【作者】 王向；

【导师】 刘才；

【作者基本信息】 燕山大学 ， 机械设计及理论， 2012， 博士

【摘要】 非线性是机械系统动力学的固有属性。大型机械系统是一个复杂的非线性动力学系统，具有不确定性、非线性、时变性的特点，故障状况复杂，干扰因素多，基于线性理论的机械故障诊断方法具有很大局限性，难以满足实际诊断需要。因此，开展复杂非线性系统的故障诊断和容错控制方法研究具有重要的理论意义和实际应用价值。在本文中，针对复杂非线性系统，研究了智能故障诊断方法和几种复杂非线性系统的容错控制方法。主要研究内容如下：本文研究了小波变换和混沌理论与神经网络相结合进行故障诊断的方法。在故障诊断中，首先利用小波对所分析的信号进行消噪预处理，有效的检测出故障信息成分，并使用关联维数等混沌特征量刻画振动信号的故障特征，对机械设备的状态进行分类和识别，克服了传统方法在故障信号特征提取和分析上的困难。然后提出了小波神经网络的网络模型，并使用改进的遗传算法对神经网络模型的权值和阈值进行优化，避免了传统BP算法的不足，并通过实验，验证了算法的可行性。提出了一类非线性时滞系统的容错控制方法。针对带有故障的非线性时滞系统和非线性切换时滞系统，设计基于自适应故障估计算法的非线性系统自适应观测器，通过对故障的自适应估计设计故障系统容错控制器，并通过李雅普诺夫有界性理论验证算法的有效性。提出了一类时滞中立系统的容错控制方法。针对带有故障的时滞中立系统，采用自适应故障估计算法，设计基于输出反馈的时滞中立系统全维观测器，并通过李雅普诺夫有界性理论验证了算法的有效性。提出了一类非仿射非线性神经网络自适应容错控制方法。由于神经网络的非线性逼近能力，用神经网络对非线性系统故障进行估计，设计了神经网络故障诊断方法，并根据神经网络的逼近能力设计了非仿射非线性系统神经网络控制器，神经网络的权值系数通过自适应算法进行在线调整，根据李雅普诺夫理论，证明了系统闭环跟踪误差是一致有界稳定的。更多还原

【Abstract】 Nonlinear is the inherent attribute of the mechanical system dynamics. Largemechanical system is a complex nonlinear dynamic system, it has uncertainty, nonlinear,time variation etc, fault condition is complex, and interference factors are great.Mechanical fault diagnosis methods based on linear theory have great limitations, it isdifficult to meet the practical needs for diagnosis. Therefore, development of complexnonlinear system fault diagnosis and fault tolerant control method research has importanttheory significance and practical application value. In this thesis, for complex nonlinearsystem, some kinds of different intelligent fault diagnosis methods are studied, and severalfault-tolerant control methods of nonlinear system are put forwarded. The main contentsof this thesis are as follows:In this thesis, wavelet transformation and chaos theory are combined with neuralnetwork for fault diagnosis. In fault diagnosis, first, wavelet analysis is used in signalde-noising pretreatment, the fault information components are detected effectively, and thecorrelation dimension and other chaotic characteristics are used to depict the fault featuresof vibration signal, the status of mechanical equipment are classified and recognized, thismethod overcomes the difficulty of the traditional method in fault signal feature extractionand analysis. Next, the network model of a wavelet neural network is put forwarded, andusing improved genetic algorithm who has global search capability on the optimization ofthe weights and thresholds, avoiding the disadvantages of BP algorithm, and throughsimulation experiment, the feasibility of this algorithm is proved.In this thesis, fault tolerant control method of a kind of nonlinear time-delay systemis put forwarded. For the nonlinear time-delay system and nonlinear switching time-delaysystem with fault, the adaptive fault estimation algorithm is adopted, and nonlinear systemadaptive observer based on the algorithm is designed, and fault-tolerant controller of faultsystem is designed according to adaptive estimation of the faults, and through thelyapunov boundedness theory, effectiveness of this algorithm are proved. Finally, thesimulation experiments show that the algorithm is feasible. In this thesis, fault tolerant control method of a kind of time-delay neutral system isput forwarded. For the time-delay neutral system with fault, the adaptive fault estimationalgorithm is adopted, and whole dimension observer of time-delay neutral system basedon output feedback is designed, and through the lyapunov boundedness theory,effectiveness of this algorithm are proved. Finally, the simulation experiments show thatthe algorithm is feasible.In the thesis, adaptive fault tolerant control method of a kind of non-affine nonlinearneural network is put forwarded. Because of the nonlinear approximation ability of neuralnetwork, neural network is used to estimate faults of nonlinear system, the neural networkfault diagnosis method is designed, and neural network controller of non-affine nonlinearsystem is designed according to the approximation ability of neural network, the weightvalues of neural network are online adjusted through the adaptive algorithm, based onLyapunov theory, the closed-loop system tracking error is uniform bounded stable isproved. Finally, the effectiveness of this adaptive algorithm is verified through thenumerical simulation results.更多还原