Development of Machine Learning Models for Critical Heat Flux predictions H/F

Description de l'unité

The internship will be conducted at the Department of Thermal-hydraulics and Fluid dynamics (STMF – Service de Thermohydraulique et de Mécanique de Fluide) within the Energy Division (DES - Direction des Énergies). The department develops and qualifies the simulation tools for the design and the safety analysis of the French nuclear power systems, such as the system thermal-hydraulic code CATHARE. These studies are performed within the framework of internal CEA projects or in collaboration with several industrial partners (e.g., EDF, Framatome and TechnicAtome) and international partners. The internship is in collaboration with the Department of Software Engineering and for Simulation (SGLS - Service de Génie Logiciel pour la Simulation).

Subject

Development of Machine Learning Models for Critical Heat Flux predictions

Job description

The Critical Heat Flux (CHF) is a physical phenomenon that may cause the deterioration of the heat transfer in the core of nuclear reactors, potentially leading to core damage. Its accurate prediction is therefore a crucial issue in nuclear reactor safety. The internship deals with the prediction of the CHF through the development of machine learning techniques and the quantification of the associated uncertainty. It will be realized in the framework of an international OECD/NEA project on the development and application of artificial intelligence methodologies for nuclear engineering. Within this framework, the intern will:

- Perform a preliminary literature review on the topic and get familiar with the previous works performed at CEA, based on experimental data in tubes with uniform axial heat flux profiles

- Analyse the new experimental CHF database (e.g., tube with non-uniform heat flux, rod-bundle, rectangular channel, …) to understand the physical phenomenon and to verify the coherence and exploitability of the experimental data.

- Perform an initial assessment of previously developed machine learning models to the new database. 

- Develop transfer learning to predict as accurately as possible the CHF for the new experiments.

- If time allows it, investigate the explainability issue and quantify the uncertainty of the models developed during the internship applying different state-of-the-art methodologies such as conformal prediction, hierarchical models and other coverage-driven techniques.

- Present and discuss the work during technical meetings and write a final report.

Applicant Profile

The ideal candidate is a MSc. student (in French system: M2 level or Ecole d’ingénieur) who has a good knowledge of artificial intelligence and neural networks with an interest in fluid mechanics. Knowledge of the physics processes underlying the CHF phenomenon will be considered a plus. Good skills in Python coding and code management are required. We welcome previous experience with any major deep learning framework (PyTorch, JAX, Tensorflow) and with uncertainty quantification approaches. Given the international context of the OECD/NEA project, fluency in English is considered necessary (French will be considered a plus).