【作者】 徐敏建；

【导师】 高维成；

【作者基本信息】 哈尔滨工业大学 ， 固体力学， 2007， 硕士

【摘要】 现代工程结构的大型化和复杂化,导致人们无法及时准确的了解结构的工作状况。自然灾害、结构的疲劳变形和自然环境的侵蚀都会导致工程结构的损伤。及时发现结构损伤,减少或避免人员伤亡和财产损失,显得尤为重要,这就需要一套结构损伤监测分析系统,来及时准确地反映结构的使用状况。本文以空间网格结构为研究对象,将整个结构损伤监测分析系统分为三个模块来研究,包括结构振动数据采集系统、结构模态参数识别系统和基于BP神经网络的结构损伤识别系统。首先,基于虚拟仪器技术LabVIEW软件平台开发了结构振动数据采集系统,实现了16个通道的实时采集、海量硬盘采集和触发采集等功能,并且设置了数字滤波功能,可以对采集数据进行滤波以减少噪声干扰,同时能较好地保证数据的完整性。然后,基于MATLAB的GUI语言开发了结构模态参数识别系统,实现了最小二乘复指数法(LSCE),特征系统实现法(ERA),有理多项式拟合法(RFPM)等参数识别功能,可以通过绘制稳态图剔除识别过程中出现的虚假模态,可以对振型结果进行三维动画显示,具有界面友好,操作简单等优点,交互式操作菜单可以引导使用者逐步完成模态参数识别的整个过程。以一桁架结构模型对其进行了仿真数据的验证,结果表明,本文所开发的模态参数识别系统识别结果可靠,快捷。最后,研究了基于BP神经网络的损伤识别算法,用共轭梯度法作为BP神经网络的训练函数,选择固有频率变化率和每个自由度在前10阶的模态分量作为BP神经网络的输入参数,并使用节点数作为输出神经元数量。采用一个桁架结构模型对其进行了仿真数据的验证,通过调整BP网络的输入和输出,实现对桁架模型的损伤位置和程度的识别仿真,结果表明本文所提损伤识别算法可以对空间网格结构进行损伤识别。本文开发的空间网格结构损伤监测分析系统,结合了LabVIEW和MATLAB各自的长处,具有采集振动信号完整,模态参数识别可靠,BP神经网络损伤识别到位的特点。更多还原

【Abstract】 People could not know the structure work condition accuracy and timely, because the Modern engineering structures are large and complicated. The structure maybe presents damage because such reason as natural calamity, tired accumulating and erosion. It’s import to find the structure damage timely for avoiding personnel’s injured and reducing property loss. And we need a structure damage monitoring and analysis system to reflect the structure work condition in time.The study object is spatial lattice structure in Paper. The whole spatial lattice structure health monitors and analysis system be consist of studying three parts. There is digital signal acquisition, modal identify, and damage identify by BP neural network. First, the structure vibration digital signal acquisition is developed with the virtual instrument technology of LabVIEW. It can acquire 16 channels digital signal at one time as real-time, saving signal data to hard disk and trigger. It have number filter set to reduce the noise influence, and keep the digital signal complete. Second, the modal parameter identify function be developed by GUI of MATLAB. It is consisted of LSCE, RFPM and ERA algorithms. Discard the false modal parameter by drawing steady-state figure. And there is a 3D drawing function to display the result of modal identifies. The modal identify function with friendly graphical user interface and simply operation; show the completion of operating step by step. The modal identify function is validated by a truss model simulation data. Come to the conclusion, the result data from the modal identify is reliable and quick. In the finally, do some study work of structure damage identify by BP neural network. Choose conjugate gradient algorithms as training function, choose the rate of change of natural frequency and modal vector of front 10 orders at a freedom degree as the input of BP neural network, and use the number of nodes of truss as the number of output. It’s validated by a truss model simulation also, and identify the damage position and extent by modulate the value of input and output. As a result, use BP neural network to identify damage is feasible.The spatial lattice structure damage monitoring and analysis system in the paper take the advantage of LabVIEW and MATLAB. The structure digital signal is complete, the modal identify result is reliable, and structure damage identify by BP neural network is feasible.更多还原