192 Chapter 9. Conclusions & Recommendations 4. Explainability: Unlike traditional heuristics, DRL policies can be complex and di"cult to interpret. Research is needed to translate these policies into robust heuristics, aligning with the principles of explainable AI (XAI) (Linardatos, Papastefanopoulos, and Kotsiantis, 2021). XAI techniques aim to make the actions of AI agents understandable, helping to answer questions like: why is the agent taking this sequence of actions, and how can this be used to improve maintenance decisions? This is key for building trust in DRL-based methods. 5. System-level analysis: While much of our focus is on component-level analysis, transitioning to system-level analysis is crucial. Sewer networks are not isolated components, and system dependencies may play an important role in strategic maintenance planning. Developing frameworks that incorporate these dependencies and evaluate their impact is an important research direction. 6. Integration with prognostics models: Integrating predictions from degradation models into the MDP state space would equip agents with information on future deterioration. This would allow more e"cient exploration of the solution space and potentially improve decision-making. Further research is needed to assess how integrating prognostic models can enhance agent performance. 9.3 References Auger, S., J.-B. Besnier, M. van Bijnen, F. Cherqui, G. Chuzeville, F. Clemens-Meyer, M. G. Jaatun, J. Langeveld, Y. Le Gat, S. Moin, G. E. Oosterom, W. van Riel, B. Roghani, M. M. Rokstad, J. Røstum, F. Tscheikner-Gratl, and R. Ugarelli (June 2024). “Data management and quality control”. In: Asset Management of Urban Drainage Systems: If anything exciting happens, we’ve done it wrong! IWA Publishing. isbn: 9781789063059. doi: 10.2166/9781789063059_0299. Cherqui, F., F. Clemens-Meyer, F. Tscheikner-Gratl, and B. van Duin (June 2024). Asset Management of Urban Drainage Systems: If anything exciting happens, we’ve done it wrong! IWA Publishing. isbn: 9781789063059. doi: 10.2166/9781789063059. de Jonge, B. and P. A. Scarf (2020). “A review on maintenance optimization”. In: European Journal of Operational Research 285.3, pp. 805–824. issn: 0377-2217. doi: 10.1016/j.ejor.2019.09.047. Guo, H. and X. Yang (2007). “A simple reliability block diagram method for safety integrity verification”. In: Reliability Engineering & System Safety 92.9. Critical Infrastructures, pp. 1267–1273. issn: 0951-8320. doi: 10.1016/j.ress.2006.08.002. Hallak, A., D. Di Castro, and S. Mannor (2015). “Contextual Markov Decision Processes”. In: arXiv preprint arXiv:1502.02259. doi: 10.48550/arXiv.1502.02259. Hout, A. van den (2016). Multi-State Survival Models for Interval-Censored Data. CRC Press, pp. 1–238. isbn: 978-146656841-9. doi: 10.1201/9781315374321. Kıvanç, *., D. Özgür-Ünlüakın, and T. Bilgiç (2022). “Maintenance policy analysis of the regenerative air heater system using factored POMDPs”. In: Reliability Engineering & System Safety 219, p. 108195. issn: 0951-8320. doi: https://doi.org/10.1016/j. ress.2021.108195.
RkJQdWJsaXNoZXIy MjY0ODMw