Deep Learning-based Stress Prediction for Short Fiber Reinforced Composites Using TabNet

Sammanfattning

This thesis investigates how deep learning models can predict stress uncertainties inshort fiber-reinforced composites (SFRCs). The central question is whether machinelearning models, particularly TabNet, can accurately predict stress variations withinSFRCs under different fiber distributions. To address this, full-field data were generatedusing Digimat and expanded through data augmentation to train the TabNetmodel. Model performance was evaluated using root mean square error (RMSE),and the results show that the TabNet model effectively predicts stress variationsacross different realizations of representative volume elements (RVEs). The modelcaptures the stress uncertainties arising from the microstructural variability of thematerial while maintaining high accuracy. This study demonstrates that combiningTabNet with data augmentation significantly reduces the computational resourcesrequired for traditional full-field simulations while providing accurate predictions ofstress uncertainties in SFRCs, highlighting its potential applications in compositematerial design and manufacturing.