Critical infrastructures (CIs) integrate physical assets and systems, providing key services to society. CIs are influenced by constant changes in technology, external threats, and operational contexts where human operators and end-users play a fundamental role. CIs can therefore be conceptualised as cyber-physical-social systems (CPSSs). The capacity of CIs to withstand or to recover quickly from impactful events is called resilience.
In a study published in the Journal of Safety Science and Resilience, an interdisciplinary team of researchers from the University of Salento and the University of Trieste (Italy) addresses a well-documented gap in the literature on CIs: the absence of standardised metrics for measuring resilience, particularly when such infrastructures are integrated with AI.
Principal investigator Alì Aghazadeh Ardebili explains, "Our systematic literature review yielded a first set of R-KPIs, categorised into specific theoretical foundations (resilience attributes, sustainability-based metrics, risk and reliability). We then applied a hybrid multi-criteria procedure, informed by twelve domain experts, to weigh the selection criteria and rank the identified R-KPIs."
Among the five selection criteria, "criticality" and "manageability" emerged as dominant, a result consistent with the high stakes of service continuity in CIs. The final ranking identifies the "probability of risk" as the most effective R-KPI, followed by "energy self-sufficiency" and "functionality loss". According to senior author Elio Padoano, the results suggest tha this ordering substantiates the centrality of risk-based and energy-autonomy considerations in AI-integrated infrastructures, while suggesting that minimum performance, although conceptually important, contributes comparatively less discriminatory power.
The proposed ten-step framework operationalises these indicators by guiding analysts through goal-setting, KPI typology, operational state, scale and stage of disturbance, monitoring, and iterative improvement. The authors demonstrate the applicability of the framework on an open dataset from a centrifugal water pump, where a support vector machine regression is used to fit the resilience curve following a disturbance on a specific day.
"The case study quantifies a recovery time of approximately 175.5 hours and zero energy self-sufficiency score, prompting concrete recommendations on redundancy, predictive maintenance, and decentralised energy provision," says Padoano.
Meanwhile, Ardebili is convinced that the study provides an advancement on knowledge about CI management in the presence of risk, because "the reader can gain a transferable methodology for resilience quantification, a defensible set of prioritised indicators, and an awareness of persisting gaps, notably the under-representation of the social dimension of CPSSs and the absence of standardised functionality-loss thresholds."
