【作者】 张丽莉；

【导师】 储江伟；

【作者基本信息】 东北林业大学 ， 载运工具运用工程， 2009， 博士

【摘要】 本文在广泛收集和整理分析国内外汽车故障诊断研究相关资料的基础上,对汽车故障诊断基础理论和技术方法应用现状进行了分类,并分析了不同阶段的研究重点以及各种不同方法的特点,论述了研究的主要发展趋势,以此为依据,确定了运用信息融合的相关理论和方法,研究汽油发动机电控系统故障诊断方法与技术为目的,以应用多种模式识别方法进行特征级融合诊断的研究方向,其主要内容包括：第一,在对现有的汽车故障诊断方法特点进行深入分析、研究和归纳分类的基础之上,提出未来汽车故障诊断方法和技术将会在传统方法的基础上,不断融合各种先进的技术和理论,并加强反馈系统的故障诊断,同时,在分析了信息融合理论的特点和适用条件的基础上,分析了信息融合理论应用于汽车电控系统的故障诊断的适用性；第二,研究了模式识别和信息融合的基本理论和技术,其中重点研究D-S证据理论及其关键问题的解决方法,包括证据体的基本可信度分配问题、证据的冲突问题、证据体的相关性问题以及不同识别框架下的证据组合问题。并研究基于D-S证据理论的信息融合方法应用于汽车电控系统故障诊断的理论基础,提出将各个独立的低维神经网络的输出值处理后作为辨识框架上命题的基本可信度分配,然后经过证据理论的再次融合后得到最终的诊断结果。第三,在深入研究汽油发动机电控系统及其控制原理以及相关数据流的基础上,分析传感器和执行器类间和类内故障均可分的理论依据,为实现基于类内和类间的特征级融合提供基础。第四,针对汽车电控系统特征参数的提取和选择问题,将传统方法和基于核的特征选择和提取方法相结合,并根据汽车故障诊断的参数要求,从传感器和执行器类间故障以及传感器类内故障和执行器类内故障分别提取出最优化的特征参数；第五,通过汽车电控系统的故障的诊断测试分析,验证基于信息融合的汽车电控系统故障诊断方法的有效性,并研究该方法用于汽车电控系统故障诊断的精度问题,实验证明,该方法可以在一定程度上提高识别的准确率,消除单一数据源包含信息的不全面性以及模糊性等等,从而有效提高故障诊断的精度。基于神经网络和D-S证据理论融合的汽油发动机电控故障模式识别是建立集成化和智能化汽车故障诊断的理论基础,相关理论的应用研究是提高汽车故障诊断精度的必要条件,该方法不仅是汽油发动机电控系统故障诊断智能化的有效方法,也是对汽车整个电控系统故障诊断方法的新探索,其研究将促进故障诊断智能化的发展进程。更多还原

【Abstract】 On the basis of extensive collection of automotive fault diagnosis study on home and on broad, the classification of basic theory and application of automotive fault diagnosis are present in this paper, and then analysis the key technologies of various periods and its character and development direction. Based on those, put forwards that the study direction would be using information fusing theory to study the automotive electric-controlled system fault, and adopting multiple pattern recognition method for character level fusing, the main content is shown as follows:At first, based on the analysis of present automotive fault diagnosis method, summarize the automotive fault mechanism and feature, put forward that the future automotive fault diagnosis method and technology would be the fusing of various advanced technology and theory which based on traditional method, and paying attention to feedback system, then, on the basis of analysis on information fusing theory’s character and applicable condition, analyze the theory applicability for automobile diagnosis. The second, the basic theory and technology of pattern recognition and information fusing are studied in this paper, especially for D-S evidence theory and its key problems resolve method, including basic evidence reliability assignment, evidence conflict, relativity and how to assemble evidence on the different recognition frame, and then study the information fusing method based on D-S evidence theory, put forward that let the output value from low-dimension neural network as the basic reliability assignment, then through D-S evidence theory to fusing again until the final diagnosis result. The third, on the basis of study on automotive electric controlled system and its control principle and data flow, analyze the theory basis of reliability between sensor and actuator, sensor itself and actuator itself, those provide the basis for character fusing. The fourth, for parameter choosing, combination traditional method with kernel method, then according to the automobile fault requirement,from sensor and actuator, sensor itself or actuator itself to choose the best character parameter for diagnosis. The fifth, through diagnosis detection system, inspect the validity, and then study the precision problem. The experiment shows that, this method could promote the rate of accuracy on some degree, eliminate the single data source’s one-sidedness and fuzzification and so on, then promote the precise efficiency.Automobile electric-controlled system fault pattern recognition based on NN and D-S evidence theory is the basic of establishment on integration and intelligent fault diagnosis, and relative theory’ application study is the necessary condition for promoting the diagnosis precise, this method is not only the efficient method about intelligent diagnosis, but also the whole vehicle’s electric-controlled system fault diagnosis’s new quest, it would be promote the development process about intelligent diagnosis.更多还原