Generative reconstruction of photoelastic fringe patterns for transparent components using pressure-derived latent features

• A generative AI framework reconstructs photoelastic fringe patterns of transparent components. • Autoencoder-based latent optical features are predicted from internal pressure signatures. • The method generalizes to transparent materials, with injection molding used as a case study. • SSIM and stripe-shift metrics validate the reconstructed stress-induced optical fields. • Data-efficient training enables accurate fringe reconstruction with minimal experimental samples. Photoelastic fringe patterns serve as an important non-destructive indicator of residual stress in transparent components, yet conventional optical measurement systems are often costly, time-consuming, and unsuitable for high-throughput industrial environments. To address this limitation, this study proposes a generalizable generative learning framework capable of reconstructing photoelastic fringe patterns directly from internal pressure-related signatures. The method integrates autoencoder-based latent optical feature extraction with a multilayer perceptron regression model that maps pressure-derived indicators to fringe characteristics, followed by a decoder to reconstruct full fringe images. This framework is applicable to transparent polymer and glass components, whose internal stress states correlate with measurable pressure or force signatures; injection molding is adopted in this work as a representative case study due to its well-defined pressure fields and birefringence behavior.