【作者】 谈理；

【导师】 刘谨；

【作者基本信息】 上海大学 ， 机械电子工程， 2008， 博士

【摘要】 在上海乃至全国许多城市的轨道交通建设的蓬勃发展中,作为隧道挖掘主要设备的盾构机起着越来越重要的作用。盾构机集机械、液压传动、自动控制、检测和计算机技术为一体。随着国民经济的高速发展和工业控制技术的不断更新,盾构机和其它工业现场机械一样,功能越来越强,设备越来越先进、复杂和昂贵。因而及时分析设备故障产生的原因或正确预报设备潜在的故障信息等,从而减少生产过程中因故障而停机的时间,这将会给生产和现代化建设带来明显的经济效益。所以,本课题通过对盾构机故障诊断智能系统的研究,探索在现代生产装备中如何融入人工智能技术,以提升现代生产装备的技术含量,保障现代生产装备的运行正常和增加现代生产装备的生产效益。由于盾构机是大型、复杂和连续运行的生产设备,具有多部件结构、多层次故障症状和故障的不确定性等特点,故本文确定了研制基于多Agent系统(Multiple Agent System,简称MAs)的故障诊断知识引擎系统的研究方向,研究了多个推理Agent协作的集成推理机制和多种知识支持Agent互补的集成知识库的构造和知识获取平台,突破了单一诊断智能系统在领域知识和推理能力方面的局限性。为了满足对盾构机故障的不确定性的模糊推理命题的推理需要和属于连续性生产设备的盾构机在故障诊断方面的实时性要求,在盾构机故障诊断知识引擎系统的推理Agent研制中,本文针对模糊性特点设计了基于综合动、静隶属度运算的模糊识别法;针对偶然性和耦合性特点设计了基于贝叶斯原理的概率统计识别法;针对如何解决盾构机所具有的大量的状态参数和控制信号等的知识获取的“瓶颈”问题以及由此而来的快速推理问题,设计了基于神经网络的并行运算方法和建立了基于遗传神经网络的机器学习平台。本文还提出了基于MAS的盾构机故障诊断知识引擎系统的结构形式为多级联盟式,研制了多个具有不同机理的子Agent、MAS的通信模块和规划统筹器等,使该知识引擎系统能按设定的通信和规划机制将若干个所需的子Agent组织成Agent联盟,驱动多个子Agent根据总任务分解和分配而分别承担各自一部分智能任务,综合各部分任务就完成了单个Agent所不能完成的对某一时段作为大型复杂设备的盾构机运行状态的判定或故障诊断的复杂的综合性的任务,以适应多部件结构、多层次故障症状、多故障特点和多型号盾构机的故障诊断的需要。论文在研究了上述一系列关键性问题的基础上,开发了一套盾构机故障诊断知识引擎系统,该系统具备了交互诊断功能、自动诊断功能和在线诊断功能,此外还具备了诊断仿真功能、查询功能和优化保养功能,以及知识库维护和更新功能,特别是通过知识重用与知识重构,能够迅速地为新增型号的盾构机在知识库中增加相应的内容,以支持推理Agent完成诊断工作。在上海某些隧道掘进施工中,盾构机故障诊断知识引擎系统经过了实际的使用,并进行了修改和完善,通过多处隧道掘进施工工程的实际检验,取得了良好的效果,得到了上海隧道工程股份有限公司盾构工程分公司和项目验收会从事盾构挖掘、机电设备、控制工程和计算机技术等领域的与会专家的好评。最后,本文提出了课题研究的结论和展望,从优化的角度论述了故障诊断知识引擎系统在现代生产装备中的应用的进一步研究的内容和所需完善的方面。更多还原

【Abstract】 In the sudden boom of urban rail transportation in Shanghai and in many other metropolises of China, the shield has been playing an increasingly important role as the main apparatus for digging tunnels. The shield, to all intents and purposes, is an amalgam of new technologies in such fields as mechanics, hydraulic transmission, automatic control, testing and computer sciences. As China is ameliorating itself rapidly as an economy and updating its technologies for industrial control on a continual basis, the shield, like many other articles of field industrial equipment, has been endowed with more and more functions - accordingly, it is getting more and more advanced, complex and high-priced. Therefore, timely analyses of existent faults and accurate predictions of potential ones will temporally shorten break-downs and thus bring tangible profits not only to the process of production but also to the construction of China’s modernism in general. In this context, the present author has set his research goal at discovering a solution to the integration of artificial intelligence into modernized production equipment, which is intended to be better technology-enabled, to operate more smoothly and to enjoy higher efficiency.Since the shield is a large complex continuously-operating mechanical device that features a multi-component structure, multi-level fault symptoms, an uncertainty of faults, etc., this paper chooses the knowledge engine system for fault diagnosis based on MAS to be its point of departure, from which to study the integrated reasoning mechanism of multiple reasoning agents and the structure of the integrated knowledge base and the knowledge-acquisition platform, both based upon the complementation of agents supported by multiple types of knowledge, with an eye to realizing a major break-through in the limited range for the specialized knowledge and reasoning ability of the single intelligent system for diagnosis.In order to meet the needs of the fuzzy reasoning proposition relevant to the uncertainty of shield faults and of the real-time fault diagnosis of the shield as an article of continuously-operating production equipment, the paper, in developing the reasoning agent of the knowledge engine system, presents its own designs of a fuzzy recognition method based upon the comprehensive static and dynamic membership computing to deal with the flizziness aspect of the object of research, of a probability and statistics method of recognition based upon the Bayesian principle to deal with the aspects of possibility and coupling, of a parallel computing method based upon neural networks and a machine learning platform based upon genetic neural networks to try to solve the "bottleneck" problem caused by the insufficient knowledge of a large number of state parameters and control signals of the shield and the problem of quick reasoning caused by this very "bottleneck."This paper also puts forward a proposition that the structural pattern of the knowledge engine system should be a multi-level alliance: after sub-agents and MAS communications modules and planners with different properties are developed, necessary agents can be organized into an alliance in accordance with the pre-fixed communications and planning mechanism, so that several sub-agents can each of them undertake some part of the intelligent task, as required by the scheme of the general task, which is capable of being divided and distributed, and that the integration of the completed sub-tasks is actually something beyond any single agent: to ascertain the operational state of or to diagnose the fault in the shield as an article of large complex equipment at a given moment or during a given span of time. Besides, in this process are met the needs for fault diagnoses that are relevant with the multi-component structure, the multi-level fault symptoms, the high fault frequency and the type variety of the shield.After conducting an in-depth study of the above-mentioned key topics, this paper goes on to create the knowledge engine system for the diagnosis of shied faults that has been developed for this research program. The Knowledge Engine System for the Diagnosis of Shield Faults, developed for different diagnosis purposes, boasts the functions of interactive, automatic and on-line diagnosis. Moreover, the flexible cooperation between the three functions, established by the communications and planning mechanism of the MAS, is reasonably expected to diagnose real faults with success. The system may also perform the functions not only of diagnosis simulation, information acquisition and optimized maintenance, but also of the sustainment and renewal of the knowledge base - in particular, through the re-use and re-construction of knowledge, the system can rapidly include information concerning the latest types of shields into its knowledge base and assist the reasoning agent in completing the diagnosis.The Knowledge Engine System for the Diagnosis of Shield Faults has got modified and improved in its constant trial operation. Applied and tested in several tunnel diggings, the System has received profuse praises from the Shield Engineering Company of the Shanghai Tunnel Engineering Co., Ltd. and from the experts in the fields of shield tunneling, mechanic and electrical equipment, control engineering and computer technology who attended the assessment conference for the Shield program.Last but not least, this paper outlines the future for further investigations into the research topic, discussing how the application of the knowledge engine system in the modernized production equipment can be further studied and improved from optimization.更多还原