Machine learning-based inverse design of auxetic metamaterial with zero Poisson's ratio

The inverse design from property to microstructure is more urgent in practical engineering than the regular design from microstructure to property. In this paper, a data-driven machine learning (ML) model based on the combination of artificial back-propagation neural network (BPNN) and genetic algorithm (GA) is developed for designing auxetic metamaterial with specific Poisson's ratio, i.e. zero Poisson's ratio. Different to topology optimization, the ML model can optimize auxetic metamaterials with higher computational efficiency, lower requirement of deep knowledge of mathematics and physical model. In the ML model, the data set prepared by solving a large number of regular design problems using finite element simulation are used to train the BPNN to establish the underlying mapping relationships from the microstructure parameters to the Poisson's ratio, and through which the GA optimization is conducted to globally seek optimal solution of the microstructure parameters related to the specific Poisson's ratio. The effectiveness of the ML model is demonstrated by comparing to the tensile experiment and the finite element simulation of the structure designed with the given prediction. The results show the ML-based method offers an efficient pathway to design the microstructure of auxetic metamaterials with arbitrary specific Poisson's ratio.

Automated summary

This paper presents a data-driven machine learning model that combines neural networks and genetic algorithms to achieve inverse design from property to microstructure, offering efficiency and accuracy in designing auxetic metamaterials with specific Poisson's ratio.