【作者】 向彦彦；

【导师】 袁雪；

【作者基本信息】 北京交通大学 ， 交通信息工程及控制， 2022， 硕士

【摘要】 近年来,在环保和能源的双重压力下,新能源汽车行业得到飞速发展。其中,纯电动汽车以及混合动力电动汽车的市场占有率较高,各厂商便大规模生产充电桩以提高新能源汽车的续航能力。但调查显示,投入使用的充电桩存在有桩没电、屏幕界面显示错误等现象,使用率不高。因此,在充电桩准备出厂前需要进行完整的功能测试,在投入使用后,还要定期对其进行检修。现阶段,充电桩功能检测方式仍然是以人工为主。由工人模拟客户,对充电桩的人机界面进行点击,完成充电流程,并检查在充电过程中有无异常。人工操作的方式效率低,存在主观误差,还会因为充电桩处于未安装状态而发生触电危险。在充电桩投入使用后,又缺乏定期检修的设备和人员,工人到固定地点对充电桩功能排查检修需耗费较长时间,成本高而且不便捷。因此,迫切需要研发一种充电桩智能检测系统。充电桩智能检测系统包含多项功能测试项目。其中,充电桩的液晶触摸屏的质量是否达标,界面是否正确显示,是否能正确切换是重要的检测项目,也是其他测试端子的应用基础。针对这个工业需求,本文提出了一种基于计算机视觉的汽车充电桩界面功能智能检测系统。该系统由界面感知模块,机械臂控制模块和充电桩管理系统云平台三部分构成。论文主要的工作有以下几方面:(1)本文分析了现阶段充电桩界面功能检测系统的研究背景及意义,发展现状及问题,提出一种基于计算机视觉的汽车充电桩界面功能智能检测系统,总结此系统涉及到的相关技术及理论基础;(2)在界面感知模块中,本文提出了一种同时实现充电桩界面定位和界面文字检测的多任务网络,还提出了一种多模型的文字识别方法来实现长序列,中文序列,数字和英文混合序列的识别。在关键信息提取部分,建立了充电桩界面模板数据库,通过模板匹配和文字识别结果分析的方法实现关键信息的提取。本文给出了界面感知模块各部分的实现方法,网络训练过程,消融实验结果分析以及与其他网络的对比实验结果分析。实验结果证明,本文提出的方法显著提高了系统的准确率和实时性;(3)本文提出整个充电桩界面功能智能检测系统的结构组成和功能实现方法,并在国家电网质量检测车间对实际设备进行检测,证明了此系统可以代替人工高效地完成充电桩界面功能检测任务。图78幅,表17个,参考文献62篇。更多还原

【Abstract】 In recent years,under the double pressure of environmental protection and energy,the new energy vehicle industry has developed rapidly.Among them,the market share of pure electric vehicles and hybrid electric vehicles is higher.Manufacturers will mass produce charging piles to improve the endurance of new energy vehicles.However,the investigation shows that there are some problems such as power failure and screen display error,so the utilization rate of charging piles is not high.Therefore,a complete functional test is required before the charging pile is ready to leave the factory,and regular maintenance is required after it is put into use.At present,the charging pile function detection method is still mainly manual.The worker simulates the customer and clicks on the man-machine interface of the charging pile to complete the charging process and check whether there are any abnormalities in the charging process.Such manual operation has low efficiency,subjective error,and electric shock risk because the charging pile is not installed.After the charging pile is put into use,there is a lack of equipment and personnel for regular maintenance.It takes a long time for workers to check and repair the charging pile function at a fixed place,which is costly and inconvenient.Therefore,it is urgent to develop an intelligent detection system for charging piles.The intelligent detection system of charging pile includes a number of functional test items.Among them,whether the quality of the LCD touch screen of the charging pile reaches the standard,whether the interface is displayed correctly and whether it can be switched correctly are important testing items,and also the application basis of other test terminals.Aiming at this industrial demand,this paper proposes an intelligent detection system based on computer vision for automobile charging pile interface function.The system consists of interface perception module,manipulator control module and charging pile management system cloud platform.The main work of the thesis has the following aspects:(1)This paper analyzes the research background,significance,development status and problems of charging pile interface function detection system at present.An intelligent detection system of interface function of automobile charging pile based on computer vision is proposed.The related technologies and theoretical basis of the system are summarized;(2)In the interface perception module,this paper proposes a multi-task network that simultaneously realizes the charging pile interface positioning and interface text detection,and also proposes a multi-model text recognition method to realize the recognition of long sequence,Chinese sequence,digital and English mixed sequence.In the key information extraction part,the charging pile interface template database is established,and the key information is extracted by template matching and text recognition result matching.This paper presents the realization method of each part of the interface sensing module,the training process of the network,the analysis of the ablation experiment results and the comparison with other networks.Experimental results show that the proposed method significantly improves the accuracy and real-time performance of the system;(3)In this paper,the structure and function realization method of the whole charging pile interface function intelligent testing system are proposed,and the actual equipment is tested in the state Grid quality testing workshop,which proves that this system can replace manual to efficiently complete the charging pile interface function testing task.There are 78 figures,17 tables and 62 references.更多还原