Development of a surrogate model for predicting fiber orientation in fiber-reinforced resin parts

Abstract

The authors developed a surrogate model that can output fiber orientation information equivalent to conventional resin flow analysis, targeting fiber-reinforced resin parts. In automotive crash safety simulations, it is important to accurately reproduce the deformation of resin parts. However, the strength characteristics of the parts differ greatly depending on the orientation of the blended fibers. Therefore, it is essential to include the influence of fiber orientation in the simulation. Conventional resin flow analysis takes a long time to analyze due to its complex operation, making it difficult to incorporate into a steady vehicle development cycle. To address this issue, the authors applied the pix2pix algorithm and developed a surrogate model that can quickly predict fiber orientation. This model reproduces the target parts with a collection of uniform rectangular parallelepipeds (voxels) to correspond to various shapes of fiber-reinforced resin parts, and reduces the impact of differences in element division for each part on learning. This method reduced the time required to obtain orientation information by 99%, and it can correspond to various shapes of resin parts.