桥梁拉索腐蚀损伤声发射监测及模式识别

发布时间：2018-04-30 10:17

本文选题：腐蚀 + 声发射　；参考：《大连理工大学》2015年硕士论文

【摘要】：斜拉桥是现代大跨度桥梁的主要类型之一,拉索是斜拉桥的重要承重构件。由于其长期暴露在外界环境之中,容易遭受氯盐等腐蚀介质的侵蚀。长年累月,拉索结构不可避免地会产生一定程度的损伤,导致力学性能衰减,使用寿命也会逐年缩短。在极端情况下,拉索发生断裂导致桥梁坍塌,引起灾难性的事故。因此,对桥梁拉索的腐蚀损伤进行监测是非常有必要的。声发射技术作为一种具有高灵敏度的无损检测技术用于监测桥梁拉索腐蚀损伤被证明是一种有效的方法,本文通过用声发射技术对拉索的腐蚀损伤过程进行实时监测,然后借助于信号处理分析技术实现声发射源的识别,进而合理评价拉索的腐蚀损伤。本文主要研究内容如下：1.通过拉索加速腐蚀实验,获取腐蚀全过程声发射信号。根据声发射撞击数和计数与时间的相关图,得到了拉索腐蚀损伤演化规律。为了确定拉索腐蚀过程中不同的损伤源,发展了基于主成分分析算法(PCA)提取声发射监测数据的特征值,应用全局寻优聚类分析算法对监测获得的大量数据进行信息挖掘、统计分析归类,找出相似损伤阶段的特征规律并分析拉索在腐蚀过程中不同损伤阶段的声源特性。2.根据拉索腐蚀损伤不同声发射源的聚类分析结果,提取不同聚类类型的声发射监测信号进行波形分析。由于声发射信号的非平稳特性,本文通过对比现行的各种时频分析技术,选择效果最好的基于经验模态分解(EMD)的Hilbert-Huang变换(HHT),将其应用于声发射监测技术中。通过研究HHT算法,应用端点延拓方法解决EMD中的端点效应；采用基于约束样条插值的包络拟合算法处理传统EMD中应用三次样条函数拟合过程中产生的过冲和欠冲问题；针对EMD模态混迭效应,采用经过改进后的总体经验模态分解(EEMD)应对,最后分析拉索不同腐蚀损伤阶段时信号的Hilbert谱特征。得到拉索腐蚀不同损伤源下声发射信号时域特征及其不同频率带上的能量分布。3.为了将拉索腐蚀声发射源聚类分析可视化、直观表征。创新地将自组织特征映像神经网络(SOFM)技术应用到拉索腐蚀损伤的模式识别中,SOFM是基于无监督学习方法的一种竞争型神经网络,通过对输入模式进行自组织训练和判断,最终将数据分为不同的类型。通过对声发射监测数据的处理,自适应得将聚类结果直观地呈现在映射到二维输出层上,实现声发射不同声源的识别。
[Abstract]:Cable-stayed bridge is one of the main types of modern long-span bridges, and cable is an important bearing member of cable-stayed bridge. Because it is exposed to external environment for a long time, it is vulnerable to corrosion by chloride and other corrosive media. Over the years, the cable structure will inevitably produce a certain degree of damage, resulting in mechanical performance attenuation, and the service life will be shortened year by year. In extreme cases, cable breaks lead to bridge collapse, causing catastrophic accidents. Therefore, it is necessary to monitor the corrosion damage of bridge cables. Acoustic emission (AE) technology, as a highly sensitive nondestructive testing technique, has been proved to be an effective method for monitoring the corrosion damage of bridge cables. In this paper, acoustic emission technology is used to monitor the corrosion damage process of cables in real time. Then the acoustic emission source identification is realized by signal processing and analysis technology, and the corrosion damage of cable is evaluated reasonably. The main contents of this paper are as follows: 1. The acoustic emission signals of the whole corrosion process were obtained by the cable accelerated corrosion experiment. According to the correlation diagram of acoustic emission impact number and counting and time, the evolution law of cable corrosion damage is obtained. In order to determine the different damage sources in the process of cable corrosion, the principal component analysis (PCA) algorithm was developed to extract the eigenvalues of acoustic emission monitoring data, and the global optimization clustering analysis algorithm was used to mine a large number of monitoring data. By statistical analysis and classification, the characteristics of similar damage stages are found out and the sound source characteristics of cable in different damage stages are analyzed. 2. According to the cluster analysis results of different acoustic emission sources of cable corrosion damage, acoustic emission monitoring signals of different clustering types were extracted for waveform analysis. Due to the non-stationary characteristics of acoustic emission signals, by comparing various time-frequency analysis techniques, the best Hilbert-Huang transform based on empirical mode decomposition (EMD) is selected and applied to acoustic emission monitoring technology. By studying the HHT algorithm, the endpoint continuation method is applied to solve the endpoint effect in EMD, and the envelope fitting algorithm based on constrained spline interpolation is used to deal with the overshoot and underimpact problems in the traditional EMD with cubic spline function fitting. In view of the EMD mode mixing effect, the improved total empirical mode decomposition (EEMD) is adopted to deal with the problem. Finally, the Hilbert spectrum characteristics of the signals at different corrosion damage stages of the cables are analyzed. The time domain characteristics of acoustic emission signals and the energy distribution in different frequency bands of cable corrosion under different damage sources are obtained. In order to visualize and visualize the cluster analysis of cable corrosion acoustic emission source. The self-organizing feature map neural network (SOFM) is applied to the pattern recognition of cable corrosion damage. SOFM is a competitive neural network based on unsupervised learning method. Finally, the data is divided into different types. By processing the acoustic emission monitoring data, the clustering results are presented intuitively on the 2-D output layer to realize the recognition of different acoustic emission sources.
【学位授予单位】：大连理工大学
【学位级别】：硕士
【学位授予年份】：2015
【分类号】：U446;TP391.4

【参考文献】