A framework to identify, characterize, and respond to interdependent infrastructure failures by integrating data-driven and physics-based approaches

Abstract

Interdependent infrastructure failure (IIF) refers to that the failure of one infrastructure system may trigger cascading impacts on other infrastructures, such as “Water pipe burst → Road damage”. Although IIFs have drawn considerable attention and efforts from researchers, there are still research gaps during identifying, characterizing, and responding to IIFs. The IIFs’ identification currently relies heavily on expertise-based methods whose processes may be subjective, biased, and error-prone, as the objective IIF evidence (e.g., official reports) is in shortage. The commonly-used IIF characterization paradigm ordinarily is applying a data-driven or physics-based method to uniformly characterize disparate infrastructures. Not only does this manipulation often encounter the challenge that certain infrastructures lack operation data or physical knowledge required by a predesignated method, but it also may significantly ignore the infrastructures’ heterogeneities regarding operation mechanisms, failure characteristics, and measurement indicators. The responses to IIFs are usually stymied by the weak perceptions of real-time infrastructure failures and conditions, and most response solutions provided by existing IIF studies are policy-level suggestions that are not immediately actionable in specific IIF incidents. In order to fill the gaps, an IIF-oriented identification-characterization-response (ICR) framework is developed in this thesis. An identification method is first proposed to harness published news articles as empirical evidence and combine text-mining with association rule learning for identifying IIFs, which could relieve the heavy reliance on expert knowledge. Then, an IIF characterization scheme is devised to avoid exploiting one uniform method to depict heterogeneous infrastructures, which includes understanding each target infrastructure, selecting applicable data-driven or physics-based methods respectively for different infrastructures, and designing interfaces to operationalize infrastructure interdependencies and connect selected methods. Finally, a response approach is developed to integrate crowdsourced social media, computer vision, and infrastructure-specific optimization with the characterization scheme predetermined in the second phase of ICR framework for prompt responses to IIF incidents, which shall make up the sparse coverage of physical sensors and help devise actionable solutions for consequence mitigations. For demonstrating the ICR framework, the case studies concerning water supply pipes and/or road transport systems are conducted throughout the identification-characterization-response pipeline. With the proposed news-based identification method, 18 IIF chains under water pipe bursts are identified, and “Water Pipe → Transport” is the most serious one. With the contrived characterization scheme, the failure patterns of water pipe bursts are extracting by news mining, while the performance deterioration of road transport is assessed by traffic simulations. With the developed response approach, the near-road infrastructure failures could be monitored through social media, and real-time parameters (e.g., traffic conditions) are captured by computer vision and injected into virtual traffic models to adaptively optimize signal timings. This thesis contributes to orchestrating data-driven methods (e.g., news mining and computer vision) and physics-based approaches (e.g., traffic model) for IIFs, which offers supplemental evidence for identifications, improves the flexibility of characterizations, and facilitates prompt responses. Beyond case infrastructures, the ICR framework holds the potential to be transferred to more infrastructures; this would help bring theoretical IIF management into practice and enable decision-makers to treat IIFs agilely.