【作者】 蔡天佑；

【导师】 田梦倩；

【作者基本信息】 东南大学 ， 机械工程， 2024， 硕士

【摘要】 在电力、通信、物流等领域,大量关键的基础设施功能都承载在封闭的工作空间中,其中,电缆沟道就是一个很典型的例子,它为城市电力系统提供了安全、有序的运行环境,从而在现代社会中发挥着不可替代的重要作用。准确可靠的电缆沟道巡检对于保障电力系统安全稳定运行至关重要。考虑到人工巡检存在效率低、有较大安全风险等缺陷,使用智能巡检机器人来进行巡检成为了现今的重点研究方向。本文针对电缆沟道巡检机器人项目,对于机器人通信模块、建图定位导航模块以及目标识别模块进行了设计,使得该机器人能够实现电缆沟智能巡检的任务目标。首先,从电缆沟道的实际环境出发,对巡检机器人的各功能模块进行了设计。采用通过网桥中继的WiFi通信方案,构建上位机与机器人在封闭空间内的通信架构;采用以激光雷达为主、里程计为辅的建图定位方案,实现点对点导航;采用以激光雷达为主,单目相机为辅的识别方案,构建目标识别架构以及采用ROS系统,用Qt实现软件设计,实现机器人的远程控制。其次,给出了机器人自主定位建图导航的框架。在LOAM算法的基础上,实现机器人的建图功能;通过HDL＿localization算法,构建代价地图,实现机器人的定位功能;配置movebase模块,实现点到点的全局路径规划以及局部的实时避障功能。然后,确定机器人的目标识别方案。通过内外参标定的形式,获取相机到激光雷达坐标系的坐标变换矩阵。采用DBSCAN算法,实现激光雷达点云聚类,调节参数优化聚类效果。在获得点云侦测框以后,根据已有的坐标变换矩阵和点云3D侦测框,绘制2D目标侦测框,辅助人眼实现目标识别的功能。最后,确定了上位机双线程,双子界面的设计框架,完成Qt上位机软件设计。并在室内、模拟电缆沟环境以及实际电缆隧道环境对于本文提出的机器人整机系统进行测试。实验结果表明,本文提出的通信模块可适应电缆沟道实际工作环境,建图定位导航模块可实现机器人的地图构建以及基于地图的定位和路径规划,目标识别模块可有效识别环境中可能存在的各种尺寸的目标,上位机控制软件可以实现机器人的远程控制。更多还原

【Abstract】 In fields such as electricity,telecommunications,and logistics,numerous critical infrastructure functions are housed within enclosed workspaces.Cable trenches,serving as a quintessential example,provide a secure and organized operational environment for urban electrical systems,playing an indispensable role in modern society.Accurate and reliable inspection of cable trenches is pivotal for ensuring the safe and stable operation of electrical systems.Given the limitations of manual inspection,such as inefficiency and significant safety hazards,the emphasis is now placed on employing intelligent inspection robots.This paper addresses the design of various functional modules for an inspection robot tasked with cable trench inspection.To begin with,based on the actual environment of cable trenches,diverse functional modules of the inspection robot are devised.A WiFi communication scheme utilizing bridge relay is adopted to establish communication architecture between the upper computer and the robot within enclosed spaces.A mapping and localization scheme,primarily leveraging Li DAR with odometer supplementation,is employed to achieve point-to-point navigation.An identification scheme,mainly utilizing Li DAR and supplemented by a monocular camera,is employed to construct the target recognition architecture.ROS system is utilized,and software design is implemented with Qt to enable remote control of the robot.Subsequently,the framework for autonomous mapping,localization,and navigation of the robot is outlined.Building upon the LOAM algorithm,the robot’s mapping function is implemented.Through the HDL＿localization algorithm,a cost map is constructed to achieve the robot’s localization function.The movebase module is configured to facilitate global path planning from point to point and real-time obstacle avoidance at the local level.Next,the target recognition scheme for the robot is determined.By acquiring the coordinate transformation matrix from the camera to the Li DAR coordinate system through intrinsic and extrinsic parameter calibration,Li DAR point cloud clustering is accomplished using the DBSCAN algorithm,with parameter adjustments made to optimize clustering effectiveness.Following the acquisition of the point cloud detection box,2D target detection boxes are drawn based on the existing coordinate transformation matrix and the point cloud 3D detection box to aid human-eye target recognition.Finally,the design framework for the upper computer with dual threads and dual sub-interfaces is established,and Qt upper computer software design is completed.The overall system of the robot is tested indoors,in simulated cable trench environments,and in actual cable tunnel environments.Experimental results demonstrate that the proposed communication module can adapt to the actual working environment of cable tunnels,the mapping and localization navigation module can achieve map construction,localization,and path planning based on the map,the target recognition module can effectively identify various-sized targets in the environment,and the upper computer control software can successfully achieve remote control of the robot.更多还原