【作者】 闫小良；

【导师】 田玉玲；

【作者基本信息】 太原理工大学 ， 计算机软件与理论， 2008， 硕士

【摘要】 故障诊断是通过从故障征兆空间到故障空间的映射,实现对故障的识别和诊断。然而,复杂系统经常处在动态变化的过程中,其行为特点不好把握,各种故障的发生具有很强的不确定性,所有这些都为有效地获取、表示和利用诊断知识进行智能化推理带来了很大的困难。独特型网络理论指出:免疫网络模型中的各个细胞、分子不是处在一种孤立的状态,而是通过自我识别、相互刺激和相互制约构成了一个动态平衡的网络结构,这和大型机电设备结点之间的相互影响是完全相似的,因此通过这种动态免疫网络的研究,可望产生更有效的设备故障诊断方法。随着目前大型设备的日益复杂,传感器结点之间的数据是动态互相影响的,如果仅靠从知识库中去寻找最佳匹配来诊断故障,很容易发生漏诊或误诊。目前人工免疫算法中的相似和浓度的定义存在一些问题,例如:利用信息熵的方法计算亲和力和浓度有一些缺陷,同时在抗体评价方面,如果仅以抗体亲合度作为评价标准,不能体现抗原对抗体的影响程度,采用通常的抗体浓度定义,又不能得到抗体亲合度的信息。为此本文根据Jerne的独特型网络理论提出了一个简单、通用性好,又能通过模型的动力学行为综合反映网络细胞(抗体和抗原)的浓度、相似度信息,以提高免疫算法的全局搜索能力的模型,并利用案例检索的优势,考虑了机电设备传感器结点之间相互影响的特性。实验证明,该模型在加快免疫算法在后期的收敛速度和提高故障检测率方面均高于其他算法。本文研究工作的核心是基于独特型免疫网络的故障诊断系统的设计及其在故障诊断中的应用。首先简单介绍了生物免疫系统的一些基本概念、工作原理、生物免疫系统的一些重要机制和生物免疫网络,对故障诊断技术及电机设备故障诊断原理进行了总结。接着介绍了人工免疫网络当前的研究及应用状况,并对一些经典的免疫网络模型的基本结构和算法流程进行了研究和分析。最后基于这些工作提出了一个基于独特型免疫网络原理的故障诊断模型,并利用改进的抗原、记忆抗体在形态空间的表现形式和简化的抗体刺激度及抗体浓度计算公式以及改进的亲和力计算公式进行训练。通过仿真实验,研究了该免疫模型的性能及其在电机故障诊断中的有效性。更多还原

【Abstract】 Fault diagnosis is to identify and diagnose fault through the mapping from the defective symptoms space to the defective space. However, many complex systems often change dynamically, and their characteristics are not grasped, so some faults have very strong uncertainties. All of these brought us great difficulties for intelligent diagnostic reasoning via obtaining, expressing and using diagnosis knowledge effectively. Idiotypic-Network theory points out that the various cells and molecules are not in a state of isolation in immune-network model. They constitute a dynamic network structure through self-identification, mutual stimulation and their mutual constraints. This phenomenon is quite similar to the interaction between the nodes of large-scale mechanical and electrical equipments. So we expect to design more efficient fault diagnosis methods by studying this dynamic immune network.With the large equipments becoming more and more complicated, data collected from transducer nodes are influence each other, therefore, which lead to diagnostic errors or misdiagnosis when we only depend on matching the analogical data from fault diagnose knowledge-base. However, the definition of analogy and concentration in the artificial immune algorithms has some problems, for example, the methods for calculating affinity and concentration by using entropy of information have some defects. At the same time, when we evaluate antibodies only by antibody’s affinity-degree, we can not accurately acquired the influence between antigen and antibody. If we evaluate it by accustomed definition, we can not acquire the information of antibody’s affinity-degree.This thesis proposes a simple model which can synthetically reflect the density of network cells (antibody and antigen) and the information of alike-degree between them according to Jerne’s Idiotypic-Network theory,this model also consider the characteristic of influence between transducer nodes and utilization advantage of CBR. Experiments show that the model is faster than other immune algorithms in speeding the later rate of convergence of immune algorithms and improveing the rate of fault diagnosis.The core content of this thesis is the design of fault diagnosis system based on idiotypic immune network and its application to fault diagnosis. Firstly, some basic concepts, framework and principles of the biological immune systems are briefly introduced. We then introduce the research content, research status and basic theory of the artificial immunological network. The framework and flow of some classical algorithms of immunological network model and its structure are studied and analyzed. This thesis also summarizes the fault diagnosis technologies and the principle of electrical equipment fault diagnosis. At last, this thesis gives us a simple fault diagnosis model based on Idiotypic Immune Network, and train it by improved formula including affinity calculation, antibody concentration, antibody stimulation and the form of performance of improved antigen and memory antibody in modal space. The simulation experiment about asynchronous electromotor shows the feasibility of the artificial immune-network system proposed above.更多还原