【作者】 吴春辉；

【导师】 辛立民；

【作者基本信息】 安徽理工大学 ， 土木工程， 2007， 硕士

【摘要】 工程结构随着使用时间的延长,不可避免地发生老化,自然灾害的频繁发生也对其造成不同程度的损伤。结构损伤的检测及修复对于减少生命财产损失具有重要的作用。同时,尽早发现结构损伤,可以大大降低维修、维护的费用。因此,对工程结构的损伤识别、定位具有十分重要的意义。结构损伤检测技术已被广泛应用于航天、土木、机械和核工业中,是一门建立在损伤机理、传感器技术、信号分析技术、计算机技术及人工智能技术之上的多学科综合性技术。相对于传统的结构损伤检测方法,本论文主要对基于遗传神经网络的结构损伤检测技术理论与应用进行研究。本文通过理论分析了适合结构损伤位置和损伤程度识别的组合参数法(此组合参数是由固有频率的变化信息和少数选定点的模态分量合成的向量),在此理论的基础上,分别对一个框架结构和一个悬臂梁结构进行了损伤数值模拟,同时采取合适的方法构造改进型BP—GA神经网络的输入参数,应用训练后的神经网络对结构进行损伤检测。本论文的主要工作有下面几个内容:首先,通过对神经网络的工作原理进行分析,得出在理论上它能够对结构的损伤进行识别。利用模态参数进行结构破损诊断是国内外研究的热点和难点。本文提出了基于改进型BP神经网络进行结构破损诊断的方法。BP网络由于具有强大的映射能力、容错性和鲁棒性等优点,非常适合解决破损诊断这类问题。但随着研究的深入,BP网络在应用中遇到了两个主要问题:(1)难以确定网络结构和初始值;(2)易陷入局部最小解。针对BP网络的不足,本文提出了一种基于遗传算法GA—BP网络的混合技术进行结构破损诊断的方法。该方法采用实数编码的遗传算法优化BP网络的结构及初始参数,从而提高了网络的精度。对比遗传BP网络与普通BP网络对三个仿真算例的识别结果,遗传BP网络的稳定性更好,精度更高,对噪声有很强的鲁棒性,是一种准确有效的结构破损诊断方法。更多还原

【Abstract】 Along with the using time passing, the engineering structures inevitably get aged, and frequent natural disasters will also damage them in different degree. Detecting and repairing the damages of the structures plays a very important role in decreasing life and property loss. Meanwhile, early detection of the damages can reduce the expenses whether in repairing or maintaining. Therefore, the identity, location and estimation of the damages are of great importance. The structure damage detecting technology has been widely used in astronautics, civil engineering, mechanical engineering and nuclear industry, which is a multi-disciplinary and comprehensive technology based on damage mechanics, sensor technology, signal analysis technology, computer technology and artificial intelligence technology. Compared to traditional structure damage detecting methods, this article mainly focuses the theories and application of that based on genetic neural network.This article analyzes the combination parameter method which is suitable for the detection of location and degree of the damages. (The combination parameter is a vector consisting of changing information of the natural frequency and the modal components of some selected points) On this basis, a frame structure and a projecting beam structure are employed to receive damage numerical modeling, while some appropriate methods are adapted to construct input parameters of the improved GA—BP neural network, and the trained neural network is applied to carry damage detection to the structure.This article has the following main contents:First of all, as a result of analysis of operating principle of neural network, it’s clear that it can identify the damages of the structure.It’s a focus and difficult to identify the damages of the structure by modal parameters. This article put forward a method based on improved BP neural network on that topic. It’s the powerful mapping capacity, fault-tolerance and the robust that make BP neural network fit for dealing with damage identification. But with the proceeding of the study, there appear two main problems in applications of the neural network: (1) it’s hard to determine the initial value of the network structure; (2) it’s easy to trap in local minimal solution. For dealing with the shortages, a method based on GA-BP neural network which adapt the mixed technologies to detect the damages is put forward. Thismethod employs genetic algorithm of real coding to optimize the structure and initialparameters of the neural network and it improves the accuracy of the network.Compared to general BP neural network, the detecting results of the three emulatorexamples gained by genetic BP neural network owns better stability, higher accuracyand is much more robust, it is an accurate and effective way to detect the damages ofthe structure.更多还原