Systems today are becoming ever more complex with ever increasing mass of components and features organized in complicated structures. Therefore, notwithstanding the fact that many studies have already been conducted to enhance system reliability from different perspectives, it has become much harder to apply traditional analyses and optimization methods developed originally for simple structures or unitary characteristics. Clearly, there is an urgent need to come up with a systematic framework capable of supporting the emerging reliability-related solutions for complex systems. This study seeks to fill this gap with particular reference to the three types of systems outlined below.
General complex systems: These systems are usually multi-state systems and complex networks. The classification of states depends on the requirement of the system performance. Their structures can be represented in simple complex forms arranged in series, parallel, series-parallel, k-out-of-n or a mixture of networks. Complexity arises from their inherent properties or specific mechanisms such as for fault coverage, combating failures arising from isolated effects, and external threats, etc.
Control systems: The control systems are of growing importance in safety-critical applications. Examples can be found in nuclear power plants, chemical plants, smart grids and transportation systems. The need for feedback loops leads to unique structures enabling real-time system-state feedback. Such systems are highly dynamic and the states of their individual components are not easily observed. The reliability analysis of such systems is a difficult issue in view of considerations regarding time-dependence and the need to incorporate highly interactive innovations straddling many disciplines.
Smart grids: As future smart grids will employ more smart meters and communication networks, it is a typical application of Supervisory Control and Data Acquisition systems. Smart meters act as the controller and sensor. Communication network provides data exchange for real-time monitoring. It is also necessary to consider degradation originated from distribution generations. Therefore, it can be regard as an integration of networked control system and degraded distribution generations. Their complexity also arises from the uncertainties of distributed grid loads and capacities of power generations. The reliability framework should consider the interdependencies existing in different types of subsystems. A reliability assessment needs to be performed and energy management methodologies ensuring optimal power dispatch has to be developed in order to minimize operational costs and reduce possible power losses.
This dissertation is consisted of three parts corresponding to the reliability analyses addressing each of above three types of complex systems. We will also examine the relationships between the different types of complex systems.
A basic feature of general complex systems is that their structures usually include redundancy. However, due to imperfect fault coverage, the reliability of such a system cannot be enhanced indefinitely by adding redundancy. Therefore it is essential to determine the optimal structures when confronted with redundant systems. This study addresses multi-state series-parallel complex systems with two kinds of parallel features: redundancy and work sharing. The optimal trade-off between these two kinds of parallel features is determined to assure maximal system reliability.
Whereas most previous studies were directed towards combating internal failures, failures originating from external threats have started receiving more attention recently. The latter failures are very common in critical infrastructure systems such as web service systems, power networks and urban water systems. As an extension of redundant systems, this work investigates issue of optimal resource allocation between increasing protection of components and constructing redundant components in parallel systems subject to intentional threats. To evaluate the system ability to survive an attack, a vulnerability model is developed. The model takes into account uncertainties in the threats.
Because the goal is to encompass both internal and external failures, a model considering competing failure isolation and failure propagation with random propagation time is provided. The method proposed utilizes the total probability theorem and binary decision diagrams so as to come up with a practical procedure for extending reliability analyses to complex systems subject to competing failures and random propagation times.
Control systems have been recognized to be among the most important complex subsystems in view of their ability to undertake indispensable functions in many safety-critical applications. How traditional reliability assessment methods be modified to evaluate the dynamic properties of control systems are further investigated, from general control systems to specific control systems (networked and degraded control systems).
There is a need to modify traditional reliability assessment methods while coming up with solutions for general control systems. This study includes a modified Reliability Block Diagram method for evaluating the reliability of the power installation in NPP cooling system with multiple faults and dynamic states. The proposed framework is more computationally efficient than commonly used Markov methods in evaluating the reliability of a power installation without knowing all system states in advance.
Digital networked control systems represent an improvement over general control systems by making use of communication networks to enable data exchanges between controllers, actuators and sensors. Networked degradations such as transmission delays and packet dropouts cause such systems to fail to satisfy performance requirements, thus adversely affecting the overall reliability eventually. It is necessary to develop a model to evaluate the reliability in the phase of early design itself, prior to its implementation. However, existing probabilistic models are capable of providing only partial descriptions of coupled networks and control systems. This thesis proposes a new stochastic model which is in time-varying forms and takes into account data packet transmissions in all channels. The analysis of domain requirements proposed here for such systems represents a new contribution to the integration of control theory and reliability engineering.
Degradation is an important issue facing many a control system. The issue is particularly serious when many key components follow different degradation paths. Therefore, it is important to have an approach capable of correctly estimating the performance of control systems incorporating a variety of degraded components. From this perspective, there is a need to refine existing estimation approaches to endow the systems with the ability to cope with uncertainties and inadequate system specifications. To address this need, this thesis develops a hybrid model considering both a time-varying model of the control system and component degradation behaviors at different time slices. Having formulated the hybrid model, reliability is estimated through an event-based Monte Carlo Simulation that does not require prior knowledge of the exact system reliability function. This makes it possible to improve system reliability by optimizing the parameters of the control strategy on the basis of compensation against the losses in effectiveness caused by the degraded components.
Maintenance activities are also important in ensuring that the system continues to work properly in the presence of degraded components. Traditional maintenance models cannot be directly applied to degraded control systems in view of dynamic properties and the presence of feedback mechanisms. A performance-based maintenance model is proposed based on the real-time output performance of the degraded control system.
The question of how communication networks be integrated into a distributed power generation system is likely to be a major concern in future smart grid. Differently from previous research, which typically assumes perfect communication networks, we aim to quantitatively account for the impact of degraded communication networks on DG system performance. The degraded behavior of communication networks is modelled by stochastic continuous time transmission delays and packet dropouts. On the DG system side, we consider the inherent uncertainties of renewable energy sources, loads and energy prices. We develop a Monte Carlo simulation-optimal power flow (MCS-OPF) computational framework that is capable of generating consecutive time-dependent operating scenarios of the integrated system. Quantitative analysis is carried out to measure the impact of communication network degradation onto the DG system. For illustration, the framework is applied to a modified IEEE 13 nodes test feeder. The results show the power of the proposed framework.