Efficient multiscale modeling of heterogeneous materials using deep neural networks

Abstract Material modeling using modern numerical methods accelerates the design process and reduces costs of developing new products. However, for multiscale heterogeneous materials, well-established homogenization techniques remain computationally expensive high accuracy levels. In this contribution, a machine learning approach, convolutional neural networks (CNNs), is proposed as efficient solution method that capable providing level accuracy. work, data-set used training process, well tests, consists artificial/real microstructural images (“input”). Whereas, output homogenized stress given representative volume element $$\mathcal {RVE}$$ RVE . The model performance demonstrated by means examples compared with traditional methods. As illustrate, in predicting stresses, along significant reduction computation time, were achieved developed CNN model.