矩形板结构损伤的分区域神经网络识别方法

通过引入LM优化算法，针对矩形薄板中对称结构的损伤识别问题，提出了一种基于神经网络的分区域分步识别方法. 对于预测输出量比较多且对预测精度要求比较高的问题，常会出现网络训练时收敛速度慢、网络预测精度低，并且当网络训练达到目标误差时，输出的预测量中常有某个输出量的误差还很大的情况. 针对这些问题，利用选取的组合输入参数，提出了基于神经网络的分区域识别方法. 通过对悬臂板结构的数值模拟结果表明：提出的分区域识别方法对结构损伤的分区和预测是可行和有效的，其预测精度要明显的高于只用单个网络的预测结果，并且预测子网络对损伤的位置和程度是同步输出的，从而避免了传统分步识别理论中子网络过多的问题.

Identification of damage in rectangular plates based on neural network technique with sub-regions

This paper presents an identification approach based on neural network method with sub-regions to identify damages in a rectangular plate using the LM optimized algorithm. The numerical results of simulations for the cantilever plate with some damages show that this approach is better than that from the conventional method with single network. Since this method realizes the synchronous output of position and intensity of damages in the structure, this approach does not need many sub-networks appeared in a conventional hierarchical identification method.