Machine learning prediction of pebble history in pebble bed reactors

Ian Kolaja, Tatiana Siaraferas, Yves Robert, Jaewon Lee, Massimiliano Fratoni

Research output: Chapter in Book/Report/Conference proceedingConference contributionpeer-review

Abstract

The unique operation mode of pebble bed reactors (PBRs) makes traditional methods for measuring the core ineffective while requiring fast and accurate systems for assessing pebbles' history.This paper introduces a novel machine-learning approach to assess the pebbles' content and history during operation.The methodology incorporates data generation for individual pebbles using HxF, a hyper-fidelity depletion tool for pebble bed reactors capable of producing pebble-wise nuclide concentrations and pebble history parameters.These parameters include the number of passes through the core, the average radial path, fuel burnup, and fluence.The methodology also illustrates the process by which fuel pebble nuclide composition data can be transformed into simulated HPGe detected spectra using a detector response matrix constructed with Serpent.Machine learning models, including random forests and neural networks, are trained on the history parameters and concentrations to extract feature importance for detector energy channels and make more accurate predictions.The results show that these ML models demonstrate substantial improvements over linear regression, particularly for complex parameters determined on the last pass of the pebble through the core.This methodology paves the way for rapid and precise determination of fuel failure risk, pebble reinsertion decisions, and potentially core state characteristics, such as excess reactivity and flux profiles, crucial for PBR operation.

Original languageEnglish
Title of host publicationProceedings of the International Conference on Physics of Reactors, PHYSOR 2024
PublisherAmerican Nuclear Society
Pages1437-1446
Number of pages10
ISBN (Electronic)9780894487972
DOIs
StatePublished - 2024
Event2024 International Conference on Physics of Reactors, PHYSOR 2024 - San Francisco, United States
Duration: Apr 21 2024Apr 24 2024

Publication series

NameProceedings of the International Conference on Physics of Reactors, PHYSOR 2024

Conference

Conference2024 International Conference on Physics of Reactors, PHYSOR 2024
Country/TerritoryUnited States
CitySan Francisco
Period04/21/2404/24/24

Funding

This research uses the Savio computational cluster resource provided by the Berkeley Research Computing program at the University of California, Berkeley (supported by the UC Berkeley Chancellor, Vice Chancellor for Research and Chief Information Officer).

Keywords

  • burnup measurement
  • detector response matrix
  • Machine learning
  • pebble bed reactor
  • pebble history

Fingerprint

Dive into the research topics of 'Machine learning prediction of pebble history in pebble bed reactors'. Together they form a unique fingerprint.

Cite this