Browsing by Author "Fu, Bowen"

Now showing 1 - 3 of 3
  • Thumbnail Image
    Item
    Application of a fully polynomial randomized approximation scheme (FPRAS) to infrastructure system reliability assessments
    (8/6/2017) Fu, Bowen; Dueñas-Osorio, Leonardo
    Networked systems make the reliability assessment of critical infrastructure computationally challenging given the combinatorial nature of system-level states. Several methods from numerical schemes to analytical approaches, such as Monte Carlo Simulation (MCS) and recursive decomposition algorithms (RDA), respectively, have been applied to this stochastic network problem. Despite progress over several decades, the problem remains open because of its intrinsic computational complexity. As the structural facilities of infrastructure systems continue to in terconnect in network forms, their study steers analysts to develop system reliability assessment methods based on graph theory and network science. A fully polynomial randomized approximation scheme (FPRAS) based on Karger’s graph contraction algorithm is an approximating method for reliability evaluation, which has a unique property rarely exploited in engineering reliability: that by performing a number of experiments in polynomial time (as a function of system size), it provides an a priori theoretical guarantee that the reliability estimate falls into the ϵ-neighborhood of its true value with (1−δ) confidence. We build upon the FPRAS ideas to develop an s-t reliability version that has practical appeal. Focusing on the relevant-cut enumeration stage of the FPRAS, we find correlations between the recurrence frequencies of links in minimum cuts within the randomization phase of the contraction algorithm, and typical network topological properties. We employ LASSO regression analysis to approximate the relationship between link recurrence frequencies and such topological metrics. With the topology-informed link recurrence frequencies, obtained at a much lower computational cost, we use a new biased contraction probability yielding 16.9% more distinct minimum cuts (MinCuts) than the original random contraction scheme. The biased contraction scheme proposed here can significantly improve the efficiency of reliability evaluation of networked infrastructure systems, while supporting infrastructure systems design, maintenance and restoration given its ability to offer error guarantees, which are ideal for future prescriptive guidelines in practice.
  • Thumbnail Image
    ItemEmbargo
    Network science algorithms for the reliability and resilience of engineered infrastructure networks
    (2023-08-11) Fu, Bowen; Dueñas-Osorio, Leonardo
    Critical infrastructure networks are essential engineered systems for modern society, encompassing power grids, telecommunication networks, transportation systems, and water distribution networks, among others. Despite their crucial role in maintaining community normalcy, critical infrastructure networks face a variety of threats, such as aging, extreme weather, intentional attacks, chronic hazards, and catastrophic disasters. Any damage to critical infrastructure networks may be significantly amplified by their interdependencies and the increasing interactions with users and operators, ultimately leading to serious economic losses and potential loss of life. As enhancing the reliability and resilience of critical infrastructure systems is a priority for governance and society’s wellbeing,, different fields of research and practice are needed to develop ideas, test them, and implement them. Traditionally, efforts to improve the reliability and resilience of engineered infrastructure systems have focused on individual components within the systems, such as retrofitting and hardening of bridges in highway networks or substations in power grids. However, infrastructure systems are interconnected in network form, and decisions for retrofit or hardening should consider the broader network perspective. Furthermore, infrastructure systems are interconnected among each other, forming a network of networks (NoNs) due to the existence of cross-network dependencies. Thus, it is necessary to investigate the reliability and resilience of infrastructure systems from a network-level perspective. Network science is a dynamic field that has developed sophisticated techniques and algorithms to investigate the properties of networks, structures in networks, and processes on networks. However, the development of advanced techniques inspired by network science to complex problems in engineered infrastructure systems is still limited. We leverage significant developments in algorithms and techniques from network science and understanding of infrastructure systems to benefit their reliability and resilience. Currently, exact reliability computation methods and optimization-based methods are among the most effective strategies for reliability and resilience management of infrastructure systems, due to their guarantees on the accuracy or optimality of the solution. However, the computational complexity of these algorithms for reliability and resilience, is usually beyond affordable. The conflict between desirable models and practical limitations calls for a rational framework which allows decision-makers to choose an effective alternative option when the optimal solution is not feasible. We address this tension with Simon’s theory of satisficing. By selecting a solution that is ‘good enough’ rather than optimal, satisficing theory is applied to the reliability and resilience of engineered infrastructure systems, on problems such as reliability estimators with guarantees on interval and confidence, and close-to-optimal solutions for optimization models, among others. Inspired by the satisficing theory, we build upon techniques and algorithms from network science and proposed algorithms on multiple computationally intractable problems related to the reliability and resilience of infrastructure systems. Funding allocation on retrofit of bridges in highway networks are usually heavily impacted by the influence of politics, which is at odds to the optimal solution from engineering perspective. Our proposed socio-technical ranking algorithm based on a framework inspired by the Katz centrality from network science can effectively integrate network topology, bridge vulnerability information and impact of politics, and provide a compromise between optimality and practicality. We also proposed a principled recovery algorithm based on network partitioning and percolation process in networks, to restore damaged infrastructure systems quickly with supply-demand balance, which is similar to solutions from mixed-integer optimization models. In addition, we introduce the dimension of resilience into the network dismantling problem in network science to identify a dismantling solution that not only breaks the network, but also delays its recovery. By developing an adversarial-dismantling-retrofit strategy, we reveal critical information for long-term resilience enhancement of infrastructure systems. Finally, we investigate the functional relationship between different components of a transportation network, by extracting its dynamical backbones from real-time traffic data, which supports congestion mitigation, traffic intervention and transportation network design. Overall, inspired by community reliability and resilience challenges, my research builds upon network science methods to develop novel algorithms that address resource allocation to bridges in transportation networks, resilience-based restoration using distributed percolation process, resilience-informed network dismantling algorithms and adversarial-dismantling-retrofit strategies, along with mechanisms of resilience for infrastructure systems via functional decompositions to support dynamics-based design and operation. All these algorithms are implementable in practice, unlike their parent methods which remain intractable for large infrastructure problems of today. By being integrated to the Interdependent Networked Community Resilience Modeling Environment (IN-CORE) platform developed by the National Institute of Standards and Technology (NIST), our proposed algorithms will be implemented to support community resilience. Furthermore, as network science is also at the intersection of physical systems and their information processing capabilities, rendering insights from this research useful for emerging developments in system automation, decentralized consensus, and the development of quantum algorithms implementable in noisy-intermediate scale quantum devices.
  • Thumbnail Image
    Item
    Resilience-informed infrastructure network dismantling
    (2022-09-13) Fu, Bowen; Dueñas-Osorio, Leonardo
    Large-scale networked infrastructure systems contribute significantly to modern society. Highly intra- and interconnnected systems enable communities to be more productive, at the expense of becoming more vulnerable to extreme events, cascading failures, and operational demands, including random failures and even targeted attacks. The resilience of infrastructure systems against common but random failure and rare but intentional attacks is critical for safe communities, as it covers multiple other types of contingencies in between. Network dismantling is a process to make the network dysfunctional by removi ng a fraction of components, which provides insights for robustness and resilience under many events, from common to rare. In particular, to protect networks from uncertain dismantling, we need to understand how to optimally fragment networks into small clusters by removing a fraction of their assets with minimal cost. Approximation methods are desirable because finding the optimal dismantling strategy is NP-hard, thus impractical on infrastructure networks. First attempts rely on iterative removal of the nodes with the highest adaptive importance, either from basic centralities, such as degree and betweeness, or from some more advanced metrics like collective influence. However, the additive nature of such methods fails to capture the synergistic nature of the dismantling problem. An algorithm connecting network dismantling problems with network decycling problems, identifies better the collective dismantling set. Other recent strategies add realism by adopting nonuniform node remo val costs, and applying a bisecting algorithm based on weighted spectral approximations iteratively. Despite these efforts, the combinatorial optimization nature of the network dismantling problem still requires global solutions, even if approximated. Additionally, the cost to remove components is the only factor considered in most previous methods. Network resilience, which can inform what to protect from dismantling to facilitate recovery, is seldom included as part of the cost. In this work, we propose a method employing Karger`s contraction algorithm and node-transferring heuristic optimization to approximate the optimal dismantling set, considering both component removal cost and network resilience after dismantling. The proposed method, resilDism, obtains good performance compared to state-of-the-art network dismantling methods, and provides valuable insights to guide network design and resilience enhancement in practice.