Bayesian Network Methods for Modeling and Reliability Assessment of Infrastructure Systems

Bayesian Network Methods for Modeling and Reliability Assessment of Infrastructure Systemsby Iris TienDoctor of Philosophy in Civil and Environmental EngineeringUniversity of California, BerkeleyInfrastructure systems are essential for a functioning society. As these systems age, however, system reliability analyses are required to identify the critical components and make decisions regarding inspection, repair, and replacement to minimize the risk of system failure. In this study, we present novel Bayesian network (BN) methodologies for the modeling and reliability assessment of infrastructure systems. In an environment where information about a system is evolving and is oftentimes uncertain, BNs are able to both update the network when new information becomes available, and handle information probabilistically to support engineering decision making under conditions of uncertainty. One of the major limitations of the BN framework, however, is the size and complexity of the system that can be tractably modeled as a BN.In this study, we propose a novel compression algorithm that significantly reduces the memory storage requirements for a BN model, along with an inference algorithm that performs both forward and backward inference on the compressed matrices. We also present several heuristics to improve the computational efficiency of the algorithms. Through the application of these algorithms and heuristics to example systems, we show the proposed methodology to achieve significant gains in both memory storage and computation time. Together, these algorithms enable larger systems to be modeled as BNs for system reliability analysis.In addition, we propose a methodology based on the dynamic BN (DBN) to assess the response of a structure as it evolves through time under an excitation that is stochastic, e.g., an earthquake ground motion, based on sensor measurements that are uncertain. We look at the maximum response in particular, and derive an analytical solution for estimating the distribution of the peak response. In applying the proposed DBN framework to a multi-story shear-type building, we show the method to be robust relative to uncertainties in the structural characteristics, ground characteristics, and input motion parameters. This work informs decision making in the management of structures subject to seismic hazard and for the development and design of smart structural health monitoring systems