【作者】 李俊松；

【导师】 宋仲康；

【作者基本信息】 武汉理工大学 ， 控制理论与控制工程， 2003， 硕士

【摘要】 随着21世纪的到来，人类对生存环境、资源的有效利用、空气污染越来越关注。为了节约能源、保护环境、实现社会可持续发展，我国汽车工业开始对混合动力汽车(HEV)进行积极的研究与开发。为了提高HEV整体和各系统的安全性，迫切需要在产品开发研制阶段就建立一系列汽车故障诊断系统，得出结论以求设计改进或采取必要的措施，防止系统灾难性事故的发生，这就需要研究和应用汽车故障诊断技术。 汽车故障诊断技术是以工程数学、可靠性理论、信息理论为基础；以电子技术、计算机技术、人工智能技术为手段的一门综合应用技术。在过去的十几年里得到了飞速发展，产生了一些新的理论与方法如：主元分析、遗传算法、小波变换、神经网络、模糊系统、模式识别、自适应理论、非线性理论等。其中，人工神经网络的研究迅速发展，为系统的故障诊断开辟了一条新的途径。 本文对国内外汽车故障诊断技术的发展、汽车故障的种类与特点、汽车故障的诊断方法、故障诊断系统的基本诊断过程和理论方法进行了深入的分析。根据神经网络的特点，指出神经网络与故障诊断结合的可行性和必然性。在讨论了神经网络的基本理论基础上，明确神经网络应用于故障诊断的三种途径。 前馈型BP网络具有极强的模式识别和分类能力。本文从应用角度分析了网络设计中的网络的层数、隐含层的神经元数、初始权值、学习速率、期望误差的选取问题，并提出了相应的改进方法。通过汽车故障诊断专家系统实例的应用与仿真，表明其方法的实用性以及诊断结果的准确性和可靠性。 汽车故障诊断中可利用的信息很多，只有充分有效融合有用的信息来对设备的故障进行诊断才能提高故障诊断的精度和可靠性。本文分析了单子神经网络进行故障诊断的特点，构架了集成神经网络的故障诊断模型，研究了集成神经网络的建模方法、组建原则和实现策略，并结合汽车发动机故障诊断实例进行了仿真分析，结果表明利用神经网络信息融合进行汽车故障诊断是一种有效的方法，能够获得对故障状态的最优估计与判决。更多还原

【Abstract】 With the coming of the 21 centuries, mankind pays more attention to the environment of survival, the valid utilization of resources, the air pollution. For the economy energy , environmental protection , realizing the society can keep on developing, automobile industry have begun to research and develop the hybrid electric vehicles (HEV) actively in our country. For ensuring safety to HEV’s the whole and the part system, it is urgent to set up a series of auto fault diagnosis system in the process of its developing and researching, in order to getting the conclusion to improve design or adopt the necessary measure to prevent the occurrence of the system accident, which need to research and apply the fault diagnosis technique.The auto fault diagnosis technique is a kind of synthesizes applied technique with regarding engineering mathematics , reliability theories , information theories as the foundation and regarding electronics technique , computer technique , artificial intelligence technique as the means. It has developed quickly in past decade. The some new theories and methods have produced, such as principle component analyses , genetic algorithm , wavelet theory , artificial neural network N fuzzy system , pattern classification adaptive control theory, nonlinear system theory, etc. Among them the research of the artificial neural network develops quickly, which make a new way for fault diagnoses.This paper analyses the development of fault diagnosis technique in domestic and international, the category and characteristics of auto fault , the method of auto fault diagnosis , the basic diagnosis process and theories method of the fault diagnosis system. According to characteristics of the neural network, it is indicated that the combination between the neural network and the fault diagnosis have possibility and inevitability. Basing on discussing the basic theories of neural network, it is defined that neural network can be applied for fault diagnosis in three ways.The BP network of forward type have the very strong mode identification and classific ability. Form appliance aspect this paper analyzes network layer number , nerve cell number of inside laye, original weigh , train speed, expecting error when designing BP network and puts forward the improvement method. The result based on the simulation of the auto fault diagnosis expert system example shows it has practicability and diagnosis result have veracity and reliability.There are much available information in auto fault diagnosis, only when the available and mulriple information is fused, can the precision and credibility be improved. Based on the analyses of single neural network characteristic, the model of integrated neural network is put forward in this paper. At the same time, the way to establish model, the principle to compose and the strategy to realize are given in this paper. The result based on the simulation of the auto engine fault diagnosis example shows it is a effective way and can acquire superior estimate and sentence of fault information.更多还原