基于特征向量简化的复杂土木结构损伤检测  

Damage detection of the complex civil engineering structures based on feature vectors reduction

作　　者：杨晓楠[1] 姜绍飞[1] 倪一清[2] 马传政[3] 

机构地区：[1]沈阳建筑工程学院土木系,辽宁沈阳110015 [2]香港理工大学土木与结构工程系 [3]沈阳建筑工程学院监理公司,辽宁沈阳110015

出　　处：《沈阳建筑工程学院学报（自然科学版）》2003年第2期81-84,共4页Journal of Shenyang Architectural and Civil Engineering University(Nature Science)

基　　金：国家十五科技攻关项目(2002BA806B-4);建设部科技项目;教育部留学基金;辽宁省自然科学基金;香港RGC项目

摘　　要：针对PNN在有噪声情况下进行结构损伤检测时输入矢量数目大,匹配模式多,训练时间长的缺点,提出了运用主组分分析法来处理数据的概率神经网络损伤定位方法.研究发现,运用该方法进行损伤定位,不仅输入矢量数目和训练时间少于未简化的PNN,而且可以极大地提高损伤定位的识别精度.The defect of huge redundant input data limits the capacity of probabilistic neural network (PNN) seriously when it is applied to the structural damage detection in noisy conditions. In this study, a principal component analysis (PCA) method is presented to deal with data for feature vectors reduction, and a PNN is applied to the damage location. The experimental result shows that not only the training time and the number of input vectors are less than those without vectors reduction, but also the identification accuarcy (IA) for damage location is greatly increased.

关 键 词：土木工程结构 损伤定位 特征向量简化 主组分分析 神经网络 结构损伤检测 

分 类 号：TU317[建筑科学—结构工程]