Establishing the interfacial microstructure-behavior relations in composites via stochastic morphology reconstruction and deep learning

Microstructure design of composite material has recently drawn great attention, driving both computational and manufacturing innovations. The preponderance of previous studies has focused on establishing the relationship between the spatial arrangement of the reinforcement phases and the corresponding mechanical behaviors. In this work, we focus on the subtle but important effects of the reinforcement/matrix interface of ceramic-matrix composite and propose a deep convolutional-neural-network-based model to probe the interfacial microstructure-behavior relation at the nanoscale. The contact-surface morphologies of the reinforcement and the matrix are modeled and quantified via the statistical correlation functions. We employ the stochastic morphology reconstruction technique to generate the representative interface elements (RIEs) of the reinforcement/matrix interface, with the consideration of designable thicknesses and point defects of the coating interphase. The shear behaviors of various RIEs are investigated by high-throughput molecular dynamics (MD) simulations performed under a series of temperatures. The interfacial element density fields and the shear behaviors acquired from the MD simulations are employed to train the proposed deep-learning model, which shows an overall accuracy of 97 % to predict the temperature-encoded shear behavior from the interfacial element density fields. The proposed model allows us to quantitatively understand the dependence of the interfacial shear behavior on the synergistic effect of temperature, contact-surface morphology, and configuration of the coating interphase. It can be employed for the computer-aided design and optimization of the composite`s interface.