基于LSTM深度学习网络的分布动载荷识别

Distributed dynamic load identification based on LSTM deep learning network

提出一种基于LSTM深度学习网络的分布动载荷识别新方法。首先,建立结构有限元模型并对载荷作用区域进行平面化和子区域网格划分,构建子区域上以形函数形式分布的动载荷和有限元模型节点动响应之间的传递关系,建立节点处应变动响应与对应子区域上分布动载荷的样本库;其次,利用Meyer小波对样本库中的时域样本进行特征提取,并基于LSTM深度学习网络训练子区域上分布动载荷与有限元模型节点应变动响应的传递关系;最后,开展了数值仿真研究,利用有限元模型仿真应变动响应识别了三维壁板结构表面的分布动载荷,验证了所提出方法的有效性。研究旨在为服役状态下壁板结构上动载荷环境预示提供技术支撑。

Here,a new method for identifying distributed dynamic load based on LSTM deep learning network was proposed.Firstly,a structural finite element(FE)model was established and planarization and sub-region meshing were performed on load action area.The transfer relation between dynamic load distributed in the form of shape functions on sub-regions and dynamic responses of FE model nodes was constructed.Sample database containing nodal strain dynamic responses of FE model and distributed dynamic load on the corresponding sub-regions was built.Secondly,Meyer wavelet was used to extract features from time-domain samples in sample database,and LSTM deep learning network was used to train the transfer relation between distributed dynamic load in sub-regions and nodal strain responses of FE model.Finally,numerical simulation study was performed,and simulated nodal strain dynamic responses of FE model were used to identify distributed dynamic load on a 3-D wall panel structure surface,and verify the effectiveness of the proposed method.It was shown that the study results can provide a technical support for predicting dynamic load environment on wall panel structures under service conditions.