一种可用于机车部件故障诊断的神经网络专系统研究

【作者】 刘伟荣；

【导师】 蒋新华；

【作者基本信息】 中南大学 ， 计算机应用技术， 2003， 硕士

【摘要】 本文针对内燃机车的并发故障和交叉故障多的特点，提出了诊断机车故障的一种神经网络专家系统模型，首先依据机车的结构和各个部分的组成设计分布式协同专家系统，将专家系统的各种知识和规则分布于各个协作的子专家系统中。在子专家系统中采用了传统的基于规则的故障诊断方式和神经网络故障诊断方式。各个协同工作的专家系统利用规则推理来解释神经网络的识别结果，神经网络用来避免烦琐的推理而直接识别故障类型。因此本系统通过在传统的专家系统中引入神经网络提高了专家系统的适应性和灵活性，并具有了一定的联想功能，可以识别带有噪声和残缺的数据。在本系统中选用ART神经网络，ART网络具有自动聚类的功能，特别适合于发现偶发性故障。 在推理过程中为了比较灵活的处理各种情况，系统使用了不精确推理、非单调推理等高级推理技术，利用可信度因子，严重程度因子来解决推理过程中可能出现冲突问题。 本论文分析了机车的结构和组成，给出了专家系统模型诊断的机理和实现，对系统的诊断过程做出说明。并指出了本系统的优缺点、适用范围以及将来的发展方向。更多还原

【Abstract】 The paper proposes a model to support fault diagnostic expert system for large and complex diesel locomotive. The model based on the structure and the parts of diesel locomotive is a distributed and cooperated expert system. The rules and knowledge is distributed many subordinate expert system. In the subordinate expert system, there are two ratiocinating modes. The one is the mode based on rules by usual expert system another is mode using neural network to recognize the type of the faults . That is giving the system the ability to recognize the fragmentary data or the data with noise. The system used the ART neural network because study of the ART is without teachers and can learn real time, So it is suited to recognize the faults that happen by accident.the system uses the advanced ratiocination such as uncertain ratiocinations and the default ratiocinations. By used the degree of reliability and the degree of importance to solve the confliction.The paper analysis the structure and makeup of diesel locomotive firstly, gives the the principle and realization of the expert system and gives the concrete design and implement of subordinate expert system, it evaluates the advantages and disadvantages of the system and give the area suited for this system and look forward the development of fault diagnosis system.更多还原