【作者】 王晓垠；

【导师】 曹效英； 秦文虎；

【作者基本信息】 东南大学 ， 测试计量技术及仪器， 2005， 硕士

【摘要】 故障诊断是指在一定工作环境下查明导致系统某种功能失调的原因或性质,判断劣化状态发生的部位或部件以及预测状态劣化的发展趋势等。故障诊断技术是40年来发展起来的一门新学科,它具有很强的工程背景和重要的实用价值。近年来,为了满足复杂系统的诊断要求,随着计算机及人工智能的发展,诊断技术进入以知识处理为核心,信号处理、建模处理与知识处理相结合的智能诊断技术阶段。基于理论上较成熟的专家系统和BP神经网络,本文对专家系统和BP神经网络在液压设备故障诊断方面的应用进行了有益的探讨,提出了以专家系统为主,采用BP神经网络作为有益补充的智能故障诊断系统的基本结构。系统根据接收到的液体加注系统的运行参数与知识库中预先存储好的故障征兆表搜索出已发生的故障征兆,接着根据专家系统知识库中的规则表搜索出所有可能的故障原因,然后逐一验证所有可能的故障原因中其它征兆是否成立,直到最终推理出诊断结果。如果在专家系统知识库中搜索不到完全匹配的故障,则调用神经网络知识库进行推理,得出诊断结果。在专家系统部分,进行了知识库的设计,提出了综合采用面向对象方法和产生式规则表示法的知识表示方法,并采用数据库技术建立和维护知识库;在推理机设计中,发挥专家系统良好的人机交互功能,采用正反双向混合推理方法进行推理;在解释程序的设计中,采用预置文本与路径跟踪法,可以在用户需要时给出对推理结果和过程的详细解释。在神经网络部分,通过对液体加注系统的分析,建立了BP神经网络的基本结构,并提取专家系统知识库中的规则作为样本集和测试集,经过MALTAB的仿真训练和测试,得到最终的正确率较高的权值和阈值系数集合作为神经网络的知识库,在进行故障诊断时可由已建好的神经网络结构进行正向计算,推理出最终结果。在系统具体实现中,采用Visual C++ 6.0进行人机界面和整体控制程序的设计,采用ACCESS 2000进行知识库的建立、维护和其它数据文件的管理,利用Matlab进行神经网络部分的训练、仿真和计算。完成整个系统设计后,我们利用MCGS组态软件建立了燃料加注系统的二维虚拟模型,以模拟实际液体加注系统的运行对智能故障诊断系统进行测试,使我们能更加直观地监测智能故障系统的运行状态和故障诊断结果。经过测试实验,证明本系统达到实际应用要求,具有较高的准确率。本文提供了一种用于液压设备故障诊断的通用方案,对于类似系统的设计开发具有借鉴意义。本系统的投入使用可以提高实时故障诊断的准确性和速度,节约人力、物力,为防止重大生产事故的发生起到积极的作用。更多还原

【Abstract】 Fault diagnosis is finding out the reasons that caused the system maladjusted and judging the places and components where the faults happened. The technology of fault diagnosis is a new subject, which has developed for more than forty years. It has very important practical value. In recent years, in order to deal with the more and more complicated systems, with the development of computer and artificial intelligence, the technology of fault diagnosis comes into another phase——intelligent fault diagnosis. It regards knowledge management as the kernel and combines signal processing and modeling processing.Based on the expert system and BP neural network, the application of expert system and BP neural network in the fault diagnosis of hydraulic pressure system and the design of intelligent fault diagnosis system using expert system and BP neural network are introduced in this paper. According to the obtained operational parameter from the hydraulic pressure system and the fault symptom table in knowledge base, the system searches every generating fault symptom. Based on the rule table in expert system it can find all the possible fault causes. Verifying each possible fault cause and comparing the other fault symptom in it, thereby it infers the diagnosis result. The result also obtains from calling neural network knowledge base when it can not search the right match fault cause in the expert system knowledge base.In the expert system, a method which combines the technology of objected-oriented and the method of if-then rule to present knowledge and a construction of knowledge base based on relational database are introduced. And then we use data base technology to built and maintain knowledge base in the part of expert system. In the design of inference machine, display the good man-machine interactive function, using backward and forward ratiocination to infer. In the design of interpretive program, employ preset text and path tagging method to get inference result and explanation of process in detail when users demand. In the neural network, analyzing the hydraulic pressure system to build a basic structure of BP neural network and extract rules of expert system knowledge base as the set of sample and test. After emulations of training and testing in MALTAB, the better valid coefficient set of significance value and threshold value can be adopted as knowledge base of neural network. In the fault diagnosis, proceeds positive-going calculation by using the created neural network structure to infer the final result.After fulfilling all the design of the system, we utilize MCGS configuration software to build a two-dimension dummy model of propellant-servicing system. The emulation of the real operation of hydraulic pressure system is to test the intelligent fault diagnosis system. From the result of experiment, the system is approved to agree the application requirement with better accuracy.This system uses VC++ 6.0 to build the human-machine interface and the whole controlling program, ACCESS 2000 to design the knowledge base and Matlab to complete the design、training and emulation of artificial neural network.The paper distributes a general precept of fault diagnosis for hydraulic pressure system, which is of great significance for designing other similar system. After applying the real time fault diagnosis of fuel-adding system, improves the veracity and speed of faultdiagnosis, saving a large of manpower and material resources and have an effect on safe production.更多还原