Modeling Fission Gas Release at the Mesoscale using Multiscale DenseNet Regression with Attention Mechanism and Inception Blocks

TL;DR

This paper introduces a deep learning approach using multiscale DenseNet regression with attention and inception blocks to accurately predict fission gas release flux from nuclear fuel microstructure images, significantly reducing computational costs.

Contribution

It develops a novel CNN architecture combining attention and inception mechanisms for mesoscale FGR prediction, demonstrating high accuracy and robustness on simulated data.

Findings

High predictive accuracy with R^2 above 98%

Best model achieves 4.4% mean absolute percentage error

Model maintains robustness on low FGR flux values

Abstract

Mesoscale simulations of fission gas release (FGR) in nuclear fuel provide a powerful tool for understanding how microstructure evolution impacts FGR, but they are computationally intensive. In this study, we present an alternate, data-driven approach, using deep learning to predict instantaneous FGR flux from 2D nuclear fuel microstructure images. Four convolutional neural network (CNN) architectures with multiscale regression are trained and evaluated on simulated FGR data generated using a hybrid phase field/cluster dynamics model. All four networks show high predictive power, with $R^{2}$ values above 98%. The best performing network combine a Convolutional Block Attention Module (CBAM) and InceptionNet mechanisms to provide superior accuracy (mean absolute percentage error of 4.4%), training stability, and robustness on very low instantaneous FGR flux values.