Abstract
Bayesian network (BN) has emerged as a fundamental mathematical tool for implementing digital twins (DTs) in complex systems. Despite substantial progress in building DTs via BNs, challenges remain in storage and computational efficiency of BNs as complexity of the system scales. This study develops a scalable BN-based framework to customize sub-DTs from system digital twins (SDTs) based on specified performance measures, termed performance-oriented SDT. The SDT uses mutable knowledge graphs (KGs) to represent complex systems, enabling the capture of statistical correlations across different sub-systems while providing flexibility to model the dynamic evolution and transformation of infrastructure systems. To support event-responsive decision under any scenarios, a generic algorithm is proposed to create custom sub-DTs from KGs for selected performance measures and available data sources. For proof-of-concept, a hypothetical graph is first presented as a numerical example to demonstrate the versability and flexibility of the proposed model/algorithm for DT-supported management of infrastructure systems. Then model/algorithm is applied to the bridge network in Miami-Dade County: results show that a custom SDT with the size of 1% of the original KG can be created for practical use of the approach that significantly improves the computational efficiency in DT applications.