Inverse Critical Experiment Design via Gradient Optimization and a Multigroup Attention-Based Neural Network Architecture
Researchers have developed a new methodology for designing critical experiments for advanced nuclear reactors using deep learning and gradient optimization. A novel neural network architecture, featuring a multigroup attention pooling layer, was trained on simulation data to predict neutronic similarity. This approach allows for the direct optimization of experiment geometries to maximize similarity, achieving high correlation coefficients for specific transportation cask designs. AI
IMPACT This AI-driven design methodology could accelerate the development and validation of advanced nuclear technologies.