Congratulations to Miss Meiyu Liu and Mr. Deyong Sun, Ph.D. students of Professor Weizhao Zhang, on receiving the Award for Excellence at the 3rd National College Students Intelligent Terminal Simulation Technology Competition (第三屆全國大學生智能終端仿真技術大賽) held on 27 September 2025!
Project Title: Integrated multiscale and multiphysics modeling method for the manufacturing of woven CFRP based on LSTM-FEA
Project Description:
Both the pre-forming and curing processes of composite materials significantly affect the performance of components. After preforming, the fiber bundle angles, material thickness, and resin distribution all undergo reconfiguration; the curing process involves chemical-thermal-mechanical coupling and is affected by the anisotropy and inhomogeneity induced by preforming. Numerical simulation in describing these processes requires a large number of iterative updates of constitutive model parameters, which seriously impairs computational efficiency. To address this issue, this project establishes a method that integrates long short-term memory (LSTM) and finite element analysis (FEA) to enable rapid prediction of material constitutive behavior and accelerate the FEA simulation process.
1) Integrated mesoscopic modeling scheme of woven CFRPs during forming processes
This modeling scheme employs micro-CT-based geometry reconstruction, continuum elements, and a transversely isotropic hyperelastic constitutive model to simulate yarns as continuous bodies. Physically meaningful parameters are input to the constitutive model to elucidate the deformation mechanism. A finite element (FE) homogenization technique based on reaction force was also established to facilitate correct meso-to-macro transfer of material properties for multiscale simulation, as well as comparison with experimental data, for the fabric composites. Once completed, simulation results from this FEA modeling scheme were validated against a series of experiments typically utilized to characterize prepregs being formed, including uniaxial tension, bias-extension, and out-of-plane compaction. The validation demonstrates that this modeling scheme can accurately capture key 3D deformation of the woven composite prepregs at the mesoscale under various process conditions, providing a comprehensive tool to numerically identify forming behavior of the prepregs while minimizing the expensive experiments.
The simulation results for mesoscopic deformation modes of woven CFRPs.
2) LSTM-FEA integrated simulation framework
The framework utilizes mesoscale representative volume element (RVE) data for training to acquire material model parameters, enabling cross-scale generalization of geometric structures and boundary conditions. The trained LSTM model is converted to Fortran code and embedded into the user-defined subroutines, which supports the FEA solver to achieve multi-physics data exchange and iteration.
LSTM-FEA integrated simulation framework.
3) Validation using the single-dome composite part
The performing simulation acquires the yarn angle distribution of the woven composite in the open hemispherical structure after molding. The post-forming geometric model is imported into the curing simulation, where the integrated LSTM-FEA analysis is performed based on the curing process parameters.
LSTM-FEA simulation results for the single-dome composite part.
Miss Meiyu Liu and Mr. Deyong Sun were awarded the Award for Excellence at the competition.