【作者】 庄小燕；

【导师】 陈浩然；

【作者基本信息】 大连理工大学 ， 工程力学， 2005， 硕士

【摘要】 本文将神经网络应用于复合材料层合结构的分层损伤诊断研究。复合材料层合结构由于其特殊的性能特点,在使用中甚至在制造过程中都会出现各种缺陷和损伤,这些缺陷和损伤将导致结构在服役期承载能力弱化。本文首先对神经网络算法的基本理论及其相关的研究成果作了介绍,总结了几种典型的网络结构,以及各自的学习方法和训练仿真过程;然后,建立了含贯穿分层的复合材料层合板悬臂板的动力分析模型,同时根据耗散理论描述了结构的阻尼,应用Adams模型,得到了复合材料层合板耗散能表达式,通过典型实验证明了本文提出的有限元方法的可靠性;在此基础上,采用了大量的参数讨论,分析了分层长度和位置对结构的固有频率、振型以及模态阻尼值的影响,并形成了神经网络训练所需要的频率样本库和阻尼样本库;最后,应用Matlab的神经网络工具箱分别建立了人工神经网络,由仿真结果比较说明了作者所建立的BP网络是适宜于复合材料层合板结构的分层损伤诊断。更多还原

【Abstract】 A strategy for detecting delamination of composite laminates is proposed in this paper, which is based on dynamic finite element method in conjunction of Neural Networks scheme. Because of the special performance of composite laminates, various forms of damage will appear, this damage will cause the infirmness of the carrying capacity of the structure. Firstly the author introduce some basic theory and some interrelated conclusion of the Neural Networks scheme, summarize several typical structures of Neural Networks and their method of learning and their train and simulation process. Then a dynamic finite element method is developed for the composite laminates with through-width delamination, and the boundary condition of the composite laminates is one side clamped. And the model of damping is describe, which is based on the theory of the dissipated energy. From the typical examination it can be show that the dynamic finite element method in this paper is reliable. According to the discussion of the input parameter, and the analyse of influences of the length and location of delamination upon the natural frequency and natural damping , the stylebook for the training of the Neural Network is founded. At last back propagation (BP) neural network in the Toolbox of MATLAB software is established for training and testing the network. From the simulating results it can be shown that present strategy is efficient for detecting delamination of the composite laminates.更多还原