Deep neural network method for predicting the mechanical properties of composites

• Published in: Applied physics letters Vol. 115; no. 16
• Main Authors: Ye, Sang, Li, Bo, Li, Qunyang, Zhao, Hong-Ping, Feng, Xi-Qiao
• Format: Journal Article
• Language: English
• Published: Melville American Institute of Physics 14-10-2019
• Subjects: Applied physics; Artificial neural networks; Composite materials; Elastic properties; Finite element method; Inclusions; Machine learning; Mapping; Mechanical properties; Microstructure; Modulus of elasticity; Neural networks; Poisson's ratio
• ISSN: 0003-6951; 1077-3118
• DOI: 10.1063/1.5124529

Determining the macroscopic mechanical properties of composites with complex microstructures is a key issue in many of their applications. In this Letter, a machine learning-based approach is proposed to predict the effective elastic properties of composites with arbitrary shapes and distributions of inclusions. Using several data sets generated from the finite element method, a convolutional neural network method is developed to predict the effective Young's modulus and Poisson's ratio of composites directly from a window of their microstructural image. Through numerical experiments, we demonstrate that the trained network can efficiently provide an accurate mapping between the effective mechanical property and the microstructures of composites with complex structures. This study paves a way for characterizing heterogeneous materials in big data-driven material design.