Yaw-attitude regulation of compaction roller for suppressing out-of-plane wrinkling in thermoplastic automated fiber placement on three-dimensional curved surfaces

ABSTRACT Automatic fiber placement (AFP) is an advanced composite manufacturing technology with precise fiber steering and rapid in-situ forming capability. However, out-of-plane wrinkling caused by geometric incompatibility during curved-surface placement remains a major limitation. This study proposes an integrated optimization framework combining geometric contact mechanics, high-fidelity finite element simulation, and deep learning to investigate wrinkle suppression under curved-surface conditions. A finite element model incorporating the nonlinear shear behavior of thermoplastic prepregs and interfacial viscous contact was developed. A MeshGraphNets-based graph neural network was further established as a physics-aware surrogate model for the placement process. The results indicate that the roller yaw attitude reshapes the stress state in wrinkle-prone regions by regulating shear deformation and stress paths, while compaction force enhances interlaminar constraint and consolidation. Their synergistic effect suppresses fiber buckling and interrupts wrinkle formation. The surrogate model achieved a mean prediction error of 3.95% and improved computational efficiency by 1224 times, providing a rapid analysis method for wrinkle suppression in curved-surface AFP. This method broadens the wrinkle-free processing window and improves placement quality.