【作者】 常虹；

【导师】 殷琨；

【作者基本信息】 吉林大学 ， 地质工程， 2010， 博士

【摘要】 论文基于石油钻井平台从一四层钢框架结构模型出发,围绕该模型发生不同程度的损伤展开识别和诊断工作,提出了基于小波分析和神经网络识别损伤的两步法。在求解结构的动力响应时,引入了计算精度高、无条件稳定且无超越现象的高阶单步-β算法,对得到的相应结点加速度响应进行一层离散小波变换,对所得到的细节信号尖峰高度的变化规律进行分析,考虑了小波函数选取、荷载强度、噪声标准及损伤程度等因素对小波识别损伤能力的影响,提出了适用于工程应用的加速度响应小波变换细节信号的尖峰高度确定损伤位置的方法。为了更真实模拟环境激励,在求解动力响应过程中加入了一定量的噪音干扰,因此论文在对相关数据进行小波变换前需要对所测得的数据进行小波降噪预处理。论文阐述了小波阈值法降噪原理,考虑了小波函数的选取、分解的尺度及阈值选取对信号降噪的影响,以信噪比和均方根误差两个指标作为评价降噪效果的优劣,提出了适用于环境激励的结构动力响应降噪流程。论文将传统的单一损伤识别方法发展为损伤识别两步法。在利用小波分析确定了损伤发生的位置后,引入BP神经网络,采用与损伤程度相关的指标固有频率平方变化比作为特征参数输入网络,从结构发生不同程度的损伤建立了一个三层BP网络,并利用训练好的网络对结构的损伤程度做出判断。更多还原

【Abstract】 During the exploration and development of the oilfield,the large rise rig derrick structure is the essential component that bears the loads.For the environment that it works is far more abominable than the general steel structure thing,especially with the development of marine petroleum industry, oil rig derrick structure will suffer from the influence of earthquake, window storm, fire, flood and such natural disasters in it’s complex working environment. The various kind of negative influence will also made the building structure damaged in different level. All this influence and damage will threaten the Structural Security.This paper carryed out analysis and research on a four floor steel structure health monitoring based on oil rig derrick structure. On the basis of drawing into high degree of accuracy and unconditional stable high ordered single-stepβ-method to count structural dynamic response, this paper pointed out two steps for analysis of structural damage. That is, to use wavelet analysis to supervise the damage on-line and to locate the damage position. Then to use BP neural net-work to define the degree of damage further. The concrete works are as follow：(1) Based on the widely reading, the paper made a summary about the method of large-scale structural’s damage’s detection at the moment, introduced the basis of wavelet theory, and the employment of wavelet analysis and neural network in the field of civil engineering.(2) Since damage of structural will lead to the weakness of natural frequency while calculating the structural’s response. Under the stimulation of environment, the shape of viberation is quite sensitive for partial rigidity, so the paper used high order single-stepβmethod to calculate nonlinear structure dynamic response, because this method is more accuracy in ccounting, unconditional stable and unexceed phenomenon over other methods.(3) Because there are all kinds of noise disturbing during the real surveying, the paper added different standard noise when imitating the stimulation of environment, before the wavelet transform of the acceleration response the date should have been treated of noise reduction. For using wavelet to reduce noise, the paper made a detailed contrast-ing and analysis on the selection of wavelet function, decomposition of scale, the method of noise reduction and so on. using two index, that is, the ratio of signal and noise and the roof mean square error to judge the effection of noise reduction. Finally, put forward the technological process for reducing the noise:using the bior6.8 wavelet 2-scale rigrsure rules to estimate the threshold to pretreat the noise reduction, and have get satisfactive results.(4) In the supervision of the structural damage online and locating of the damage position, the selection of wavelet function is very important, it may influence the result of the damage detection directly, so this paper has made detailed contracting and analysis, through the acceleration response’s discrete decomposition to obtain the detail signal spike height to judge the ability of the wavelet’s function in damage detection by a lot of analysis, and decided to adopt Doubechies wavelet to monitor the structure and judge the damage position.(5) This paper considered the different structural condition and damage condition did a discrete wavelet transform to the interrelated node acceleration response, considing the wavelet functiln, load intensity, noise standard and damage degree etc. factors, put forward the method of using the signal spike heigh to define the position of damaget to decide.(6) In order to analyze the influence to the distinguishing ability of wavelet analysis in a whole, the paper made much analysis and contracting in different sides such as the standard of noise, load intensity and damage degree, and gained a conclusion that increasing of load intensity and damage degree will increase the distinguishing ability of wavelet, on the contrary, the increasing of the noise standard will make the useful signal covered by noise and decrease the distinguishing ability of wavelet, even made it undistinctive.(7) Paper put forward two step method for identifying the damage based on wavelet analysis and the BP neural net-work. Since the neural frequency of structural will decrease while the structure being damaged. Only by the change of natural frequency the damage can not be found obviously, the paper used an index related to the damage degrees, that is, the change rate of squared natural frequency, as the feature parameter inputing the BP net. Bulit BP neural net-work from different sides such as the happening of different damage situation in single supporter and different damage degree at the same time, then using the well trained net-work to judge the damage degree, the error of the result is small. This proved the feasibility of the usage of neural net-work in judging the damage degree.更多还原