Adaptive momentum-based optimization to train deep neural network for simulating the static stability of the composite structure

Abstract

This article is the first attempt to employ deep learning to estimate the mechanical performance of multi-phase systems. Features of the design-points are obtained with the aid of the fast-converging numerical method used to solve the governing motion equations developed according to the kinematics of shear deformable structures. The optimum values of the parameters involved in the mechanism of the fully-connected neural network are determined through the momentum-based optimizer. The strength of the method applied in this survey comes from the high accuracy besides lower epochs needed to train the multi-layered network. It should be mentioned that the mechanical characteristics of the structure are computed through a two-step micromechanical scheme including the Halpin–Tsai method. The accuracy of the employed approach is examined and verified through the comparison of the results with those published in the literature. The numerical results give the practical hint that increasing the content of the reinforcement phase not always equal to increasing the resistance of the composite structure toward static instability. Thus, designers must choose the weight content of nano or macro-scale reinforcements by considering the shape factors of these materials to boost the strength of the system appropriately.