Study On Impact Resistance Of Transparent Sandwich Structure Based On BP Neural Network

Transparent materials are widely used in special protection,traffic vehicles,optical devices and other fields with a certain stiffness and durability while providing good visual effects.Traditional transparent materials,such as glass and ceramics,are brittle materials.Once they fail,they will break quickly and produce sharp fragments,which can easily cause secondary damage.With the continuous development of composite materials,researchers are trying to combine a variety of transparent materials to prepare transparent sandwich structures with the advantages of multiple materials.Traditional composite structures follow the process of design-preparation-simulationexperiment-adjustment and optimization,and the research and development cycle is long,and it is difficult to give a rapid design scheme when facing the design requirements of multi-objective performance.With the continuous development of computing science,artificial neural network has been widely used in the field of composite material design.The prediction method based on pure data not only makes the previous experimental data can be used more deeply,but also avoids the complex nonlinear problems in the traditional simulation fitting process,and greatly shortens the development process.This article in view of the alumina ceramics as face layer,silica inorganic glass and polycarbonate organic glass as an energy absorption layer,through the transparent sandwich polyurethane cementation,for its shock resistance experiment and simulation research,and set up the BP neural network model,selection of structure layer thickness and projectile impact speed as input characteristic value,through the simulation data for training,The peak deflection of transparent sandwich structure under projectile impact load is predicted.The conclusions of relevant studies are as follows:(1)The first stage light gas gun was used to conduct projectile impact tests on transparent sandwich structure samples,and the dynamic crack propagation process was recorded by high-speed camera.Finally,two different failure modes are presented,which are dominated by ceramic layer bending failure and ceramic layer impact compression failure respectively.The ceramic layer remained intact under the dominant mode of bending failure,and the lower inorganic glass cracked at the four corners of the ceramic layer due to stress concentration.In the mode dominated by impact compression,the ceramic layer is broken and fails at the impact point,and the inorganic glass in the lower layer is broken and expanded from the center to the periphery.(2)Abaqus finite element software was used to simulate the projectile impact of 120,150 and 180m/s for transparent sandwich structures with multiple layers’ thickness ratio.For ceramic materials,a subroutine based on JH-2 constitutive model was introduced to simulate the process of crack propagation and debris splashing combined with element deletion method.The numerical simulation results are in good agreement with the experimental results.With the increase of the velocity of impact projectile,the ceramic fracture zone expands and the crack growth rate increases,and the cruciform crack takes shape first,followed by the X-type crack.A low stress region with radial distribution accompanied by energy release during crack growth can be observed from the stress cloud.The simulation of Cohesive contact based on interlayer bonding is also evident in the peak displacement curve of the back impact point of transparent sandwich structure.(3)BP neural network algorithm is used to conduct deep analysis of the simulation results,and a neural network model is established,which takes the thickness of three structural layers and the impact velocity of the projectile as the input characteristic values,and the deflection peak value at the impact point as the output,to predict the response of the structure under the impact load of the projectile.The hidden layer of the model contains six artificial neuron nodes.The performance of the artificial god external network model is compared by comparing the single hidden layer model with the double hidden layer model.The activation function of artificial neuron is "tansig",and the training function of model is "trainlm".The calculation time and accuracy of the model meet the requirements.The average calculation time of single-layer and multi-layer neural network models is 1 minute and 3minutes,and the average error rate is 7.6% and 3.2%,respectively.Visible with the increase of the number of hidden layer neural network model of the calculation accuracy is improved significantly,but the calculation is also a substantial increase,compared with the traditional spent many hours in terms of finite element calculation greatly saves time,and with the increase of data samples,the neural network model of computation efficiency and multi-objective mapping ability can also continue to improve,Better to provide guidance and help to the design and development of transparent sandwich structure.