Efficient Reliability Modelling & Analysis of Complex Systems with Application to Nuclear Power Plant Safety

Nuclear power may be our best chance at a permanent solution to the world's energy challenges, owing to its sustainability and environmental friendliness. However, it also poses a great risk to life, property, and the economy, given the possibility of severe accidents during its generation. These accidents are a result of the susceptibility of the generating plants to component failure, human error, extreme environmental events, targeted attacks, and natural disasters. Given the complexity and high interconnectivity of the systems in question, a small glitch, otherwise known as an initiating event, could cascade to catastrophic consequences. It is, therefore, vital that the vulnerability of a plant to these glitches and their ensuing consequences be ascertained, to ensure that the appropriate mitigating actions are taken. The reliability of a system is the likelihood that it survives a defined period and its availability is the likelihood of it being capable of performing its required functions on demand. These quantities are important to a nuclear power plant's safety because, a nuclear power plant by default is equipped with safety systems to inhibit the propagation of an initiating event. An accident ensues if the safety systems required to mitigate some initiating event are unavailable or incapacitated by the initiating event. It is, therefore, easy to see that the reliability, as well as the availability of these systems, shape the safety of the plant. These crucial quantities, currently, are estimated using legacy techniques like static fault and event tree analyses or their derivatives. Despite their popularity and widely acclaimed success, these legacy techniques lack the flexibility to implement fully the operational dynamics of the majority of systems. Most importantly, their ease of application deteriorates with increasing system size and complexity, such that the analyst is often forced to make unrealistic assumptions. These unrealistic assumptions sometimes compromise the accuracy of the results obtained and subsequently, the quality of the risk management decisions reached. Their inadequacy is often amplified if the system is composed of multi-state components or characterised by epistemic uncertainties, induced by vague or imprecise data. The ideal approach, therefore, should be sufficiently robust to not necessitate unrealistic assumptions but flexible enough to accommodate realistic system attributes, while guaranteeing accuracy. This dissertation provides a detailed account of a series of computationally efficient system reliability analysis techniques proposed to address the limitations of the existing probabilistic risk assessment approaches. The proposed techniques are based mainly, on an advanced hybrid event-driven Monte Carlo simulation technique that invokes load-flow principles to resolve, intuitively, the difficulties associated with the topological complexity of systems and the multi-state attributes of their components. In addition to their intuitiveness and relative completeness, a key advantage of the proposed techniques is their general applicability. They have been applied, for instance, to a variety of problems, ranging from the production availability of an offshore oil installation and the maintenance strategy optimization of the IEEE-24 bus test system to the probabilistic risk assessment of station blackout accidents at the Maanshan nuclear power plant in Taiwan. The proposed techniques, therefore, should influence robust decisions in the risk management of not only nuclear power plants but other critical systems as well. They have been incorporated into the open-source uncertainty quantification tool, OpenCossan, to render them readily available to industry and other researchers