基于振型差值曲率与神经网络的海洋平台结构损伤识别研究

摘要

利用神经网络进行复杂结构损伤识别时,会遇到网络规模过大以致难以收敛的问题.为此,提出了一种基于振型差值曲率与神经网络的结构损伤识别两步法.第一步,利用振型差值曲率得到损伤的大致区域.第二步,在第一步选定的区域内,利用BP神经网络确定损伤构件的准确位置.四层海洋平台数值模拟和实验结果验证了该方法的有效性.%When a neural network is used to identify damage of complex structures, a problem that the network size is too large to converge will be encountered. So, a two-step structural damage identification approach based on curvature of modal shape difference and BP neural network was proposed here. Firstly, the damage areas were localized approximately using curvature of modal shape difference. Secondly, the damaged elements were determined accurately with a BP neural network in the selected areas. The effectiveness of the method was validated with the results of numerical simulation and experiments for a four-floor offshore platform.