PhD | Machine learning for accurate and efficient uncertainty quantification in radiological waste...

Quantifying the activity levels of activated and contaminated materials, in various radioactive waste package types and geometries is of paramount importance for the safe and effective decommissioning of nuclear installations. This PhD focusses on using probabilistic modeling, Bayesian data inversion and machine learning (ML) to develop automated codes for deriving the activity distribution of radionuclides in radioactive objects from radiological measurements, accounting for relevant sources of uncertainty. The latter include the spatial distribution of activities within a waste package, as this can have large effects on the overall measurement efficiency and the corresponding total activity estimates. The work will be primarily focused on inferring the gamma-emitting radionuclide inventory (in terms of total activity concentrations of a given object) using gamma spectrometry. Furthermore, this study will also attempt to develop a method for the Bayesian inference of the 3D spatial distribution of radionuclides within the considered objects using an enhanced gamma dataset (i.e., by using total gamma measurements and angular segmented gamma scanning) together with X-ray tomography (which can provide an estimate of the 3D density distribution based on the obtained linear attenuation coefficients). Lastly, the potential of completely bypassing the data inversion step by using ML to directly predict the total object’s activity from the measured raw gamma spectra will be explored as well.