Machine Learning For The Analysis And Design Of Auxetic Metamaterials

As a new type of artificial functional materials,auxetic metamaterials can produce lateral expansion/contraction deformations different from conventional materials when they are axially stretched/compressed,so they have a negative Poisson’s ratio effect.Its mechanical properties are regulated by changing its microstructure design,thereby realizing a conventional forward design from structure to performance.But in practical engineering,the inverse design from performance to structure is often more urgent than the conventional forward design from structure to performance.Compared with the one-to-one conventional forward mapping design from structure to performance,inverse design is often a one-tomany inverse mapping process,that is,the preset target performance will correspond to multiple sets of microstructure parameter solutions,which is more difficult to achieve.At present,the mature and effective inverse design method of auxetic metamaterials is mainly topology optimization method,but its theory is complex,the process is cumbersome,and it has high requirements for designers’ professional skills in mathematics and physics.To this end,the thesis proposes a data-driven machine learning model based on the backpropagation of artificial neural network(Back-Propagation Neural Network,BPNN)and genetic algorithm(Genetic Algorithm,GA),which can achieve multiple Efficient design of auxetic metamaterials with design parameters and customized mechanical properties.Mechanical properties can be a single Poisson’s ratio or multiple performance targets such as Young’s modulus and Poisson’s ratio of the material.In the established machine learning model,the finite element simulation method is used to analyze a large number of conventional design problems to obtain a sample data set,so as to use the sample data to train the BPNN neural network and establish the mapping relationship between the microstructure parameters and the target mechanical properties.Then,through the optimization function of the genetic algorithm,the optimal solution of the microstructure parameters close to the target mechanical properties is globally searched.The validity and design accuracy of the auxetic metamaterial machine learning model is verified by tensile tests and finite element simulations of the designed microstructures.The results show that the machine learning approach provides an efficient way to design metamaterial with specified mechanical property parameters.This method not only avoids the tedious calculation formula derivation,but also reduces the depth requirements of mathematical and mechanical theoretical knowledge in conventional optimization design,and has strong designability and operability.Finally,this thesis extends the machine learning-based metamaterial design method to other novel auxetic metamaterial structures,which further verifies the advantages of the machine learning method.