AI applications for Pebble Bed Nuclear Reactor Control and Fuel Management
Massimiliano Fratoni, Professor
Nuclear Engineering
Applications for Fall 2024 are closed for this project.
Pebble bed reactors (PBRs) use small spherical fuel elements that are piled in a cylindric core to achieve critical mass and energy production. The pebbles are circulated in and out of the core until the inner fuel is exhausted and at that point they are replaced with new pebbles. Such design greatly for A PBR core contains reduce the potential of unwanted power excursions and maximizes fuel utilization. Furthermore, the use of ceramic material for the pebble and molten salt or gas for cooling allows to achieve 25% higher efficiency in the heat to electricity conversion process.
Due to the high number of pebbles (hundreds of thousands) and to the different paths they can take while circulating through the reactor core, its internal states can be difficult to assess. For this purpose, computationally-expensive physically-based particle-tracking models are typically used. This project aims to use Data Science (DS) and Deep Learning (DL) techniques to emulate such models and provide an operational framework that informs operators for dynamically adjusting reactor parameters to meet required energy demands while maximizing efficiency and maintaining the reactor within the safe operational constraints.
In the first phase of this project, we successfully developed a proof-of-concept model to predict a few of the important reactor parameters based on the modeled isotopic concentration history of inserted fuel pebbles using a Long-Short-Term-Memory recurrent neural network (LSTM). The next phase of this project aims to further diminish the reliance on expensive models and more closely simulate operational conditions using inputs such as the detected energy spectra of the recirculating fuel pebbles. Further goals include predicting a more complete assessment of the reactor state, guiding their operation in real-time for optimal energy production.
Role: We are seeking motivated students with a DL/DS background willing to work collaboratively with a diverse team offering engineering and machine learning expertise to participate in model design and development, contributing to the design and implementation of safe clean energy production. Responsibilities will include data pre-processing, feature extraction, as well as neural network design and implementation. Students are expected to meet weekly with project supervisors and are highly encouraged to meet independently among each other for effective research collaboration.
Qualifications: Desired expertise includes familiarity with Python and Jupiter notebooks, as well as dedicated Machine Learning (ML) libraries such as PyTorch, Keras, and SkLearn. Familiarity with parallel-computing environments and other programming languages are a plus.
Day-to-day supervisor for this project: Ian Kolaja, Ph.D. candidate
Hours: 9-11 hrs
Engineering, Design & Technologies Digital Humanities and Data Science