Investigation and Tailoring of Rotating Squares' and Rectangles' Auxetic Structure Behavior through Computational...

Auxetic structures, renowned for their unique lateral expansion under longitudinal strain, have attracted significant research interest due to their extraordinary mechanical characteristics, such as enhanced toughness and shear resistance. This study provides a systematic exploration of these structures, constructed from rigid rotating square or rectangular unit cells. Incremental alterations were applied to key geometrical parameters, including the angle (θ) between connected units, the side length (a), the side width (b) of the rotating rigid unit, and the overlap distance (t). This resulted in a broad tunable range of negative Poisson's ratio values from -0.43 to -1.78. Through comprehensive three-dimensional finite-element analyses, the intricate relationships between the geometric variables and the resulting bulk Poisson's ratio of the modeled auxetic structure were elucidated. This analysis affirmed the auxetic behavior of all investigated samples, characterized by lateral expansion under tensile force. The study also revealed potential stress concentration points at interconnections between rotating units, which could impact the material's performance under high load conditions. A detailed investigation of various geometrical parameters yielded fifty unique samples, enabling in-depth observation of the impacts of geometric modifications on the overall behavior of the structures. Notably, an increase in the side width significantly enhanced the Poisson's ratio, while an increase in the overlap distance notably reduced it. The greatest observable change in the Poisson's ratio was a remarkable 202.8%, emphasizing the profound influence of geometric parameter manipulation. A cascaded forward propagation-backpropagation neural network model was deployed to determine the Poisson's ratio for auxetic structures, based on the geometric parameters and material properties of the structure. The model's architecture consisted of five layers with varying numbers of neurons. The model's validity was affirmed by comparing its predictions with FEA simulations, with the maximum error observed in the predicted Poisson's ratio being 8.62%.