【作者】 曹龙汉；

【导师】 曹长修；

【作者基本信息】 重庆大学 ， 控制理论与控制工程， 2001， 博士

【摘要】 柴油机故障诊断技术是利用柴油机的状态信息和历史状况，通过分析和处理来定量识别其实时技术状态，并预测异常故障的未来技术状态的一门建立在多学科基础上的综合技术。本文试图从工程应用的角度出发，以柴油发电机组为研究对象，以测试技术、信号处理、小波分析、模糊聚类、人工神经网络以及粗糙集等理论为基础，深入研究柴油机的故障机理、故障特征提取和诊断方法。论文主要由八部分内容组成：1)、简述了论文的研究目的和意义。综述了柴油机的各种状态监测方法和故障诊断技术的研究现状，分析了各种诊断方法的特点及适用范围，指出了存在的问题和将来的发展趋势。2)、为了提高FFT的谱分析的精度，分析了加窗频谱分析方法误差产生的原因。针对加窗傅氏变换的栅栏效应，研究了各种窗谱校正和频率细化方法。分析了样本长度对窗谱校正精度的影响。3)、从故障诊断角度出发，给出了柴油机缸盖振动信号的信息模型。分析了柴油机缸盖振动信号的时域、频域和循环波动特性，提出了抽区间采样分析和参数平均相结合的振动诊断方法，研究了柴油机缸盖表面振动信号的检测及实现方法。在此基础上，研究了气门间隙异常和气门漏气等典型故障的振动诊断机理和诊断方法，揭示了这些故障与缸盖振动信号之间的本质联系。4)、在介绍柴油机供油系统的组成及燃油喷射过程的基础上，对高压油路中的三偶件磨损、喷油过程等的故障机理进行了深入的分析，对其造成的影响进行了讨论。根据供油系统的工作特点，讨论了供油系统各种典型故障状态下的燃油压力波动情况，分析了燃油压力波形和故障之间的对应关系。5)、研究了柴油机故障诊断的特征信号提取方法，包括基于短时AR分析的信号特征提取方法和基于小波多分辨率分析的特征提取方法。利用短时AR分析，给出了从柴油机缸盖振动信号中提取整循环诊断征兆的实现方法。在介绍连续小波变换、二进离散小波变换和小波包变换的基本理论的基础上，给出了小波分析中最重要的算法—Mallat快速算法。研究了小波包分解结果中存在的频率混叠现象，提出了一种抗频率混叠的小波包改进算法。利用小波多分辨分析或小波包分析以及Kllback-Leibler信息量最小，通过对柴油机表面振动信号进行小波分解与分析，确定了各故障状态的特征频带、进而可用特征频带的时间序列的时序模型参数作为特征矢量，实现对柴油机运行状态故障的诊断。阐明了燃油压力波形进行小波分解后，可以消除噪声干扰，提高信噪比。选择小波逼近似信号和某些分解信号，提取相应的诊断特征参数，可以作为每一个输入模式的属性值，实现常见供油系统故障的诊断。6)、以模糊数学为基础，介绍了模糊C-均值聚类算法的基本原理，给出了模糊C-均值聚类算法的一般计算步骤，对模<WP=4>糊聚类的有效性进行了深入分析。提出了应用模糊C-均值聚类算法实现对柴油机活塞—缸套磨损的故障分类和诊断，实际应用证明了其诊断方法的有效性。7)、介绍了粗糙集理论的核心内容，给出了连续属性值的离散化方法。论文应用粗糙集理论对反映柴油机运行工况的特征参数进行了属性简化，通过人工神经网络实现了对柴油机故障的自动分类和诊断。8)、论文的最后一章对全文的主要研究内容进行了总结，给出了主要研究结论和创新点，指出了论文的不足和今后的研究重点。更多还原

【Abstract】 The fault diagnosis technology of diesel engine is a comprehensive technology built upon many subjects. It can be quantificationally identified the technical state of real time of diesel engine, and the future technical states of abnormal fault are forecasted with the status messages and history status of diesel engine through analysis and handling. This paper attempts to research profoundly fault mechanism, fault feature extraction and fault diagnosis method of diesel engine from the view of project application. Its research object is diesel-dynamotor unit and its academic foundation is test technology, signal processing, wavelet analysis, fuzzy clustering, artificial neural network and rough sets theory, etc.This paper consists of eight parts: i) The meaning and research purpose of this paper is described briefly, the research present situation of fault diagnosis technology and the various state inspection methods of diesel engine are summarized. The suitable scope and the characteristic of various diagnosis methods are analyzed. The future development and the existed problems are pointed out. ii) To raise the precision of frequency spectrum analysis of FFT, the leakage error reasons of the window spectrum analytic method are analyzed. Aimed at the fence effect of window Fourier transform, the various methods of window spectrum emendation and frequency subdivision are studied, and the influence of the sample’s length for window spectrum calibration accuracy is analyzed. iii) From the point of view of fault diagnosis, the information model is given and the frequency domain, time domain and circular fluctuation characteristic of the cylinder cap vibration signal of diesel engine are analyzed. A method of combining sampling analysis by means of section and parameter average is put forward, the measure method and realization of the cylinder cap surface vibration signal of diesel engine are studied. On this foundation, the vibration diagnosis mechanism and diagnosis methods of some typical fault, such as an abnormal valve gap and gas leakage from exhaust valve, are also studied, and the essential connection between these faults and the cylinder cap vibration signal of diesel engine is revealed. iv) On the foundation of introducing the composition of fuel injection system and the fuel injection process of diesel engine, the fault mechanism of 3 coupling parts abrasion and fuel injection process is analyzed deeply, and its influence is discussed in the same time. According to the working characteristic of fuel injection system, its fuel pressure wave that various typical fault states are discussed, and the corresponding relation between fuel pressure waveform and faults is analyzed. v) Many kinds of methods<WP=6>for the feature extraction of fault diagnosis to diesel engine, including the methods based on short time signal AR analysis and the methods based on wavelet multi-distinguish rate analysis, are studied. Using short time signal AR analysis, the realization method of full circulating feature extraction from the cylinder cap vibration signal of diesel engine is given. After introducing some basic theoretical foundation such as continuous wavelet transform, binary discrete wavelet transform and wavelet packet transform, a most important algorithm in wavelet analysis－Mallat fast algorithm is given. The mixed-frequency phenomenon existed in the result of wavelet packet decomposition is studied. An improved algorithm of wavelet packet transform that can prevent frequency alias is put forward. After using wavelet decomposition for the surface vibration signal of diesel engine, the characteristic frequency band of each fault state can be confirmed with the use of multi-resolution wavelet analysis or wavelet packet decomposition as well as least Kllback-Leibler information. Using its AR model parameter as characteristic vector, the fault diagnosis of diesel engine can be realized. It is expounded that fuel pressure waveform carried out in wavelet decomposition can remove noise interference and enhance S/N ratio. S更多还原