An efficient data-driven framework for inverse design of modified reentrant structure with segmented rods meeting extreme negative Poisson’s ratio
Abstract
Flexible devices usually require structures under tension capable of meeting demands of significant or zero lateral expansion in practical applications, where auxetic structures with giant negative Poisson’s ratio (NPR) and zero Poisson’s ratio (ZPR) provide promising solutions. However, highly-efficient customized inverse design of auxetic structures targeting specific mechanical responses remains challenging, resulting in a prominent research gap in the customized development of auxetic structures. In this study, a novel inverse design framework coupling a support vector machine (SVM) with a genetic algorithm (GA) is proposed for the accelerated development of segmented-rod modified reentrant structures (MRS) that satisfy extreme NPR and ZPR performance requirements. This inverse design method achieves optimized prediction accuracy and delivers more reliable structural inverse design results. First, the energy-based theoretical solution in Poisson’s ratio is derived and verified by comparing to the finite element (FE) solution. Then the GA-SVM algorithm is constructed and trained using the prepared dataset, which shows higher accurate and efficient prediction of Poisson’s ratio with R2 value of 0.999 87 than the other four common data-driven algorithms. Finally, an inverse design integrating GA and GA-SVM is realized to generate more optimal structural configurations according to the specific NPR and ZPR targets. The validity of generated optimal designs is experimentally examined. The results demonstrate that the present data-driven design framework is effective and robust for the modified reentrant design. The optimal NPR can reach −7, which is beyond the conventional Poisson’s ratio space. Also, the ZPR designs are created to meet the transverse non-deformation demand. This work offers an efficient machine learning framework tailored for the MRS inverse design, extending the exploitable space of auxetic structures.