Scalable criticality analysis in interdependent infrastructure systems using functionality graphs

Critical infrastructure systems (e.g., power, drinking water, wastewater, transportation, etc) have transitioned from isolated to highly interconnected, interdependent systems supporting essential services, economic productivity, and public safety. Strong interdependencies between these networks result in disruptions that no longer stay local to that system but propagate as cascading failures across infrastructures. Furthermore, community lifelines (e.g., hospitals, fire stations, police stations, etc) require several resources to operate and provide essential services (functionalities). These functionalities also have different impacts on the communities they serve. Therefore, identifying critical functionalities in interconnected infrastructure systems is crucial for downstream tasks such as resilience enhancement via asset hardening. Traditional modeling approaches often fail to capture functionality-level interactions and cannot quantify their community-level consequences. Therefore, this work introduces a scalable framework that integrates Hetero-Functional Graphs (HFGs) for community-centric modeling of interdependent infrastructure systems and Graph Neural Networks (GNNs) to identify critical functionalities. The HFG-based approach enables a richer representation by modeling system functionalities and dependencies as graphs while utilizing service and resource requirements as node features. GNN models trained on a seed set of nodes (using expert inputs) in the HFG can generalize and identify critical functionalities, crucial for downstream tasks such as resilience enhancement via asset hardening. Further, we provide an approach to assess the intra-asset type criticality using the Access Degradation Factor (ADF) metric. The proposed approach applied on data for a county in Western Kansas identified essential functionalities required to operate critical assets, which go beyond those identified by taking a single infrastructure perspective.