image: The AI-driven framework of optimizing the mechanical property of 2D-PHS.
Credit: This study was a joint effort between Professor Shengjie Ling’s team (currently at Fudan University, formerly at the School of Physical Science and Technology, ShanghaiTech University) and Dr. Yu Wang from Shihezi University. The first author is Yicheng Shan, and the corresponding authors are Dr. Yu Wang, Associate Professor Wenli Gao, and Professor Shengjie Ling.
Two-dimensional patterned hollow structures (2D-PHS) are an advanced class of metamaterials known for their unique mechanical properties and lightweight nature. Comprising a solid matrix with periodically arranged hollows, 2D-PHS effectively reduce material weight while optimizing stress and strain distribution to maintain structural integrity and strength. This precise geometric control offers superior tunability in strength-to-weight ratios, deformability, and stretchability compared to traditional solid materials. These attributes make 2D-PHS particularly valuable in high-performance lightweight systems. In the aerospace industry, these materials could improve components like aircraft wings and fuselage panels by maintaining high strength with reduced weight. Additionally, their excellent fatigue resistance and energy dissipation capabilities make 2D-PHS ideal for applications subjected to repetitive or cyclic stresses, such as biological tissue engineering scaffolds and impact-resistant devices.
In this context, research team from ShanghaiTech University reported an AI-driven material design framework that combines experimental and computational methodologies to enhance two-dimensional patterned hollow structures (2D-PHS). The framework examines critical factors influencing mechanical properties, including the arrangement, size, and shape of hollow structures, and, by applying AI-driven strategies, the algorithm tailor these parameters to meet practical application needs. By integrating computational modeling with experimental testing, the study seeks to boost the mechanical performance of 2D-PHS and expand their use across various engineering domains. The framework resulted in a 4.3% improvement in average stress uniformity and a 23.1% reduction in maximum stress concentrations. The tensile strength of optimized samples increased from an initial average of 5.9 MPa to 6.6 MPa under 100% strain, demonstrating enhanced mechanical resilience.
Future research will aim to improve the model's generalization abilities, for instance by designing a universal neural network architectures or using transfer learning methods to lessen the reliance on large volumes of specific training data. Additionally, broadening the model to incorporate optimization parameters from various physical domains can increase the framework's versatility and usefulness. Moreover, the inclusion of nonlinear simulations and the execution of destructive experiments will be vital for thoroughly investigating the failure mechanisms of materials and structures under different loading and material scenarios. Moving forward, extending this framework to three-dimensional structures represents a promising direction, providing greater complexity and functionality tailored to a wide range of engineering applications.
The combined AI-driven system not only streamlines the material design process but also serves as a powerful tool for creating lightweight materials with customized mechanical properties, ideal for essential engineering application.
The combined AI-driven system not only streamlines the material design process but also serves as a powerful tool for creating lightweight materials with customized mechanical properties, ideal for essential engineering applications in the aerospace and automotive sectors.
The research has been recently published in the online edition of Materials Futures, a prominent international journal in the field of interdisciplinary materials science research.
Reference: Yicheng Shan, Leitao Cao, Yu Wang, Jing Ren, Chen Huang, Wenli Gao, Shengjie Ling. AI-Driven Generative and Reinforcement Learning for Mechanical Optimization of Two-Dimensional Patterned Hollow Structures[J]. Materials Futures. DOI: 10.1088/2752-5724/ade732