结构紧固连接件损伤的神经网络诊断研究

Detection of Fastener Failures in Structures Using Neural Networks

在结构紧固连接件损伤诊断中采用了函数连接型神经网络。与常用的ＢＰ网络不同，函数连接网络通过对输入模式的增强来实现复杂模式的分类。它借助某种函数变换，将原始的模式特征向量变换成另一个更高维的形式，使原始空间里不能区分的类别，在这个高维空间中得以区分，减少了神经网络的层数，提高了网络的学习速度。诊断时选用正弦激振下结构中多个位置上传感器稳态响应的幅值构成表征损伤的特征向量，简化了特征提取的过程。数值模拟和实验结果表明。

The application of neural networks in fastener damage detection in structures is investigated. The functionallink net is used which performs classification of complicated patterns by means of pattern enhancement. Compared with traditional BP net, the functionallink network can decrease the necessary layers of the net and thus increases the learning speed. In our research, amplitude responses of structures under the excitation of a sine function with fix frequency is adopted as the feature vector that reflects the structural damage. This approach simplifies the process of feature extraction. Numerical simulation and experimental results show that this method can identify damages in different sites correctly.