Search

• A neural-network-based framework is proposed for performance prediction and inverse design of a porous auxetic metamaterial. • The real-time prediction towards the auxeticity and stiffness can improve the inverse design. • The framework can be extended to structural material design with complex characteristics. Auxetic metamaterials are widely applied in energy-absorbing systems and soft robots due to their superior mechanical properties. The properties of auxetic metamaterials are determined by their architectures, which means it is feasible to obtain metamaterials with target performance through structural design. This paper proposes a framework with neural networks to generate a nearly real-time prediction of the auxeticity and stiffness simultaneously of porous materials. Trained neural networks accurately provide computationally inexpensive predictions on response histories. Based on the prediction database given by the neural network, the parametric analysis can be conducted to provide the mutual influence and trend relationship of auxeticity and stiffness of the porous metamaterial. In addition, inverse design can also be conducted for seeking optimal architecture in terms of the auxeticity and stiffness quickly and accurately. With the method, the target nonlinear properties can be accurately and rapidly designed for porous metamaterials. The relative error of neural network predictions is less than 1.5% compared with experimental tests, and less than 8.0% compared with the numerical simulations obtained by the finite element method. The proposed framework can be extended to more structural materials to provide design guideline. [Display omitted] [ABSTRACT FROM AUTHOR]