Research On Shape Optimization And Mechanical Properties Of Mechanical Metamaterials Based On Deep Learning

Metamaterials are artificially designed composite structures that can achieve extraordinary physical properties that ordinary materials do not possess.As one of the branches,mechanical metamaterials can achieve extraordinary mechanical properties through predefined geometric configurations.Both the negative stiffness mechanical metamaterial and the negative Poisson’s ratio mechanical metamaterial have important applications in the field of energy absorption.The metamaterial structures with the two characteristics can be topologically combined to obtain a new metamaterial structure with composite characteristics.In this paper,a new mechanical metamaterial structure with both negative stiffness and negative Poisson ratio is proposed,by topologically combining the traditional negative stiffness curve beam structure and the traditional negative Poisson’s ratio arrow structure.The theoretical model of the cell structure is established,and the deformation of the structure under the compressive load is divided into two stages when the constraint condition formula is satisfied.The first stage is the negative stiffness of snap through deformation of the curved beam structure,and the second stage is the negative Poisson’s ratio deformation of the curved beam structure and the arrow structure.The deformation mechanism of the two-dimensional composite metamaterial structure is studied,and the influence of structural geometric parameters on the mechanical properties of the metamaterial structure is analyzed.During the deformation process of the structure,the occurrence of the compactness of the arrow structure is delayed,and the reaction force of the structure does not increase rapidly after all the curve beam structures are completely snaped.The number of vertical arrays controls whether there is a snap-back behavior after unloading at the end of stage-1 of the structure.The number of horizontal arrays controls the peak load of stage-1 of the structure.The geometric parameters of the cell structure can significantly affect the specific energy absorption of the structure,the percentage of the stage energy absorption,and the equivalent Poisson’s ratio,and it has good programmability.A deep learning-based full-process design scheme suitable for mechanical metamaterials is researched and developed,including the forward prediction of mechanical properties and the reverse design of geometric parameters.When building a forward prediction neural network model for mechanical properties,the model based on long short-term memory neural network has the highest prediction accuracy.When building a reverse design neural network model for geometric parameters,using a model based on a common fully connected neural network is sufficient to ensure accuracy.Based on the purpose of expanding the programmability range of the structure,on the one hand,the mechanical properties of stage-2 of the composite structure are improved by introducing elastoplastic materials,and on the other hand,the number of structural shrinkage directions is increased by improving the shape of the composite structure.After adding elastoplastic materials,the structural response remains basically unchanged in stage-1,and the bearing capacity of stage-2increases,but the phenomenon of negative Poisson’s ratio weakens.After the composite structure is improved,the structure array direction expands from two-dimensional to three-dimensional.Under vertical compressive load,the deformation process of the three-dimensional composite metamaterial structure is like that of the two-dimensional composite metamaterial structure,and the deformation is divided into two stages.