9 9.2. Recommendations 189 a single gate and event might be optimal in size but poor in accuracy. The FT-MOEA-CMextension (Chapter 4) addresses this by introducing extra metrics, but other solutions could be explored. 2. Scalability: Scalability is a major issue as complexity increases exponentially with the number of basic events. Although SymLearn (Chapter 3) and FT-MOEA-CM (Chapter 4) demonstrated improvements, performance with larger datasets remains uncertain. Evolutionary algorithms struggle with the vast solution space and require excessive computation time for larger FT populations, complicating convergence. More e"cient methods are needed for larger problems. 3. Noisy and incomplete data: Real-world datasets are often noisy or incomplete, lacking records of rare events and containing inaccuracies. While this was briefly discussed in Chapter 2, more research is needed to address these issues e!ectively. 4. Comparison benchmark: A systematic comparison of algorithms for FT model inference is lacking. Such a benchmark could highlight strengths and weaknesses, suggesting improvements. 5. Additional logic gates: We focused on static FTs (using AND and OR gates). Including more gate types (e.g., voting gates) and considering dynamic FTs (Ruijters and Stoelinga, 2015) could lead to more e"cient and representative models. 6. Application to other formalisms: Our approach, based on Multi-Objective Evolutionary Algorithm, may also apply to other formalisms with similar challenges (i.e., automatically inferring the graph structure from data), such as Attack trees (Kordy, Piètre-Cambacédès, and Schweitzer, 2014), dynamic FTs, and Reliability Block Diagrams (Guo and Yang, 2007). Research in this direction is valuable. 9.2.2 Multi-State Deterioration Modelling of Sewer Mains In Chapters 5 and 6, we explored the use of Markov chains to model damage severity in sewer mains. Below, we outline the remaining challenges: 1. Data collection: A key challenge in data collection is the ‘data-loop problem’ (Auger, Besnier, Bijnen, et al., 2024), where the lack of evidence for the benefits of data collection discourages further investment. This leads to a cycle of declining data quality and quantity, reducing the e!ectiveness of systematic data collection. This problem is not exclusive to sewer systems but is common in civil infrastructure. Breaking this cycle is crucial to address data scarcity and build trust in the methods. 2. Extending the inspection goal: Currently, sewer main inspections focus on assessing pipe conditions for maintenance or replacement, rather than feeding degradation models. To address this, inspection campaigns should
RkJQdWJsaXNoZXIy MjY0ODMw