Applications of Bayesian networks in complex system reliability

Estimating system reliability is an important and challenging problem for system engineers. System reliability may be defined as the probability that a system will perform its intended functions during a specified period of time under stated conditions. Current approaches for reliability analysis use specialized networks, each of which is designed for a specific system. This assumption of specialized networks presupposes that the BN is built by an expert who has "adequate" knowledge about the behavior of the system. However, finding a system expert may not be possible at times for every system under consideration. The number of system experts is limited and finding one is usually difficult and costly.;This dissertation introduces a methodology for reliability estimation, sensitivity analysis, and fault diagnosis in complex systems. A complex system is any system with a large number of interacting components and typically includes various subsystems. First, to solve the problem of reliability estimation in complex systems, the methodology introduced in this dissertation uses Bayesian networks (BN), which is a probabilistic approach that is used to model and predict the behavior of a system based on observed stochastic events. Second, this dissertation introduces a holistic method for automated construction of the BN model for estimating reliability in complex systems. Third, as suggested by the previous studies, this dissertation provides a new algorithm for sensitivity analysis of complex systems using BN. Fourth, this dissertation also introduces an algorithm for efficient fault diagnosis in complex systems using BN with heuristics to reduce the time to diagnose the unprecedented changes in the complex system reliability.;In addition to these, this dissertation presents a case study on estimating grid system reliability with the use of BN. Due to the size and complexity of grid systems, traditional methods for system reliability estimation cannot be used. In this case study, the methodology for reliability estimation is applied to the grid systems and effectively used in estimating grid systems' reliability. In order to validate the results and demonstrate the performance of these new methodologies, this dissertation presents a real-life application.