A Hybrid and Adaptive Decision Support System for Protection from Radioactive Threats

Protecting society and the environment from radioactive threats is a multidimensional challenge that spans natural and artificial sources, medical and industrial applications, and long-term infrastructure monitoring. Modern radiological protection requires methodologies that can integrate heterogeneous information, adapt to evolving conditions, and combine physics-based understanding with data-driven intelligence. This thesis proposes a hybrid, adaptive, and heterogeneous framework for radioactive risk protection, structured around four foundational pillars: (P1) Monitoring, (P2) Modelling \& Assessment, (P3) Prediction, and (P4) Decision Support. The hybrid nature of the approach lies in the integration of formal modelling techniques (e.g., Bayesian networks, Petri nets, simulation-based methods) with data-driven and AI approaches for time series analysis and inference. Adaptivity is achieved through continuous recalibration of models and predictions as new data and environmental conditions emerge. Heterogeneity is explicitly addressed at multiple levels, including sensor technologies, data sources, physical processes, and decision-making contexts, enabling the framework to operate across diverse scenarios and scales. This general framework is applied to the domain of radioactive waste pre-disposal within the European PREDIS project. Here, custom low-power monitoring architectures (P1) are developed using scintillating fibre (SciFi) and solid-state neutron (SiLiF) detectors for long-term deployment in hostile environments. A modelling and assessment layer (P2) formalises source behaviour, sensor performance, and safety/sustainability metrics through scenario analysis and simulation. The prediction pillar (P3) incorporates simulation-based inference and machine learning to characterise and forecast process dynamics with quantified uncertainty. Finally, the decision support pillar (P4) integrates these elements within a novel hybrid architecture inspired by Kahneman’s Fast and Slow Thinking, combining expert rules with adaptive data-driven reasoning to support both immediate responses and strategic planning. Together, these contributions establish a unified, methodological and technological basis for modern radioactive protection, bridging the gap between real-time monitoring, predictive modelling, and operational decision-making in complex and heterogeneous scenarios.