Committee:
Pingfeng Wang (Chair, Director)
Harrison Hyung Min Kim (Member)
Zahra Mohaghegh (Member)
Chenhui Shao (Member)
Xin Chen (Member)
Design & Manufacturing
Analysis for Resilience of Complex Engineering Systems: Operations and Designs
With the growth of complexity and extent, large-scale interconnected network systems, such asinfrastructure networks, become more vulnerable to external disturbances. Hence, managing potentialdisruptive events of an engineered system and therefore improving the system’s resilience is animportant yet challenging task. To ensure system resilience after the occurrence of failure events, thisthesis proposes mechanisms across different phases: design, operation, and failure recovery.
As for enhancing the designs, challenges have arisen due to the increasing scale of modern systemsand the complicated underlying physical constraints. To tackle these challenges, we present agenerative design method that utilizes graph learning algorithms. The generative design frameworkcontains a performance estimator and a candidate design generator. The generator can intelligentlymine good properties from existing systems and output new designs that meet predefined performancecriteria. While the estimator can efficiently predict the performance of the generated design for a fastiterative learning process.
Besides, during the operation stage, intentional islanding is commonly applied in practical applicationsand attracts abundant interest in the literature. Thus, we propose a hierarchical spectral clustering-based intentional islanding strategy for interconnected systems. And various system onlinemeasurements are used as embedded information in the clustering algorithm to enrich the modelingcapability of the proposed framework.
Moreover, we establish a mixed-integer linear programming (MILP) based failure recovery frameworkusing heterogeneous dispatchable agents. The scenario-based stochastic optimization (SO) techniqueis adopted to deal with the inherent uncertainties imposed on the recovery process from nature. And theCVaR risk measure is implemented because of the temporal sparsity of the decision-making. Theresulting restoration framework involves a large-scale MILP problem and thus an adequatedecomposition technique i.e. modified Lagrangian dual decomposition, is also employed to achievetractable computational complexity.
We consider case studies for complex engineering systems, such as synthetic supply chain networksand power systems from the IEEE dataset to illustrate the applicability of the proposed methods forimproving system resilience.
