结构损伤存在检测的两种神经网络的比较研究

Comparative study of two kinds of neural network used for structural damage occurrence detection

目前采用的新奇检测方法———自联想记忆神经网络方法,当所用的测试数据具有不同噪声水平或为非正态分布时,会得出错误的结果.为此,提出了一种新的方法———统计神经网络方法,用于结构的损伤存在检测,并用"可能性"来描述结构损伤的存在.通过一个两层框架的数值模拟和一个简支梁的实验数据进行对比性研究表明,统计神经网络可以用来检测结构的损伤存在,具有比自联想记忆神经网络更好的检测效果.

Novelty detection is an important means to detect structural damage occurrence. Nevertheless auto-associative memory neural network, the current novelty detection method, may give faulty information in structural novelty detection when the training and testing data are un-normally distributed or polluted by noise with different levels. Present a new approach called statistical neural network for structural damage detection, based on the neural network and statistical principle. The conception of possibility is applied in the sequence to represent the occurrence of structural damage. Both the numerical simulation results of a two-story steel frame and the experimental results of a simple beam prove that the performance of statistical neural network are better than the auto-associative memory neural network.