Background
Aircraft maintenance planning involves scheduling maintenance tasks and allocating the technicians, equipment, hangar slots and spare parts they require, under strict safety and availability constraints. In practice, schedules are still built from fixed intervals and static rules, and revised manually when conditions change: unexpected findings, delays, or a part that is not in stock. Prognostics and Health Management (PHM) now provides component-level health and remaining useful life estimates, which creates the opportunity to adapt maintenance and spare part decisions to the actual condition of the fleet. However, maintenance scheduling and spare part inventory are mostly optimised separately, despite their strong interdependence [1]. Joint, condition-driven formulations for multi-component aircraft systems with limited spares exist [2], and deep reinforcement learning has shown substantial cost reductions for engine maintenance [3], but such approaches remain far from the real-world scheduling systems used by operators.
The gap lies in dynamic scheduling: policies that update as the state of the aircraft and the operational context evolve. The literature is rich in mathematical frameworks for this purpose, from Markov decision processes and their contextual extensions [4] to stochastic programming and reinforcement learning, but implementing them in realistic setups and demonstrating their value against standard baselines is still a major challenge [5]. Three difficulties stand out. First, the action space is large and structured: executing a maintenance task requires preparatory actions, such as collecting equipment, securing the spare part and booking a technician, with precedence and resource constraints. Second, policies must be updated fast, which forces a trade-off between computationally expensive but precise models and lighter but approximate ones. Third, fair benchmarks against the rule-based practice are rarely reported. This project therefore formulates a mathematical framework for dynamic, joint maintenance scheduling and spare part management, starting from a small NLR use case and scaling up, and provides a comparative benchmark of several solution methods.
References
Van Horenbeek, A., Buré, J., Cattrysse, D., Pintelon, L., Vansteenwegen, P. (2013). Joint maintenance and inventory optimization systems: A review. International Journal of Production Economics , 143(2), 499-508.
de Pater, I., Mitici, M. (2021). Predictive maintenance for multi-component systems of repairables with Remaining-Useful-Life prognostics and a limited stock of spare components. Reliability Engineering & System Safety , 214, 107761.
Lee, J., Mitici, M. (2023). Deep reinforcement learning for predictive aircraft maintenance using probabilistic Remaining-Useful-Life prognostics. Reliability Engineering & System Safety , 230, 108908.
Hallak, A., Di Castro, D., Mannor, S. (2015). Contextual Markov Decision Processes. arXiv preprint arXiv:1502.02259.
Pinciroli, L., Baraldi, P., Zio, E. (2023). Maintenance optimization in industry 4.0. Reliability Engineering & System Safety , 234, 109204.
Assignment
The assignment will include the following tasks:
§ A literature study on dynamic maintenance scheduling and spare part management, covering sequential decision-making frameworks (e.g., Markov decision processes and their contextual extensions, stochastic programming, reinforcement learning) and their application to aircraft maintenance.
§ Selection of a use case provided by NLR, and formulation of a mathematical framework that jointly handles maintenance scheduling and spare part management, including preparatory actions and resource constraints.
§ Developing dynamic solution methods for the framework and analysing the trade-off between computational cost and solution quality when policies must be updated fast.
§ Applying the framework to the use case, starting small and scaling up, and providing a comparative benchmark of several…