Speaker
Description
To enable planning and dose assessment in Decommissioning and Dismantling (D&D) environments, we are developing a computational pipeline, based on Monte Carlo (MC) methods, through the application of Machine Learning (ML) techniques. The characterization of radiation fields is the first fundamental step in creating such pipeline.
In this work, we present a case study that demonstrates the feasibility of combining measurement-informed MC simulations with ML-based interpolation, to reconstruct 3D dose fields. The dataset used for this case study was collected in a temporary storage facility within SCK CEN, where a radioactive 200L waste drum was present. This study was structured in three phases.
The first phase concerned the measurements survey: dose rate measurements and gamma spectra have been collected, on a grid surrounding the drum. The dose rate and the gamma spectra were measured, respectively, with AccuRad PRD dosemeters (by Mirion) and CZT GR05 detector (by Kromek).
Afterwards, Monte Carlo simulations have been run, using PHITS particle transport code. A digital twin of the drum was made with MC methods using available information: involved isotopes and their activities, material composition, dimensions of the drum, and geometry of the facility. However, the source location and the material arrangement inside the drum were not known as a priori. Therefore, these parameters have been optimized for the best correspondence among simulated dose rate and measurements.
Lastly, interpolation algorithms were trained for predicting the 3D dose rate map of the facility, based both on simulations output and measurements. The efficacy of different ML regressor models, at different resolution levels, was tested for the general predictions of doses starting from measurements, or from measurement-informed MC simulations.
Eventually, the MC optimization provided deviations from measurements within 20%, in the best correspondence cases, and the ML-based interpolation techniques showed promising results.