Title: Predicting Tensor-Valued Anisotropy in Steel Fiber-Reinforced Concrete with a Graph Fourier Neural Operator

Presented by Guanming Zhang, Computing PhD student, Data Science emphasis

Abstract

Steel fiber-reinforced concrete exhibits nonlinear and directional mechanical behavior that arises from fiber orientation and mesoscale heterogeneity. This talk presents a Graph Fourier Neural Operator that maps mesostructural graphs and loading conditions to interpretable field-level nonlinear anisotropy indices. The model predicts qualitative measures of yield and post-yield behavior, stiffness degradation, and principal directions under physics-based constraints that preserve mechanical plausibility. Trained on mesoscale simulations covering a range of fiber contents and orientation distributions, the surrogate provides a fast and explainable means of evaluating nonlinear anisotropy for use in design-space exploration and performance assessment of steel fiber-reinforced concrete.

Advisor: Dr. Yang Lu

Committee Members: Dr. Eric Henderson, Dr. Jun Zhuang

External Examiner: Dr. Michal Kopera