Analysis of the strength-ductility balance of dual-phase steel using a combination of generative adversarial networks and finite element method

This study investigates the potential of generative adversarial networks (GANs) and finite element method (FEM) for assessing the tensile properties of dual-phase (DP) steels, which are essential for lightweight and crashresistant automotive applications. DP steels are renowned for their high strength and ductility, and their microstructure plays a crucial role in determining these tensile properties. GANs are a promising approach to learning and extracting the complex morphological features of real microstructure and precisely generating realistic images with diversity. This study employs GANs to produce synthetic microstructural images of DP steels, allowing the rapid generation of a diverse dataset. These GAN-generated microstructural images accurately reflect the metallurgical and morphological features of real DP steels. FEM simulations are then utilized to determine the corresponding tensile properties. The simulation results reveal that the martensite fraction is highly correlated with tensile strength, and uniform elongation depends on the microstructural morphology as well as the martensite fraction. Generally, GAN-generated microstructures exhibit properties similar to those of real microstructures, with some generated microstructures surpassing their real counterparts in strength-ductility products. This study emphasizes the critical role of microstructural morphology in determining tensile properties and demonstrates the potential of GANs and FEM for the efficient evaluation of the properties of DP steel and optimization of microstructures to improve performance. These insights offer valuable contributions to materials research in the automotive industry.