Towards a Neuro-Symbolic Approach to Risk Assessment in Railway Systems

Modern railway systems face increasing exposure to natural disasters, anthropogenic hazards, cyber threats, such as cyber attacks, and mechanical failures due to the growing complexity and interconnectivity of physical, digital, and operational system components. This paper presents a neuro-symbolic framework that integrates semantic technologies, such as those based on ontologies and knowledge graphs, with Graph Neural Networks (GNNs) to support multidimensional risk assessment. Unlike previous approaches that treat infrastructure components in isolation or focus primarily on delay prediction, a neuro-symbolic approach based on GGNs has the potential to capture interdependencies and simulate the cascading effects of heterogeneous threats through propagation rules and deep learning models trained on historical data. Our proposal presents the guidelines to build a risk assessment system to combine rule-based reasoning with deep learning to identify vulnerabilities, anticipate disruptions, and support decision-makers in enhancing system resilience. The description of a case study provides a scenario description that can be used as reference to assess the combined framework. Our proposal opens new avenues for predictive safety analytics in critical infrastructure and is designed with adaptability in mind, allowing extension into other networked domains such as maritime and aviation infrastructures.