【作者】 董广明；

【导师】 陈进；

【作者基本信息】 上海交通大学 ， 机械设计及其理论， 2007， 博士

【摘要】 区别于传统的结构局部无损检测技术(NDE)的结构损伤全局检测技术一般可以分为两种:基于动力学模型的损伤检测方案和基于信号分析的无模型损伤检测方案。本文第二章首先研究了基于结构动力学模型的裂纹轴损伤位置和深度的识别问题,然后在第三章以后,主要篇幅用于研究基于信号分析,无模型的损伤全局检测方案。针对结构损伤全局检测基本问题,采用仿真和试验手段分别研究了如何识别结构体系中损伤的发生、损伤位置和损伤程度的不同方法。第二章主要研究基于结构动力学模型的结构损伤全局检测方法。结构裂纹检测是一个较难的工程问题,为有效检测具有单一开裂纹的圆轴中的裂纹位置和深度,首先建立了基于Euler-Bernoulli梁和局部柔度理论的单一开裂纹轴连续模型,并通过振动分析研究了裂纹位置、深度与裂纹轴前两阶固有频率、振型的关系。根据研究结果,提出了一种根据裂纹轴的前两阶固有频率以及相应第一阶振型来识别裂纹位置和深度的方法。算例分析表明,在建立了完善的裂纹轴有限元模型或解析模型后,通过裂纹轴的模态参数分析,可以有效地确定裂纹的位置和深度。结构发生损伤后,如连接结构的松动会产生间隙而导致结构的撞击和摩擦,结构的振动响应信号会产生某些奇异性,本文第三章利用小波奇异性检测理论分析结构在受损时的振动信号,从而判断结构受损伤时刻,并着重研究了结构在高斯白噪声随机激励下,消除激励对响应信号奇异性影响的小波变换方法。支撑座结构的简单仿真模型分析以及钢框架结构仿真算例表明了该方法的有效性。本文第四章根据经验模态分解(EMD)得到的信号分量特点,研究了两种用于识别结构损伤发生的方法。采用EMD和Wigner-Ville分布(WVD)结合的时频分析方法,可有效抑制交叉项干扰,同时不降低分析结果的时频分辨率,这种方法适用于检测旋转机械结构损伤导致的倍频成分;针对非旋转机械结构损伤前后频率变化不大的特点,采用限带宽EMD分析并结合随机减量方法,得到对应结构各阶模态的自由衰减振动信号,从而实现结构损伤前后固有频率和振型阻尼的识别,进而判断损伤的发生。钢框架结构仿真算例表明,EMD-RDT方法可准确识别结构不同状态下的各阶固有频率与振型阻尼;支撑座结构模型实验分析表明该结构的频率与阻尼辨识结果可以作为判定螺栓组松动发生的依据。当结构发生早期损伤时,由于结构的基频下降很少,因此有必要寻找对结构早期损伤敏感的特征参数;另一方面,工程结构及其输出信号因受许多随机干扰而存在许多不确定性,于是采用结构损伤统计模式识别方法也是一个必然。因此,第五章基于单传感器测量信号的特征提取及统计模式识别来识别结构的早期损伤及其损伤程度,研究并对比了ARX以及AR-ARX模型损伤特征参数序列在结构不同损伤程度下的统计分析结果,并进行了不同参数下的讨论。钢框架结构仿真算例和支撑座结构模型实验分析表明均表明两者可以及时反映结构早期损伤以及早期损伤的不同程度,且对噪声和传感器的放置位置不敏感。研究基于全局的结构损伤位置识别方法很多,作为方法的补充,并出于博士论文完整性的考虑,第六章对钢框架模型结构,首先建立了钢框架结构连接螺栓松动的动力学模型,然后根据多传感器信号双谱特征并结合神经网络分类方法,对钢框架结构不同螺栓松动位置进行了识别。更多还原

【Abstract】 The global-based structural damage detection techniques, which are different from the commonly used local based non-destructive evaluation (NDE) methods, can be generally classified into two kinds: one kind of methods depends on the structural dynamic model, while the other kind of methods is independent of the structural model and only the measured structural response signals are included to obtain the aim of damage examination. In this paper, model based methods and non-model based methods are all studied: firstly, the crack depth and position identification problem of a cracked shaft is studied based on its dynamic model in chapter 2; secondly, different non-model based methods, which compose main part of this dissertation, are studied to detect the structural damage existence, degree and position via simulation and experiment.In the second chapter, a structural dynamic model is established to study the model based method: the analytical continual model of a shaft with an open crack,which is based on the Euler-Bernoulli and local flexibility theories. Vibration analysis of the cracked shaft based on the continual model is carried out and relations between crack parameters and modal parameters are studied. A crack depth and position identification method is put forward based on the first two natural frequencies and the first mode shape of cracked shaft. Simulations show that the method, which is based on the finite element model or the analytical continual model, is effective and accurate in determining the crack position and depth of a shaft with an open crack.Singularities of structural response signal will result from damage occurrence, for example, loosed connecting components will induce structural impact and rub. In chapter 3 of this dissertation, wavelet singularity detection theory is used to analyze the structural responses when damage occurs, by which the time of damage occurrence can be obtained. The emphasis of this chapter focuses on the method separating the singularities due to the excitation and the singularities due to the system’s changes when a damaged structure is subject to Gauss white noise excitation, and coefficients of the wavelet transform at a small scale corresponding to the system’s change are separated from the output. Simulations on a three DOF system representing the support model and a FEM model of steel frame structure show that, the proposed method is more effective than frequency analysis and previously used wavelet singularity detection methods.Base on the characteristics of Empirical Mode Decomposition (EMD), two methods are studied to identify the incident of structural damage in chapter 4. The signal time-frequency representation method based on the combination of EMD and WVD is presented, which can reduce the interference effect while does not decrease the time and frequency resolution. The EMD-WVD combined method is effective in detecting multiple frequencies in the rotating machinery damage detection. The structural natural frequency varies little before and after damage for the non-rotating mechanical machinery, therefore, band-limited EMD method combined with the random decrement technique (RDT) is used to determine the response of each mode and obtain the free vibration modal response, which is utilized to identify natural frequencies and damping ratios of structures in various states. Simulation and experimental cases are studied, which show the validity of the proposed EMD-RDT method.The structural modal parameters such as the first natural frequency decreases little for the small incipient damage, which makes it essential to find sensitive damage patterns; on the other hand, there are many kinds of uncertainties in structural responses caused by many random disturbances, which make it necessary to use statistical methods for damage pattern analysis. In chapter 5, ARX and AR-ARX model damage pattern sequences are studied and compared through statistical analysis and parametric discussion. Simulation and experimental results show that the two damage pattern sequences are both sensitive to the early structural damage and able to identify different degree of early damage, moreover, the two damage pattern sequences are both insensitive to noise and sensor position.There are many global based methods studying structural damage position identification. As a method implementation and completeness of the whole dissertation, firstly a FEM model of a steel frame with loosed connecting joint is established in chapter 6; then joint looseness position identification method is put forward based on the combination of bispectrum and network. Simulation results show its validity under large noise conditions.更多还原