REPRESENTATION OF HOMOGENIZED CROSS SECTIONS BY REDUCED MODELS AND MACHINE-LEARNING FOR MULTI-PHYSICS SIMULATION OF NUCLEAR REACTORS
This PhD program focuses on algorithms of machine-learning and data compression for model reduction, in order to optimize the reconstruction of few-group homogenized cross sections used for the computer simulation of nuclear reactors.
These data are fundamental for all coupled multi-physics calculations performed to study reactor design and safety. The use of data assimilation techniques will be considered to enhance the process of cross section preparation by lattice transport computer codes, thus allowing to properly address big data problems.
The improvements sought by the application of innovative mathematical methods aim at achieving a significant reduction in memory storage and in the overall computational time, still ensuring the best accuracy for the reconstructed cross sections.
Département de Modélisation des Systèmes et Structures
Service des Réacteurs et de Mathématiques Appliquées
Laboratoire de Protection d’Etudes et de Conception
Place: Saclay, France
Start date of the thesis: Sept/Oct 2021
Daniele Tomatis, Dr
e-mail : email@example.com
CEA Saclay, France
Sciences Mathématiques de Paris Centre
Proposal in PDF format, French and English descriptions (one after the other) available for download , HERE