基于神经网络的复合材料应变场反演方法研究

Research on Neural Network Based Strain Field Inversion Method for Composite Materials

针对飞机蒙皮关键位置载荷预测困难,难以实现结构全局应变场监测的问题,文中提出了一种基于深度神经网络的应变场重构方法。针对飞机蒙皮的易损伤特征设计了含孔试验件,通过疲劳试验机和应变片搭建了应变采集系统。针对有限元模型和真实响应之间的误差问题,采用最小二乘法修正有限元分析结果。建立神经网络训练数据集,并且使用多层算法针对各数据集进行神经网络架构及参数优化,实现了结构到应变的精确等效。结果表明:应变场重构模型预测准确率在90%以上,实现了从局部有限测量点到整体结构的全局应变重构理论。

In order to solve the problem that it is difficult to predict the load at the key position of the aircraft skin,it is difficult to realize the global strain field monitoring of the structure.In this paper,a strain field reconstruction method based on deep neural network was proposed.According to the fragile characteristics of aircraft skin,a perforated test piece was designed,and a strain acquisition system was built by a fatigue testing machine and strain gauges.In order to solve the problem of error between the finite element model and the real response,the least squares method was used to correct the finite element analysis results.The neural network training dataset was established,and the multi-layer algorithm was used to optimize the neural network architecture and parameters for each dataset,which realized the exact equivalence from structure to strain.The results show that the prediction accuracy of the strain field reconstruction model is more than 90%.The global strain reconstruction theory from the local finite measurement point to the global structure is realized.