20 Chapter 1. Introduction - Part III: how can near-optimal maintenance strategies be devised for components with Multi-State Deterioration such as sewer mains using Deep Reinforcement Learning? It is worth noting that the answers from Part II on sewer mains degradation modelling are utilised in Part III on maintenance optimisation. In the following section, we discuss the research methodology employed in this dissertation. 1.5 Research methodology The methodology to answer the research questions in Section 1.4 is divided into three main phases (Figure 1.12), with each phase associated with a part of the dissertation. Phase I: Data-driven Inference of Fault Tree models (Part I) Ch.2: Implement Multi-Objective Evolutionary algorithms for inference of FTs Ch.3: Explore symmetries and modules to enhance scalability Ch.4: Explores the use of metrics derived from the confusion-matrix Phase II: Multi-state deterioration modelling (Part II) Ch.5: Presents sewer network case study and the use of discrete-time Markov chains for degradation modelling Phase III: Maintenance optimisation of multi-state components (Part III) Ch.7: Explores the use of Deep Reinforcement Learning for maintenance optimisation in sewer mains under different model assumptions Ch.6: Evaluates homogeneous- vs inhomogeneous-time Markov chains for degradation modelling in sewer mains Figure 1.12: General methodology used in this dissertation. Phase I focuses on the data-driven inference of Fault Tree (FT) models (Part I). For this, we employ multi-objective optimisation and evolutionary algorithms, leading to the development of the FT-MOEA algorithm (Jimenez-Roa, Heskes, Tinga, et al., 2023) (see Chapter 2). This phase produced two extensions: the investigation of symmetries and modules to enhance scalability in systems with symmetrical failure data, resulting in the SymLearn tool-chain (see Chapter 3); and the exploration of metrics derived from the confusion matrix to improve scalability and robustness (see Chapter 4). Phase II focuses on Multi-State Deterioration modelling (Part II), with a particular emphasis on sewer main degradation. Initially, we present a real-world case study from a sewer network in the Netherlands and explore Discrete-Time Markov Chains to model deterioration in sewer mains across di!erent cohorts (see Chapter 5), evaluating two DTMCs types. Subsequently, we extend this work by
RkJQdWJsaXNoZXIy MjY0ODMw