| About this Abstract |
| Meeting |
MS&T23: Materials Science & Technology
|
| Symposium
|
Leveraging Integrated Computational Materials Engineering for High-fidelity Physics-based and Machine Learning Models
|
| Presentation Title |
Novel Convolutional-Recurrent Hybrid Neural Network for Predicting Fission Gas Release in UO2 Nuclear Fuel |
| Author(s) |
Peter Toma, Md Ali Muntaha, Joel B. Harley, Michael R. Tonks |
| On-Site Speaker (Planned) |
Peter Toma |
| Abstract Scope |
The mechanism of fission gas release (FGR) in UO2 nuclear fuel significantly impacts the longevity of nuclear fuel rods. Gaining insight into the fundamental mechanism and forecasting the behavior of fission gas transport and release is of paramount importance. Traditionally, the phase field method has been employed to compute FGR from a UO2 microstructure, but this method is highly computationally intensive and time-consuming. Neural networks, a machine learning architecture, provide an alternative means of determining both the FGR of a microstructure and its evolution over time. In this research, a hybrid convolutional-recurrent neural network was designed to forecast the development of a simulated microstructure and the subsequent FGR over time, utilizing training data from a multiphysics code that couples the phase field and cluster dynamics methods. This framework offers a novel way of simulating time-dependent physical processes with reduced computational expense. |