Prediction of cure-induced deformation of carbon fiber reinforced polymer composites based on transfer learning

The cure-induced deformation (CID) significantly influences the final quality of carbon fiber reinforced polymer (CFRP) composites. It is essential to ensure the deformation within specification by cure optimisation, of which the key is to construct the efficient predictive model between the cure cycle and CID. Conventional numerical methods require wellmeshed grids and fine-grained time steps, resulting in an expensive computational cost. Furthermore, data-driven methods have shown great potential in building surrogate models to accelerate the massive forward evaluations. However, their remarkable performance depends on massive labelled data. Therefore, this paper proposes to anticipate the cure-induced deformation of carbon fiber reinforced polymer composites based on transfer learning to reduce the data requirement of data-driven methods. First, sufficient simulation data are gathered as the source domain and are utilized for training the source model. Then, a parameter-based transfer learning method is introduced to obtain the final target model by finetuning the source model utilizing a limited amount of target labelled data. Finally, the validation numerical experiment is conducted on a case of CFRP aircraft skin. The results show that the proposed method based on transfer learning can significantly reduce the amount of target labelled data by 90% compared to the original data-driven method.