Predicting the effects of volume fraction on mechanical behaviour of the polymer composites

This study investigates the application of artificial intelligence (AI) to analyse the mechanical behaviour of short fibre reinforced polymer (FRP) composites, aimed on predicting stress-field distribution in particulate polymer composites (PPC). Traditional experimental and computational approaches often face challenges to accurately quantify and optimise the behaviour of such complex materials. This research utilises a paired image-to-image translation (Pix2Pix) deep learning algorithm, which employs a conditional generative adversarial network (cGAN) to predict stress patterns, particularly in the principal direction (S yy ) under tensile loading. Key input parameters for the AI model include fibre volume fraction, Young’s modulus, and time steps (frame numbers). The study models a 2D representative area element (RAE) for glass fibre-filled epoxy specimens using ABAQUS/CAE v2021. The corresponding stress patterns are used as reference for the AI model. The PPC specimens consist of rod shape glass fibres with an aspect ratio of 80, randomly oriented and positioned (at 1%, 2%, and 3% volume fractions) within epoxy matrices. Tensorflow, with GPU acceleration, is utilised to optimise model training and improve computational efficiency. A train-test split of 80:20 is used, and the model is trained with 2000 and 5000 iterations to analyse the impact of volume fraction on stress distribution. The stress field is partitioned into four distinct spectra based on the upper and lower bound stresses in the glass fibre and epoxy. This partitioning scheme improves the model’s predictive accuracy for stress distributions across varying material properties, fibre orientations, and volume fractions. The results are visualised through histograms of L2 norm and correlation metrics, with improved model performance observed as L2 norm values approach zero and correlation values approach one. The results demonstrate that the proposed approach provides a more efficient and less labour-intensive alternative to conventional experimental and numerical methods in analysing the mechanical behaviour of polymer composite materials.