Before, during, and after... the characterization of the radiological state is essential at all stages of the decommissioning scenario of a nuclear facility. Can we intervene directly on-site, or is teleoperation necessary? Has the contamination of a given area been completely eliminated? How should we categorize a particular nuclear waste to optimize its future management?
In-situ non-destructive nuclear measurements aim to evaluate the radiological state of processes and equipment in real time, while meeting criteria of efficiency, safety, flexibility, and reliability, and reducing costs through rapid, precise, and non-invasive analyses. While characterization techniques for gamma emitters are well mastered, those for pure beta emitters remain a significant challenge due to the low range of beta radiation in matter and the ambient gamma noise, which makes in-situ detection particularly complex.
The integration of artificial intelligence (AI) tools, such as machine learning or deep learning, in this field opens new perspectives. These technologies enable the automation of the analysis of large amounts of data while extracting complex information that is often difficult to interpret manually, particularly for deconvoluting continuous beta radiation spectra. Initial results obtained in the framework of L. Fleres' thesis have shown that AI can effectively predict and quantify the beta-emitting radionuclides present in a mixture. Although promising, this approach, tested in laboratory conditions, still needs to be qualified in real-world field conditions.
The proposed thesis aims to continue and refine these developments. It will involve integrating new algorithms, exploring various neural network architectures, and enriching learning databases to improve the performance of current systems for the in-situ characterization of beta emitters. This will include scenarios where the beta/gamma signal-to-noise ratio is degraded, as well as the detection of low levels of activity in the presence of natural radioactivity. Other research avenues will include the detection of low-energy radionuclides and the adaptation of deconvolution tools for large-surface detectors.
The characterization methodology developed at the end of the project will have strong potential for industrial valorization, particularly in the fields of decontamination and decommissioning. The doctoral candidate will join a team with extensive experience in the implementation of non-destructive radiological characterization techniques and methods in-situ and will have the opportunity to evaluate the proposed solutions on some of the largest decommissioning projects in the world.

Desired Profile: The ideal candidate holds a degree from an engineering school or a Master's (M2) with solid knowledge of nuclear measurement, particularly regarding the physical phenomena related to the interactions of ionizing radiation with matter. Skills in statistical data processing methods and programming (Python, C++) would also be appreciated.