1 1.3. Research gaps 17 from failure datasets in a robust and scalable manner; Part II—with the main concepts discussed in Section 1.2.3—focuses on how and to what extent it is possible to accurately model Multi-State Deterioration with applications in sewer mains; and Part III—with the main concepts discussed in Section 1.2.4—explores how to devise near-optimal maintenance strategies for components with Multi-State Deterioration such as sewer mains using Deep Reinforcement Learning. Each problem and its associated research gaps are further detailed in the subsequent sections. Part I: Data-driven Inference of Fault Tree models One challenge in Fault Tree Analysis (FTA) is constructing accurate, e"cient, and reliable Fault Tree (FT) models. This process is referred to as construction, synthesis, or induction of FTs (Salem, Apostolakis, and Okrent, 1976; Hunt, Kelly, Mullhi, et al., 1993; Madden and Nolan, 1994). In this research, we refer to it as inference of FT models, which involves deducing the structure of an FT based on observed data, identifying relationships and dependencies among basic and intermediate events. Research in this area is expanding, driven by the principles of Industry 4.0 (Oztemel and Gursev, 2020) and 5.0 (Maddikunta, Pham, B, et al., 2022), where modern societies are increasingly enhancing processes and automation, partly by leveraging large, structured datasets. The inference of FT models has been explored since the 1970s through three main approaches: knowledge-based, model-based, and data-driven. These methods are discussed in detail in Section I.3, with an overview provided below. Knowledge-based approaches primarily rely on di!erent heuristics for knowledge representation and domain expertise. While they were among the earliest methods explored, these approaches are constrained by the experts’ knowledge, which may be biased, incomplete, or unable to capture unseen relations in FT models. Modelbased approaches involve translating existing system models into FTs, utilising frameworks like Simulink (Karris, 2006) and SysML (Friedenthal, A. Moore, and Steiner, 2015), but they require pre-existing system models. Data-driven approaches have gained prominence with the growth in data collection, automatically generating FTs by analysing structured datasets. In Part I of this dissertation, we focus on data-driven approaches for the inference of FT models, as these methods may require minimal domain expertise and reduce human intervention in identifying causal relationships within the data. However, we identify challenges in terms of scalability, robustness and completeness. Scalability is the ability to maintain e"ciency and performance as the dataset size increases. In our context, it denotes the algorithm’s capacity to infer a FT structure for
RkJQdWJsaXNoZXIy MjY0ODMw