【作者】 林吉良；

【导师】 蒋静坪；

【作者基本信息】 浙江大学 ， 控制理论与控制工程， 2009， 博士

【摘要】 随着计算机技术、电子技术、控制理论、机械工程等学科的发展和新材料、新元件的应用,移动机器人技术日新月异。移动机器人已被广泛应用在工业、农业、军事、太空探索、深海探索、医疗、救援和日常生活等各个方面。现代工农业、军事等领域日趋复杂以及人们对生产生活质量的不断追求,使得对移动机器人要求越来越高,机器人技术也越来越先进、复杂。从简单应用到智能化、从人工参预到自组织运行、从已知环境到未知环境应用、从单机器人操作到多机器人协调工作和编队工作等等,这些应用都对移动机器人的稳定性提出了一定要求。如果移动机器人在带故障状态下运行,一方面会使移动机器人的寿命缩短,另一方面还可能带来不利影响,有时可能产生灾难性的后果。实际工程运行表明,移动机器人容易出现故障。即使花费大量人力物力精心设计和制造的移动机器人,在面对未知的、复杂的应用环境时,也经常会出现故障。特别是应用在太空、深海或危险环境中,移动机器人出现故障时人们无法直接维修或维修代价太高,因此移动机器人故障诊断与容错控制技术的研究具有重要意义。本文在国内外学者对移动机器人故障诊断研究的基础上,结合移动机器人应用发展的趋势,引用了相关领域的新技术,对未知环境下的移动机器人故障诊断、移动机器人并发故障诊断、编队运行的移动机器人故障诊断与容错控制、群集(flocking)中的移动机器人故障诊断与容错控制进行了研究。本文的主要研究成果如下：1、分析了移动机器人故障的分类、产生原因,故障诊断的意义、研究现状以及各种移动机器人的诊断方法、特点、需要解决的问题和研究趋势；2、提出了一种基于支持向量机(SVM)的未知环境下移动机器人故障诊断的方法。支持向量机故障诊断的关键是提取特征向量、支持向量机参数选取、降低噪声对支持向量机的影响以及剔除孤立点等。针对支持向量机对噪声敏感的特点,提出了小波变换的方法重构采样信号并提取特征,用网格搜索与交叉验证的方法优化支持向量机的参数,采用投票方式的多支持向量机对故障特征进行分类。提高了故障诊断的适应性和分类正确率。3、针对移动机器人可能同时并发多种故障的特点,提出了基于模糊核聚类(KFCM)的移动机器人并发故障诊断技术。采用针对单发故障的卡尔曼滤波器对采样信号进行滤波,利用整合先验知识的模糊核聚类方法对滤波残差信号进行模糊分类,利用数据点对各聚类中心的模糊隶属度,诊断机器人是否发生了某一类故障,或者某两类并发故障。在“先锋3”号移动机器人上对12种单故障与并发故障进行了诊断并与FCM方法进行了对比,结果证明了该方法比FCM具有更好的诊断效果。4、研究了移动机器人在SBC、SSC跟随规则的编队方式运行时的故障诊断与容错控制。根据SBC跟随规则和SSC跟随规则,设计了相应的分布式扩展卡尔曼滤波器对信号进行滤波,根据滤波残差进行故障诊断。提出了编队控制时的容错控制策略和避障算法。弥补了Diagle等人算法中机器人出现故障或遭遇障碍物时无法维持编队的缺陷。分析了大量移动机器人在编队控制时可能组成的复杂网络上故障传播模型,提出了基于复杂网络传播模型的目标免疫算法,能降低故障在大规模编队网络中传播的概率。5、以α-网格模型的群集为研究对象,采用几个群集相关性能指标,研究群集中移动机器人出现故障时对群集性能的影响。提出了基于数据通信和数据关联的群集故障诊断和容错控制算法。将故障机器人看成障碍物,提出了运用栅格地图记录信息和和采用各类agent模拟受力关系的方法对复杂形状障碍物的避障算法。不仅克服了Olfati-Saber算法总对凹多边形、长墙状的障碍物无法避障的缺点,而且该算法无需事先知道机器人运行环境的先验知识,具有一定的优越性。更多还原

【Abstract】 With the development of computer technology, electronic technology, control theory, mechanical engineering and the applications of new material, new components, mobile robot technology changes quickly and continuously. Mobile robots are widely used in industry, agriculture, military, space exploration, deep sea exploration, medical care, rescue, daily life, etc. Modern applications are more and more complicated and people are in pursuit of life quality, which lead to more and more requirement and complication of mobile robot technology. From simple application to intelligent ones, from man intervened mode to autonoums mode, from simple environment to unknown environment, from single robot to multi-robots work correspondly and in formation, the reliability of mobile robots are required in all these applications. If mobile robots are operated with fault, the lifetime of them will be decreased, as well as they may bring adverse impact, sometimes disaster consequences. Unfortunately, studies show that mobile robots are often in fault states when they are applied in complex unknown environments especially in space, deep sea and dangerous environments, though they are well designed and manufactured. So studies on fault diagnosis and fault tolerant control of mobile robots are extremely important because they are unreachable from human kind or too expensive.According to the trend of mobile robot applications, this thesis studies fault diagnosis of mobile robot in unknown environment, simultaneous diagnosis of mobile robot, fault diagnosis and fault tolerance control of mobile robot in formation, fault diagnosis and fault tolerance control of mobile robot in flocking based on the research works from scholars all over the world, and with new technique in relative disciplines adopted. The major work and result are represented as follows:1、Failure classification, main reasons, significance, research status, fault diagnosis methods and their characteristics for mobile robot are surveyed. The problems to be resolved for fault diagnosis of mobile robot and the research trend are studied. 2、A new method for fault diagnosis of mobile robots in unknown environment based on Support Vector Machines(SVMs) is proposed. The key points of fault diagnosis method based on SVM are feature extraction, parameter selection, decreasing noise and rejecting outliers. According to noise sensitivity characteristics of SVM, reconstruction of sampling signals and feature extraction by wavelet method are adopted. Grid searching method and cross validation are introduced to optimize the parameters of SVMs. Fault features are classified by multiple SVMs based on voting system. The adaptability and correct rate for classification is increased by adopting the methods above.3、According to the characteristics of multiple faults may occur simultaneous on a mobile robot, simultaneous fault diagnosis technique based on Kernel Fuzzy Clustering Method is proposed. According to the kinematic model of a mobile robot, a specific Kalman Filter (KF) is designed for each single fault state to filter the fault data of the mobile robot. Residuals of the KFs are classified by Kernel Fuzzy Clustering Method (KFCM) with prior knowledge. Simultaneous faults are diagnosed whether one or two faults occurred according to the fuzzy membership to each single fault set. Simulation has been implemented on a 3-wheeled mobile robot named Pioneer 3 to diagnose 12 common single faults and simultaneous multi-faults, and compare the result with FCM, which shows KFCM for simultaneous fault diagnosis technique is better than FCM.4、Fault diagnosis and tolerant control method is studied when mobile robots are walking in SBC and SSC formation modes. Correlative distributed extended Kalman filters are designed according to SBC and SSC following laws. Faults are diagnosed according to the filtering residuals. Fault tolerant control strategy and obstacle avoidance algorithm is proposed remedy the defects of Diagle’s algorithm in which formation is not able to maintain when faults occur or obstacles encounter. Fault spreading model on complex network composed by mass mobile robots in formation control is studied and targeted vaccination algorithm based on spread model on complex network is proposed to decrease the faults spreading probability in formation. 5、Takingα-lattice flocking as research object, the influence and fault tolerance control algorithm when faults occurs in flock are studied. The impact to flocking performance is analyzed by means of flocking property indexes when faults occur. A flocking fault diagnosis method and fault tolerance control strategy based on communication and data association are introduced. Considering failure mobile robots as obstacles, an obstacle avoidance algorithm against complex shaped obstacles is proposed which based on recording the information by grid maps and simulating forces by 4 kinds of agents. The algorithm is well performaned that overcomes the shortage of Olfati-Saber’s algorithm that is not able to avoid concave shaped obstacles and long walls without prior knowledge.更多还原