CGAN-Based Surrogate Model for Predicting Von Mises Stress Distribution in 3D Structures

Abstract

U-Net branches. Encoder features from the first two-time branches are forwarded and reused in the decoder of the third branch, promoting cross-layer feature sharing and more coherent stress patterns along the time-like axis. Numerical experiments show that the proposed surrogate reproduces the reference stress fields with good accuracy, yielding an MSE of 0.0011 and an MAE of 0.0138. The resulting cGAN-based surrogate provides a practical and computationally efficient tool for von Mises stress evaluation in complex 3D structures, suitable for integration into engineering design and analysis workflows.