基于有限元与RBF神经网络的结构健康监测研究

【作者】 潘瑞松；

【导师】 杜永峰； 李慧；

【作者基本信息】 兰州理工大学 ， 结构工程， 2005， 硕士

【摘要】 结构健康监测是当今土木工程领域的一个研究热点。其中结构损伤检测是结构健康监测的基础和核心,而振动模态分析技术和神经网络技术是解决这一问题的有效工具。 本文目的在于寻找适合的损伤诊断方法,对结构损伤进行精确定位和损伤程度识别。在结构动力信息的获取方面,本文介绍了环境脉动试验及其在结构动力特性中的应用,并对基于小波分析理论的时频分析方法进行了讨论。对于结构损伤识别,本文研究分析了包括损伤定位和损伤程度判断的结构损伤识别的实现过程,并且研究了应用模态频率进行损伤检测的方法,用悬臂梁的数值模拟进行了验证。借助有限元分析软件ADINA,对悬臂梁结构进行了有限元仿真分析。由模态分析获得模态参数,构造固有频率类损伤识别指标,利用仿真结果构造损伤样本,提出针对在损伤检测的不同阶段采用不同损伤指标分阶段识别的方法,采用MATLAB软件编制了RBF神经网络进行损伤定位和损伤程度识别。论文最后总结了基于结构动态特性分析的RBF神经网络结构损伤诊断方法,并指出了未来研究的发展方向。 本文同时在基于有限元分析的基础上进行了结构加固纠偏的研究。有限元分析结果表明,利用ADINA软件进行施工过程模拟,可以有效的预防施工过程中结构出现局部损伤。更多还原

【Abstract】 Research of structure health monitoring is one of the hot issues in current civil engineering. Among it, the research of structure damage diagnosis is the key point and difficult part of structure health monitoring. The technique of vibration modal analysis and artificial neural network is suitable to solve the problem.To accurately orient the damage location and identify the damage degree, an effective method of damage diagnosis is proposed. This paper introduced how to obtain the dynamic response of structure by environmental vibration experiment and the application of it, then discussed the method of time and frequency analysis based on the theory of wavelet. The identification of the structure damage, the process of structure damage orientation and identification of structure damage degree and the method using modal frequency to damage diagnosis are presented in this paper. A cantilever beam is chosen as the research object. The cantilever beam is simulated by using FEM analysis, the modal parameter is obtained by modal analysis, then constitute damage indices of natural frequency, so the damage mdices of different categories are made as the input parameters of RBF neural network to train the network and identify the damage. The procedure of RBF using MATLAB to orient the damage location and identify the damage degree is effective. A method that different index adopted in different damage diagnosis stage is proposed. At last the method of combination of structure dynamic characteristics analysis and neural network in structure damage diagnosis is summarized and the future research direction is presented.Lastly, using FEM method, a real damaged structure is taken as numerical example using ADINA, the studies of reinforcement and leaning rectification of the building are shown. The purpose of this simulation is to justify that through this method, the partial damage of structure can be prevented in the rectifying process.更多还原