A State-of-the-Art Review on Machine Learning-Based Multiscale Modeling, Simulation, Homogenization and Design of Materials

Multiscale simulation and homogenization of materials have become the major computational technology as well as engineering tools in material modeling and material design. However, the concurrent multiscale simulations require extensive computational resources, in which the CPU time increases exponentially as the spatial and temporal scale increase. In fact, with only a few exceptions, both hierarchical and concurrent multiscale modeling techniques have not been adopted in the industrial sector, primarily because of their computational cost. Recently, the rapid developments in artificial intelligence technology as well as the fast growth in computational resources and data have stimulated a widespread adoption of machine learning-based methodologies to enhance the computational efficiency and accuracy in multiscale simulations and their applications. Even though there is a great expectation of a revolution propelled by the artificial intelligence and machine learning technology in computational materials and computational mechanics, the machine learning-based multiscale modeling and simulation is still at its infant stage. In this paper, we aim at a state-of-the-art review on the machine learning-based multiscale modeling and simulation of materials, and its applications in composite homogenization, defect mechanics modeling, and material design, to provide an overview as well as perspectives on these innovative techniques, which may soon replace the conventional multiscale modeling methods.