New AI-Aided Design Framework Creates Auxetic Meta-Laminae
MIE Associate Professor Yaning Li, her former postdoc Shujing Dong (now a professor at Shanghai Polytechnic University), and recent graduate Ammar Batwa, PhD’25, (now a professor at King Abdulaziz University), published their research on “Tuning Local Anisotropy for Macroscopic Auxeticity: Design Auxetic Meta-Lamina via Systematic Finite Element Simulations and Machine Learning Approach” in Materials & Design. Their research unveils an AI-aided design framework for meta-laminae, a breakthrough that enables the generation of an infinite variety of auxetic meta-laminae by systematically tuning local anisotropy.
Abstract:
This paper aims to design meta-lamina for desired properties including overall auxeticity via tuning local anisotropy of each discretized meta-patch. As an example, material system, meta-lamina with square patches is explored. By tuning the local anisotropy in each patch, desired overall elastic material constants, including the effective stiffnesses and effective Poisson’s ratios can be achieved. Interestingly, a large design pool for negative in-plane Poisson’s ratio are discovered and identified via systematic Finite Element (FE) simulations. This investigation delves into patch pattern-property relationships, machine learning methods, and inverse design approaches tailored for specific properties. Thorough investigations are conducted into the relationship between patch patterns, local anisotropy, and overall elastic properties, with concrete summaries provided for several typical patched metamaterial designs exhibiting varying properties. Four machine learning methods are used to train datasets from systematic finite element simulations. In addition, Bayesian optimization method is used to solve the inverse problem. Validation of the proposed designs is substantiated through finite element simulation, affirming the accuracy of both the design methodologies and the associated machine learning techniques. The design method proposed, alongside the developed machine learning techniques, offers an effective approach for the design of patch-wise meta-laminae.