Discrete element method-generated data enhance CNN-based stress field prediction accuracy and efficiency in unidirectional fibre-reinforced polymer composites

Zhao, Siyu and Ding, Xiaoxuan and Ye, Jianqiao and Hou, Xiaonan (2026) Discrete element method-generated data enhance CNN-based stress field prediction accuracy and efficiency in unidirectional fibre-reinforced polymer composites. Applied Mathematical Modelling: 116962. ISSN 0307-904X (In Press)

Text (Revised Manuscript without Track Changes)
Revised_Manuscript_without_Track_Changes_.pdf - Accepted Version
Available under License Creative Commons Attribution.
Download (1MB)

Abstract

Accurate analysis of mechanical fields, such as stresses, strains, and deformation fields, is essential in the design and repair of composite materials. Compared to the time-consuming computational micromechanics methods, there is a high demand for developing a fast, efficient and accurate surrogate model to conduct mechanical fields analysis. This paper introduces a CNN-based U-Net model, mapping the stiffness distribution of unidirectional fibre-reinforced polymer composite laminae to their corresponding linear and nonlinear stress fields. The model is based on a dataset generated through discrete element method (DEM) simulations, which includes the simulation results from 50 representative volume elements with 60% fibre volume fraction and 50 with 40% fibre volume fraction for training, and 50 for each fibre volume fraction (60%, 50%, 40% and 30%) for testing. In comparison with existing studies, the proposed U-Net model, trained on DEM-generated data, demonstrates superior accuracy and efficiency in predicting both linear and nonlinear stress fields of laminae, including those with fibre volume fractions different from the training data, without requiring a large training dataset.