Data-driven design of high ductile metamaterials under uniaxial tension

A data-driven quasi-disordered Face Cubic Centre (FCC) metamaterial design for improved ductility, based on a deep learning framework has been developed. The method relies on Artificial Neural Network (ANN) fundamentals to train a transfer function between input and output variables. A range of finite element (FE) calculations is used to generate macroscopic stress-strain responses of FCC metamaterials under uniaxial tension. It is found that FCC metamaterial lattices can enhance the ductility of the structure compared to FCC periodic lattices. The quasi-disordered FCC metamaterial design is achieved through distortion of coordinates and varying strut diameters in an FCC periodic lattice. The distorted coordinates and strut diameters are taken as input variables to an ANN which is trained to predict the stress responses of metamaterials and to optimize the lattice structure for high ductility. Two functions are used to optimize the structure: 1) ductility function and 2) strain energy density function. Both methods use a simulated annealing algorithm to optimize the coordinates and strut diameters. The quasi-disordered FCC metamaterials after optimization are validated by FE simulation which shows an increase in ductility of the structure of 30-40%. The ANN framework is computationally robust and can be extended to more complex geometries or multiple loading scenarios.