Thermal Load Feature Identification Of Honeycomb Sandwich Structure Based On Temperature Response And Deep Learning Technique

In recent years,with the continuous development of air and space technology and the increasing pursuit of limits in the aerospace field,the damage and destructive behavior of air and space device structures caused by thermal loads such as aerodynamic heat,radiant heat,and atmospheric re-entry heat have also attracted increasing attention,and the identification of their thermal load characteristic parameters has become particularly important.Thermal loads can cause degradation of material properties and generate thermal stresses,which in turn accelerate the rapid failure of structures.Extracting the thermal response of a structure to accurately invert the various characteristic parameters of unknown thermal loads is important for subsequent structural health assessment.Honeycomb sandwich structure,because of its high strength,high specific stiffness,light weight,good compression and bending resistance,heat insulation,sound insulation,impact resistance,fatigue resistance and many other excellent properties,the plate structure on the spacecraft widely used this structure,so this paper takes the honeycomb sandwich structure as the research object,for the honeycomb sandwich structure of thermal load parameter inversion identification problem,this paper proposes a spatio-temporal sequence based This paper proposes a thermal damage identification method based on the fusion of temperature field response and deep learning;introducing deep learning into the thermal load parameter inversion identification of honeycomb sandwich structures will greatly reduce the speed of inversion identification and reduce the difficulty.However,to apply deep learning to the thermal load parameter inversion recognition of honeycomb sandwich panels,two key core problems must be solved,one is the construction of the dataset and the other is how to select the network structure model suitable for the dataset.The thermal response data of laser load loading on honeycomb sandwich panels at different locations were first obtained through numerical simulations,and these spatio-temporal sequential temperature field data were simulated and extracted in batches through secondary development.Based on this data,datasets were produced and a variety of characteristic field data were obtained by data feature engineering using known thermal physical parameters,which can better characterize the damage response.From the obtained data,suitable deep learning network structures,such as ConvLSTM and LRCN,selected a suitable deep learning regression network structure is built for the obtained data,which can simultaneously identify the spot radius and power magnitude of the unknown laser loaded at any position on the target plate.By comparing the recognition performance of ConvLSTM and LRCN in the same network structure,the most suitable structure is selected,and a physically guided neural network model is proposed based on the real thermal property parameters of the honeycomb sandwich panel material,which enhances the physical interpretability of the network by adding a physical non-consistent term to the loss function in the neural network,and after introducing this term,The network has better recognition results.The optimal parameters are set by adjusting the hyperparameters in the network to make the best convergence and recognition performance,and various possible influencing factors,such as noise and different materials,are discussed to test whether the network has strong robustness and generalization performance.At the end of this paper,the practical feasibility and effectiveness of the method are verified through experiments,and the trained model can identify the unknown thermal load characteristic parameters more accurately.