Mechanical and Civil Engineering Seminar

Multiscale modeling aims to describe the behavior of a complex system by linking models of the system's response at relevant scales. The approach has been successful in many areas of science and engineering, for example materials science, chemical engineering or climate modeling, and has yielded many excellent models. Yet, the success of multiscale modeling is frequently hindered by the often-staggering computational cost of at-scale models. We start by discussing a computational framework for scale bridging in multiscale modeling, where at-scale models are replaced by surrogate models based on adaptive online Gaussian process regression. We assess the accuracy and computational cost of such surrogate models. Next, we present an alternate approach for constructing surrogate models with sparse Gaussian process regression. We demonstrate that both methods lead to significant reductions in computational cost when applied to a multiscale model of 1,3,5-trinitrohexahydro-s-triazine (RDX), in which a continuum finite element macroscale solver is coupled with a microscale dissipative particle dynamics model. In addition, we chart a path forward for a multiscale model of RDX incorporating both deformation and chemical decomposition. Finally, we enumerate some of the open problems in both scale bridging and surrogate modeling in the context of multiscale modeling.