Generative Adversarial Network-based Modeling of Stress and Strain in Hierarchical Composite Materials

This study provides a framework that uses Generative Adversarial Networks (GANs) to describe stress and strain in hierarchical composite materials. Traditionally, techniques for modeling stress and strain, including finite-element analysis (FEA), are accurate and computationally demanding. The GAN model also models multi-scale interactions at micro, meso, and macro levels. It has been trained using simulated and experimental high-resolution datasets. The model achieved an accuracy of 89.97%. A recall of 88.76%. An inception score of 8.24. A Fréchet inception distance of 3.83. The results clearly show it can make predictions that are reliable and realistic. The cost-effective and scalable framework provides a superior alternative strategy to conventional prediction methods. As a result, this framework will ultimately transform materials performance analysis and reporting, as well as design materials and structures. The data diversity and reliability also present an advantage to overcome the shortcomings of today's modeling frameworks.