Search

Maintenance of structures and infrastructures is of increasing importance in order to reach acceptable level of safety despite the unavoidable uncertainty, and the economic efforts have to be reasonable. These two goals represent competing objectives in an overall optimization of very complex system and structure, which involve significant uncertainties. In fact, all civil engineering structures and engineering systems are subjected to degradation by fatigue cracks and corrosion due to varying loads. When the cracks propagate or corrosion grows, the structural system accumulates damage thereby leading to serviceability loss and eventual collapse. These failures can be prevented by appropriate maintenance scheduling and repair, even in the presence of uncertainties of various nature and scale, leading to a reduction in fluctuations and changes of structural and environmental parameters and conditions in the models describing the processes involved in fatigue cracks and corrosion growth. Degradation models used to predict the future state of components often involve simplifications and assumptions to compensate a lack of data, imprecision and vagueness, which cannot be ignored. To overcome these issues, the imprecise probabilities framework and markovian approach are proposed for performing reliability analysis, decision-making, and risk-based design and maintenance. It is shown how these approaches can improve the current practise based on models: B31G, Modified B31G, DNV-101 and Shell-92 failure pressure models. The reliability assessment is performed by taking into account the simultaneous action of many natural and technological loads. These loads are random by nature and can be adequately described only by stochastic processes; which are not performed due to lack of valid calculation methods. This methodology has been applied to study the reliability of arctic pipeline infrastructure. Finally, a robust and efficient probabilistic framework for optimal inspection and maintenance schedule selection for corroded pipelines and fatigue cracks in bridges is presented. Optimal solution is obtained through only one reliability assessment removing huge computational cost of the reliability-base optimization approach and making the analysis of industrial size problem feasible.

Recent technological accidents, which resulted in severe material losses, multiple fatalities and environmental damage, were deeply associated with human errors. Direct human actions or flawed decision-making processes have been increasingly tied to devastating consequences, raising major concerns regarding industry's ability to control risks. The most common approach to estimate the probability of human errors and weigh their impact to the overall risk is the application of a suitable Human Reliability Analysis (HRA) technique. However, uncertainties associated with behavioural aspects of humans dealing with advanced technology in complex organisational arrangements turn this type of evaluation into a challenging task to perform, an issue that brings difficulties to ensure sound predictions for human actions when interfacing with complex systems. Consequently, the development of innovative strategies to overcome existing limitations to understand how these sociotechnical systems could fail is of paramount importance, particularly the intricate relationship between humans, technology and organisations. This PhD research project is devoted to approach this multidisciplinary problem in a systematic manner, providing means to recognise and tackle surrounding factors and tendencies that could lead to the manifestation of human errors, improving risk communication and decision making-processes and ultimately increasing confidence in safety studies. The initial part of this thesis comprises a large-scale analysis of human errors identified during major accidents in high-technology systems. Detailed accident accounts were collected from regulators, independent investigation panels, government bodies, insurance companies and industry experts. The raw data is then scrutinised and classified under a common framework, resulting in a novel and comprehensive major-accident dataset, the Multi-attribute Technological Accidents Dataset (MATA-D). The second stage applies advanced data analytic techniques to gain further insight into the conditions leading to the genesis and perpetuation of errors, essentially making use of cluster analysis and classification. The application of different clustering methods reveals common patterns among accidents, and the usage of an artificial neural network approach (self-organising maps) algorithm allows the translation of the multidimensional data into visual representations (2-D maps) of accidents' contributing factors. This stage generates appropriate information to increase the understanding of these sociotechnical systems, to overcome barriers to communicate risk and to enable a wide-ranging 'learning from accidents' process. The final part of the research project builds upon the self-organising maps algorithm output, focusing on a deeper interpretation of specific clusters to disclose strategies to minimise human factors weaknesses and reduce major accidents. An important practical implication suggested by the data analysis is that human errors, in most of the cases, constitute reasonable responses to disruptive transactions between the technology and the organisation, which impact human cognitive functions. Accordingly, the recognition that human errors are mistakenly seen as root-causes of major accidents and the examination of these interaction problems from a new perspective provided an effective way to recognise hazards and tackle major risks, delivering realistic proposals to improve design, decision-making processes and to build trust in safety assessments.

Complex systems and networks, such as grid systems and transportation networks, are backbones of our society, so performing RAMS (Reliability, Availability, Maintainability, and Safety) analysis on them is essential. The complex system consists of multiple component types, which is time consuming to analyse by using cut sets or system signatures methods. Analytical solutions (when available) are always preferable than simulation methods since the computational time is in general negligible. However, analytical solutions are not always available or are restricted to particular cases. For instance, if there exist imprecisions within the components' failure time distributions, or empirical distribution of components failure times are used, no analytical methods can be used without resorting to some degree of simplification or approximation. In real applications, there sometimes exist common cause failures within the complex systems, which make the components' independence assumption invalid. In this dissertation, the concept of survival signature is used for performing reliability analysis on complex systems and realistic networks with multiple types of components. It opens a new pathway for a structured approach with high computational efficiency based on a complete probabilistic description of the system. An efficient algorithm for evaluating the survival signature of a complex system bases on binary decision diagrams is introduced in the thesis. In addition, the proposed novel survival signature-based simulation techniques can be applied to any systems irrespectively of the probability distribution for the component failure time used. Hence, the advantage of the simulation methods compared to the analytical methods is not on the computational times of the analysis, but on the possibility to analyse any kind of systems without introducing simplifications or unjustified assumptions. The thesis extends survival signature analysis for application to repairable systems reliability as well as illustrates imprecise probability methods for modelling uncertainty in lifetime distribution specifications. Based on the above methodologies, this dissertation proposes applications for calculation of importance measures and performing sensitivity analysis. To be specific, the novel methodologies are based on the survival signature and allow to identify the most critical component or components set at different survival times of the system. The imprecision, which is caused by limited data or incomplete information on the system, is taken into consideration when performing a sensitivity analysis and calculating the component importance index. In order to modify the above methods to analyse systems with components that are subject to common cause failures, α-factor models are presented in this dissertation. The approaches are based on the survival signature and can be applied to complex systems with multiple component types. Furthermore, the imprecision and uncertainty within the α-factor parameters or component failure distribution parameters is considered as well. Numerical examples are presented in each chapter to show the applicability and efficiency of the proposed methodologies for reliability and sensitivity analysis on complex systems and networks with imprecise probability.

This thesis applies agency theory to design risk regulatory policies in the implementation of sustainable sea defences. The aim is to investigate methods to reduce the probability of an accident inflicted on the society by resolving a downstream and an upstream moral hazard problem existing in engineering problems with an emphasis on the construction of sustainable sea defences. The risk regulation aims at reducing the probability of damage cost associated with flooding by enhancing an operator's safety operational procedures. It is defined by three policies; a safety design, a transfer payment, and a fine. In the downstream moral hazard between an operator and a regulator, a risk neutral operator under full liability will implement a sea defence which is optimal from the regulator's point of view when a fine is set equal to damage cost without a cost for the regulator. However, if the operator has limited assets to cover the damage cost, a cap on the operator's fine is placed to take into account that an operator may default if damage costs are too high. Limiting the responsibility of the operator in case of an accident decreases the strength of the incentive mechanism leading to the implementation of a sea defence below the socially optimum. This second best height of the sea defence involves an informational cost to the society in the form of a liability rent to guarantee the participation of the operator in the engineering project. The more we make the operator liable for the damage cost, the higher the liability rent. The society faces a trade off between liability rent and residual risk. The higher the residual risk, the lower the liability rent. Unlike the limited liability case, an operator with averse attitude to uncertain payoffs will implement a sea defence higher than a risk neutral operator in order to reduce the weight on the upper extreme values of the tail of the fine. This is because the marginal fine decreases under risk aversion. Similarly to the limited liability case, the implementation of a second best height of the sea defence is not free for the society. The society will have to compensate the operator for participating in the project in the presence of uncertain payoffs. This compensation takes the form of a risk premium and is subject to the risk coefficient and the variance of the damage cost distribution. Due to the lack of regulatory quality and independence, a pro-industry regulator and a government may have conflicting interest about the fine cap to impose to the operator. While the regulator prefers a higher fine cap to induce a higher liability rent, the government seeks to impose a lower fine cap. The government faces a trade off problem between how much discretion should be given up and how much expert information should be used from the regulator. In the case when the government observes that any of the fine cap choices available to the regulator is higher than its ex-ante choice of the fine cap, no discretion will be granted because the cost to the society outweighs the benefits of using the regulator's expertise. Nevertheless, when the government observes that some of the fine cap choices available to the regulator are lower than its ex-ante choice of the fine cap some level of discretion can be granted. The limit on the fine caps that the regulator can announce is determined by how much the regulator is biased towards the interests of the nuclear industry. When the regulator's range of fine caps is optimally restricted, the objectives of the regulator and the operator will be aligned by encouraging the regulator to choose a fine cap that will not exceed the optimal fine cap of the government.