基于神经网络技术的斜拉桥损伤分步识别方法  

Step-by-step Detection for Damage in Cable-stayed Bridges Based on Neural Network Technology

作　　者：赵玲[1] 李爱群[2] 

机构地区：[1]扬州大学建筑科学与工程学院,江苏扬州225009 [2]东南大学土木工程学院,江苏南京210096

出　　处：《公路交通科技》2009年第9期70-75,共6页Journal of Highway and Transportation Research and Development

基　　金：国家自然科学基金资助项目(50378017)

摘　　要：结合神经网络技术进行了斜拉桥损伤分步识别的系统性研究,提出了具体的斜拉桥损伤分步识别过程,给出了每一识别步骤中适当的损伤识别参数,可实现斜拉桥主要构件即拉索和主梁中损伤的有效识别。采用概率神经网络确定损伤构件的类型,采用径向基函数(RBF)网络实现损伤的定位定量分析。针对润扬大桥斜拉桥的损伤模拟分析表明:将测试数据进行平均计算可以大大降低噪声对于概率神经网络识别结果的影响;噪声水平对2个径向基函数网络的损伤位置和损伤程度的识别能力方面的影响较小。采用不同的神经网络分阶段实现大跨斜拉桥的损伤识别,不仅提高了损伤识别的效率和准确性,而且增强了损伤识别方法在实际结构中应用的可行性。Systematic research on the step-by-step detection for damage in cable-stayed bridge was made based on the neural network technology. The concrete procedure of step-by-step detection for damage in the cable-stayed bridge and the suitable parameters for damage identification in each step were proposed to achieve a resultful identification of damage in principal members, that is, cables and girder of cable-stayed bridge. The probabilistic neural network was adopted to determine the type of damaged elements, and the radial basis function （RBF） network was chosen to fulfill the localization and quantification of damage. Damage simulation analysis for the cable -stayed bridge of Runyang Bridge indicates that （1） the influence of noise on the identification results of the probabilistic neural network can be significantly decreased by averaging the test data; （2） the noise level slightly affects the ability of the two RBF networks to identify the damage location and extent. The step-by-step damage identification method for long-span cable-stayed bridge by use of different neural networks is beneficial to the improvement on both the efficiency and exactness of damage detection and the feasibility of applying the damage identification method in real structures.

关 键 词：桥梁工程 损伤分步识别 神经网络 斜拉桥 拉索损伤 主梁损伤 

分 类 号：U446[建筑科学—桥梁与隧道工程]