实时采集图像进行预处理、最佳阈值分割、小波模提取轮廓边缘。然后利用具有旋转、平移、缩放不变性的小波矩算法计算障碍物轮廓图像的小波矩特征向量,把特征向量输入SVM神经网络实现对障碍物图像的识别判断。在定位阶段采用霍夫变换和结构约束条件对边缘图像中的直线、圆、椭圆等几何基元进行定位,然后把几何基元图像的形心坐标代入单目测距算式可估计出机器人与障碍物的距离。以上识别与定位信息为机器人在线行走与导航提供了条件。4、在分析除冰机器人环境特点和越障机理的基础上,提出了基于图像的越障视觉伺服控制方案。首先,选取具有全局性、通用性、抗干扰性能好的图像矩特征作为反馈图像的伺服特征,而小波神经网络具有较强的学习和泛化能力,将两者结合起来设计伺服控制器。经过训练后的神经网络将具备伺服控制能力,在除冰机器人执行越障动作时,神经网络将反馈图像特征与期望特征的误差直接映射为手臂关节控制量,实现机器人越障动作的伺服控制,避免了传统视觉伺服控制中的相机标定和图像雅可比逆矩阵的求解,大大减少了计算量,提高了图像视觉伺服的响应速度。5、在应用本文以上研究技术的基础上,研制了三臂式除冰机器人样机。分析了除冰机器人研制的难点与关键技术,并从工程应用角度,重点介绍了除冰机器人本体的机械结构和设计方法、电机与控制系统的设备构成。在整机装配完成后,分别对各分部进行了测试和整体调试,最后给出了除冰机器人上线行走和除冰的实验情况。文章结尾部分,总结了全文的主要工作和创新性研究成果,并对下一步研究工作进行了展望。更多还原

【Abstract】 The transmission power line icing can cause various electrical accidents which will bring huge losses. These accidents include circuit breaker, line breaking, tower falling down, insulator flashover, communication interruption and so on. The traditional de-icing methods are not only inefficient but also unsafe. Hence, it becomes more urgent to develop new effective de-icing methods.Nowadays, more and more researchers pay attentions to a new de-icing equipment called as De-icing robot, which can realize online clearing and de-icing operation. However, as complex operating environment, there are many key technical problems to be resolved for de-icing robot researching, such as obstacle-surmounting mechanism design, sensor and control system development, and so on. The processing and perception of the environment can be achieved by visual information analysis for Vision-based robot control. The visual information can be used further to guide and control robot accomplishing the assigned tasks. To solve the on-line running and obstacle-surmounting problems, a visual control method based on vision sensor is presented in this thesis. The main research is summarized as follow:(1) Online visual images processing and analysis.(2) Guide and control de-icing robot online running and obstacle-surmounting by using feedback image information. The contents involve robotology, image processing, target recognition and space positioning technology, visual servo technology, and so on.This dissertation focuses on the methods for solving visual image servo control of de-icing robot. Main results and contributions of this dissertation are as follows:1. Based on the research experiences of the inspection robots, we design two-arm and tree-arm deicing robot mechanisms. Considering the complexity of crawling mechanism, the screw theory is used to simplify the kinematics analysis. Consequently, the forward and inverse kinematics models of robot arms are established successfully. The realizations of de-icing robot controls are on the basis of the above results.2. As there are several obstacles on the power line, such as counterweight, strain clamp, suspension clamp, the work environment of de-icing robot is very complex. Therefore, robot needs to be able to identify and locate various obstacles in front of transmission lines. By observing large number of actual images taken by camera of de-icing robot, we use the local characteristics of images to distinguish and locate the obstacles target. First, collect the sample photographs of obstacles and extract its SURF (Speeded-Up Robust Features), meanwhile build the template library of obstacles images’SURF feature. In practical applications, through matching the SURF characteristics of real-time and the template images, robot can accurately determine the image recognition results based on matching condition. If the match is successful, we think the obstacle in the current real-time image is the same as the template obstacle. Following, select more than four match points between the template image and real-time image to calculate homography matrix. Furthermore, we can estimate the distance between robot and obstacle after putting closest point coordinates into formula of distance measuring by monocular vision. Finally, robot can realize autonomous online navigation control when it determines obstacles information above.3. A recognition and location method based on external shape feature of obstacles is developed. As the external shape and contour are great different for different obstacles, robot can identify the obstacles by using its contour features. After pretreatment, optimal threshold segmentation, wavelet modulus image contour extraction for real-time image that captured by camera, the wavelet moment feature vector can be constructed after calculated the obstacle contour image’s wavelet moment. Then the SVM neural network can identify obstacles after putting image wavelet moment feature vectors input to the SVM neural network. In the process of target location, the straight line, circle, ellipse can be detected by Hough transform and the condition structure constraints in obstacle contour images. The distance between robot and obstacle also can be determined after putting the center point coordinates into formula of distance measuring. Then, the de-icing robot can realize autonomous navigation control meanwhile knowing the category and distance information.4. A visual servo control scheme is developed on the basis of analyzing the work environment characteristic and surmounting theory of de-icing robot. Firstly, since the moment feature have many superiority in global, general, it is set as the servo characteristics. Secondly, the wavelet neural network has strong learning and generalization abilities, a high performance visual servo controller can be designed by combining the two respect’s advantage. Then the trained ANN will have servo control capability. At the stage of obstacle striding movement, the visual servo controller direct mapping the error of feedback image characteristics and desired features for arm joint controller, and realize visual servo control for robot motion. This can avoid camera calibration and inverse Jocobian matrix of the traditional visual servo control. As a result, it can greatly reduce the amount of calculation and improve the response speed of the image visual servo.5. Based on above researches, a tree-arm de-icing prototype is developed. Meanwhile the difficulties and key techniques are summarized. Furthermore, the mechanical structure and design method of de-icing robot, motor device and control system are reported. Following, each division was tested respectively and the whole machine trial test. Finally, the online walking and de-icing experiments of de-icing robot are elaborated in the thesis.In this thesis ending part, the main innovations of the thesis are summarized, and prospected the next research work.更多还原