Abstract Scope |
Accurately modeling the mechanical behavior of polymers requires computationally intensive, high-fidelity continuum damage mechanics kinematics. To address the significant computational demands and time-consuming nature, our research introduces a graph neural network-based surrogate model. This approach efficiently emulates finite element analysis of semicrystalline polymers, integrating complex mesh geometry and physical inputs—coordinates, stresses, displacements, material properties, and node connectivity. It not only enhances accuracy and reduces the amount of data required for training but also significantly accelerates the modeling process, which is particularly beneficial in industrial applications where time is critical. The model is trained on 500 simulations with specific material parameters controlling the general kinematics of elasticity, yield strength, and hardening in a continuum damage mechanics model. While the physics-based model offers more complexity and variability in material properties, this reduced-order model requires less data for training yet effectively proves the concept. |