Simulação do Módulo de Young de compósitos reforçados com fibra usando aprendizado de máquina

Abstract

Young's modulus, or modulus of elasticity, is a property that measures the proportionality relationship between stresses and strains of a material in an elastic regime. When dealing with a composite material - a material consisting of the combination of two different phases: matrix and reinforcement - its calculation is not always simple. For this reason, in this paper it is proposed the use of machine learning, specifically neural networks, to estimate Young's modulus. A well established method, the multiscale modeling, was used to generate training data for the neural networks. Simulations using the licensed software Multiscale Designer, resulted in 152 different Young's modules, cathegorized by fiber percentage, laminated structure and test type (compression or tension). To define the best possible architecture for the network, seven different configurations were tested by varying the number of layers, number of neurons and activation function. From the results it was possible to pinpoint the best network, and further compare its predictions with simulated and experimental data, resulting in relative errors that in overall do not exceed 20% The results from machine learning are also more computationally efficient, obtaining results in less than a second for several configurations simultaneously, which is an advantage when compared to licensed software, which takes on the order of minutes for a single material, and when compared to the experimental one, which requires manufacturing the material.