Multiscale modeling and simulation of composite materials: Methods, AI integration, and challenges

The recent development of composite materials has presented an acute necessity for modelling and simulation architectures, which can represent the multiscale, multiphysics, and highly heterogeneous behaviour of composite materials. The review offers an overall and synthetic evaluation of modern modelling methods, including analytical expressions, numerical scales, atomistic and mesoscale models, macroscale structural modelling, and developing AI-based methods. This work compares and critically analyzes the fundamental assumptions, predictive ability, computation needs, and scale-specific constraints of these methods, unlike current reviews that consider these methods individually. The recent advances in the field of machine learning, physics-informed neural networks, and digital-twin technologies are given specific focus and are referred to as the factors that can help to increase the speed of homogenization, allow uncertainty-aware prediction, and provide real-time structural assessment. The review also reports significant research gaps, such as a lack of microstructure data, interface characterization, limited data for data-driven model generalizability, and a lack of a universal validation protocol that can allow the creation of models of entirely predictive composite simulations. This review provides strategic directions with regard to hybrid physics-data modelling, automated multiscale workflows, and predictive systems enabled by digital twins. On the whole, this piece of work presents a summarized, critical, and futuristic view toward the development of next-generation modelling and simulation technologies of high-performance composite materials.