NRE 8011/8012 Seminar


High-Fidelity Reactor Physics Parameter Prediction Using Convolutional Neutral Networks


Dr. Justin Watson


University of Florida


Thursday, March 14, 2024 at 11:00:00 AM   


Boggs Building, Room 3-47


Fan Zhang


The use of machine learning algorithms and neural networks to model human behavior, predict physics parameters, and to mine data is an ever-growing field of research. We are at the forefront of Artificial Intelligence (AI) solutions and their application to a wide array of problems. They can be used to explore new design spaces that have never been investigated, used to discover new features from data, and accelerate computationally expensive simulations. The Florida Advanced Multiphysics Modeling and Simulation group (FAMMoS) is developing a novel Hierarchical, Flexible, Integrate Neural Network (HIFI-NN) for full-core, three-dimensional reactor analysis. The result will be a hybrid deterministic-AI system that will use cutting edge AI knowledge to predict high-fidelity reactor parameters in space and energy dependance. It will be applicable to a wide range of nuclear and non-nuclear problems. LatticeNet is a hierarchical deep learning architecture which has been developed for predicting pin powers and other parameters within single or multiple 2D pressurized water reactor assemblies. It has been shown to be effective at the task of predicting distributions of reactor parameters such as normalized pin powers under changing thermal hydraulics conditions. However, deep learning models are prone to overfitting and to learning rules in the dataset which are either not applicable to the real world or not what the researchers intended. When developing deep learning architectures for reactor modeling and simulation applications it is important to investigate how well these models perform on unseen data which lies outside of their training distribution. In this seminar I will present the LatticeNet architecture, demonstrate its strengths and weaknesses. I will present multiple thermal hydraulic datasets which are specifically tailored to provide challenging inference examples and evaluate these datasets using existing trained LatticeNet models in order to determine how well these models perform on classes of examples which are either statistically unlikely or impossible to be in the training data. I will show that these models exhibit surprising generalization capabilities for data outside of their training distribution, and moreover that the error of these examples is not entirely random but semi-continuous. I will also show that at least some variants of LatticeNet are particularly vulnerable to adversarial inputs which cause them to produce non-physical answers and demonstrate a simple method to detect these non-physical regions which requires no generation of new data.


Dr. Justin Watson is an Associate Professor of Nuclear Engineering, Department of Materials Science and Engineering, Herbert Wertheim College of Engineering, University of Florida. Dr. Watson received his Ph.D. in nuclear engineering from The Pennsylvania State University in 2010. Dr. Watson joined the University of Florida and founded the Florida Advanced Multiphysics Modeling and Simulation group (FAMMoS) in September of 2018. His research group focus on coupled time dependent space-kinetics/thermal hydraulics modeling of reactor systems for steady-state and transient analysis, nuclear reactor safety analysis, reactor kinetics and dynamics, Artificial Intelligence/Machine Learning, high performance computing and high throughput computing, and parallel computing software design. His current research portfolio consists of projects funded by industry, DOE, and NRC. Prior to joining the University of Florida, Dr. Watson was the head of the Computational Methods Development Department, Applied Research Laboratory, the Pennsylvania State University.


Meet the speaker