基于神经网络的翼盒结构响应重构方法应用

翼盒结构复杂，航行中承载条件恶劣，利用有限测点信息重构其它位置响应对于实时健康监测具有很强的现实意义。通过误差反向传播神经网络训练得到响应之间的非线性关系，建立基于神经网络的响应重构方法，开展有限元分析对其进行数值仿真验证，并将该方法应用于实测随机激励环境下翼盒典型承力结构的响应重构及损伤定位与判断分析。结果表明：采用该方法重构出的预测响应功率谱密度的均方根相对误差不超过1.90 dB，主要频点误差小于10%；判断出翼盒关键测点e 的损伤或故障发生在所截取片段数据3 s 后，其故障特征频率为240 Hz 左右，该方法应用于响应重构预示及健康监测分析具有可行性。

Application of response reconstruction method of wing box structure based on neural network

It is of great practical significance for real-time health monitoring to reconstruct other position responses by using limited measuring point information of wing box structure in complex navigation with harsh bearing conditions In this paper, the nonlinear relationship between the responses is obtained by training the back propagation neural network, and the response reconstruction method based on neural network is established and verified by numerical simulation by finite element analysis. Finally, the method is applied to the response reconstruction, damage location and judgment analysis of typical load-bearing structures of wing boxes under measured random excitation environment The results show that the RMS relative error of the predicted response power spectral density reconstructed by this method is less than 1.90 dB and the main frequency error is less than 10%; The damage or fault of the key measuring point E of the wing box occurred 3s after the intercepted fragment data, and its fault characteristic frequency was about 240Hz, which indicated the feasibility of applying this method to response reconstruction prediction and health monitoring analysis.