DEEP LEARNING FOR MODELING OF MULTISCALE FLOW PHENOMENA IN COMPOSITES MANUFACTURING

Manufacturing fiber-reinforced composites (FRCs) with fast and efficient resin infusion

processes, such as resin transfer molding (RTM), involves impregnating a fabric with a

polymer resin. Fabrics typically exhibit a multimodal pore distribution defined by the pores

between fibers (in a yarn) and the pores between yarns (in the fabrics or, more broadly, in the

preforms). This multimodal pore distribution and the relative orientation of individual layers

in a layup necessitate a complete understanding of multiscale flow phenomena to ensure high-

quality manufacturing in reasonable cycle times. In addition, the identification of the

permeability tensor is required for predicting the flow patterns and the fill times[1]. In

contrast, air entrapment due to the interplay of capillary and viscous effects at the flow front

needs to be well understood to predict the residual voids in composites and mitigate them [2].