【作者】 牛慧峰；

【导师】 姜万录；

【作者基本信息】 燕山大学 ， 机械电子工程， 2009， 博士

【摘要】 故障检测与诊断技术是提高系统可靠性和可维修性的有效途径,近年来智能故障诊断理论与技术发展飞快,其核心是有效的获取、传递、处理、再生和利用诊断信息,从而能够对给定环境下的诊断对象进行准确的故障模式识别和预报。目前智能故障诊断领域面临的主要难题是典型故障样本的严重不足以及诊断知识的发现困难,这两者都严重制约着智能故障诊断理论和技术的发展。人工免疫系统是对生物免疫系统的模拟,具有强大的信息处理能力。生物免疫系统主要的功能是在线检测和杀伤来自生物体内和体外的抗原,具备“自己/非己”识别能力。由此衍生而来的阴性选择算法能有效地识别被检测对象的正常模式(自己)和异常模式(非己),且不需要有异常模式的先验知识。该算法为智能故障诊断提供了新思路和新方法。支持向量机的诞生较好地解决了以往许多学习算法在对小样本、非线性和高维数据进行学习方面的不足,并克服了神经网络等学习方法中网络结构难以确定、收敛速度慢、局部极小值、过学习与欠学习以及训练时需要大量样本等不足,可以使在小样本情况下建立的分类器具有很强的推广能力,这对系统故障诊断具有很强的针对性和适应性。将支持向量机与人工免疫机理相融合用于智能故障诊断,提出了一种新的人工免疫机理与支持向量机复合的故障诊断方法,并通过对液压系统的重要元件液压泵多模式故障诊断的试验研究,验证了该方法的有效性。对丰富故障诊断的理论体系,具有重要的理论意义和工程实用价值。本文主要进行了以下几个方面的工作:(1)综述了人工免疫系统的起源、发展,以及人工免疫系统的机理、算法。重点研究了阴性选择机理及其算法的实现。对目前先进的学习机器——支持向量机的背景、理论、特点和应用进行了较为详尽的阐述。研究了支持向量机用于故障诊断的关键问题,给出了基于支持向量机故障诊断的基本实现步骤;编写了支持向量分类机应用软件,并对支持向量机的部分核函数参数选择进行了研究。(2)将支持向量机方法与实值阴性选择算法相融合,提出了一种复合的故障诊断方法,用于斜盘式轴向柱塞泵的多模式故障诊断与识别。仿真研究表明,该方法能够有效地提高故障诊断的确诊率。(3)针对振动信号出现的调制现象,研究了基于复解析小波簇的解调原理。鉴于设备出现故障时,各频带内信号的能量分布会发生较大变化的特点,提出了包络信号的小波包分解子带特征能量法,用于提取信号的特征向量,来刻画设备的运行状态。(4)针对主元分析在故障特征选择上存在的不足,提出并实现了一种有效的基于核主元分析的非线性故障特征选择方法。阐述了核主元分析的基本原理和算法,通过对故障特征向量进行核主元分析,能有效的降低特征向量的维数,从而降低了故障分类器的计算复杂度。核主元分析对于非线性数据特征较为敏感,适合于液压系统故障的非线性特征选择。(5)基于传统的油膜理论,分析了滑靴在运动过程中对斜盘产生冲击的机理,通过试验验证了理论分析的正确性。组建了基于虚拟仪器的试验系统,选择泵端盖振动信号和泵出口压力信号作为监测信号,用松靴柱塞和磨损的配流盘替换正常元件来模拟多个松靴故障和配流盘磨损故障失效形式,采集了大量各种故障模式的样本数据。由于滑靴对斜盘冲击产生的振动信号被高频谐振信号调制,通过小波簇包络解调方法将调制信号解调出来,解调后的信号包含了丰富的故障信息。对包络信号用小波包分解子带特征能量法进行特征提取,构成正常和各类故障的样本集,并用核主元分析方法对样本集进行降维处理,用免疫支持向量机复合故障诊断方法进行诊断,验证了该方法在提高故障诊断确诊率方面的独特优势。更多还原

【Abstract】 The fault detection and diagnosis technique is an effective way to enhance system reliability and maintainability. In recent years, the intelligent fault diagnosis theory and technology develops very fast. To effectively acquire, transfer, deal with, regenerate and utilize diagnostic information has been the core of intelligent fault diagnosis so that it could exhibit an ability to precisely identify the patterns of faults and predict future faults. Recently, the dominating difficulties that the field of intelligent fault diagnosis faces are terrible lack of typical fault samples and finding problem of diagnosis knowledge, both of which badly prohibit the development of intelligent fault diagnosis theory and technology.Artificial immune system, which has powerful information processing ability, simulates biology immune system. The main function of biology immune system provided with the ability of recognizing“Self”and“Non-self”is to detect, damage and kill the antigen from inside or outside body in the real time. Thus, without the transcendent knowledge of abnormal mode, the algorithm of Negative Selection (NS) derived from that can recognize the normal (Self) or abnormal (Non-self) mode of the detected objects efficiently. Therefore, this algorithm provides the new idea and method to intelligent fault diagnosis.Owing to naissance of Support Vector Machine (SVM), the shortages of learning aspect are solved in many learning algorithms using finite sample data, nonlinear data, and high dimension data. And SVM also overcomes shortcomings of the neural network learning methods, for example, determining the network structure difficultly, slow convergence, local minimum value, over-fitting & under-fitting as well as excessive need for training sample. SVM makes the generalization ability of classifier better in the finite sample condition, which has strong pertinence and applicability for the system fault diagnosis. So the composite intelligent fault diagnosis method is put forward based on support vector machine and artificial immune mechanism. The validity of the founded method is verified through the experiment research of diagnosing fault of the hydraulic pump, the important components in hydraulic systems. These researches have important theoretical significance and engineering practical application value for enriching the fault diagnosis theory.The main works in this dissertation are showed as follows:(1)The origin and development of artificial immune system is summarized and the mechanism and algorithm of artificial immune system is researched. The mechanism of NS algorithm and its realization is studied. The background, theory, characteristics and applications of support vector machine which are advanced study machine presently are expounded. The key problem is studied in applying support vector machine to fault diagnosis, and the foundational realizing steps are brought out for support vector machine applying to fault diagnosis. The application software of SVM is programmed and the choice principle of some kernel function parameters is researched.(2)A composite intelligent fault diagnosis method is brought forward using the support vector machine and the real-valued negative selection algorithm. This composite method is used to the multi-pattern fault diagnosis and recognition of swashplate axial piston pump. The simulation research is shown that the right rate of fault diagnosis is enhanced effectively using this method.(3)Aiming at the modulation phenomenon of the vibration signal, the principle of envelop demodulation based on the complex analytical wavelet cluster is researched. When the equipment generates fault, signal energy distributing each frequency band has a greater change. The wavelet packet decomposition sub-frequency bands characteristics energy method of the envelope signal is introduced for the extraction of signal eigenvectors which can reflect the status of equipment operation.(4)In allusion to some shortage of principal component analysis(PCA) in fault feature selection, a kind of effective nonlinear feature selection method, which based on kernel principal component analysis(KPCA), is suggested and realized. The basic principle and algorithm of KPCA is expatiated. Through the selection of feature vectors, dimensions of feature data and calculating complexity of classifier are decreased effectively, and experiments show that KPCA is sensitive to nonlinear features, and it is suitable for the selection of nonlinear features of hydraulic system faults. (5)Based on the traditional oil-film theory, the slipper/swashplate impact mechanism during the slipper operation is analyzed and the correctness of the theoretical analysis is verified through the experiment. The experiment system based on virtual instrument is composed and the end-cover vibration signal and the outlet pressure signal of pump are selected as the monitored signals. The loose slipper failure and the wearing valve plate are simulated artificially by replacing the normal components of pump with loose slipper piston and worn valve plate. And a mass of data samples of multi-fault pattern acquired. Because vibration signal generated from slipper/swashplate impact is modulated by high frequency resonance signal, the modulation signal has to be demodulated using the wavelet cluster envelope demodulation method, and the envelope signal contains abundant fault information. After the envelope signal is decomposed using wavelet packet, signal eigenvectors which compose the sample set (normal and fault), are extracted and the dimensions of sample set are decreased using KPCA. The samples are diagnosed using composite fault diagnosis method, whose validity is verified.更多还原