A multi-task learning model for fast prediction of mechanical behavior of UD-CFRP composites under transverse tension

Abstract

The weaker transverse mechanical response of unidirectional carbon fiber reinforced polymer (UD-CFRP) composites has raised a lot of attention and research. The single-task modeling strategies based on deep learning (DL) have achieved impressive results in terms of surrogating the finite element method to accelerate computation, but they do not sufficiently consider the data correlation between different tasks. In this study, a multi-task learning (MTL) model based on DL is developed for fast and accurate prediction of the transverse mechanical response of UD-CFRP, which includes crack morphology and stress–strain response. Different from traditional single-task learning (STL) models, MTL models can take into account the relevance of different tasks to achieve one model to predict multiple tasks simultaneously. The prediction results of the RVE model confirmed that the developed MTL model has better prediction accuracy and generalization compared to the STL model. Moreover, the model can perform inference for two subtasks simultaneously in milliseconds. The developed MTL model takes the image of the initial structure as input without any prior assumptions and simplifications. Therefore, it has good potential to be extended and transferred to other multitask scenarios in the field of composites.