Accurate prediction of deformation in plastic components is essential for improving the reliability of automobile crash safety simulations. In fiber-reinforced plastic parts, mechanical strength and failure behavior depend strongly on the orientation of blended fibers, making fiber orientation a critical factor that must be properly considered in structural analyses. However, conventional resin flow analysis used to predict fiber orientation is computationally expensive and complex, which limits its applicability within the fast-paced vehicle development cycle where rapid design iterations are required. To address this challenge, we propose a surrogate modeling approach based on the Pix2Pix algorithm for rapid prediction of fiber orientation. The proposed method represents target components as collections of uniform rectangular voxels, enabling a consistent three-dimensional data representation regardless of part geometry. This voxel-based formulation allows the model to flexibly accommodate the wide variety of shapes found in automotive fiber-reinforced plastic components. In addition, the use of voxels significantly reduces the sensitivity of the learning process to variations in mesh discretization across different parts, which is a common issue in conventional finite element–based approaches. By minimizing the influence of mesh dependency, the proposed method improves the robustness and generalization performance of the surrogate model during training. Validation results demonstrate that the proposed approach reduces peak load prediction error by 8 percentage points (from −19% to +11%) compared with conventional isotropic material models, while drastically reducing computational cost. These results indicate that the proposed voxel-based Pix2Pix surrogate model is a promising solution for integrating fiber orientation prediction into practical automobile crash safety simulations, thereby enabling faster and more efficient vehicle development.
ITO et al. (Thu,) studied this question.