Condition-based maintenance is an emerging maintenance strategy for aircraft, leveraging the constant collection of sensor information to facilitate diagnostics and prognostics of potential failures. The implementation of this strategy requires a significant initial investment and therefore, the resulting benefits should be quantified in advance. The objective of this research is to investigate the potential benefits of condition- based maintenance (CBM), and more specifically, the use of Prognostic & Health Management (PHM) systems operating with dynamic failure thresholds, with a focus on the required performance levels of the PHM systems. In this paper, a scheduling framework has been developed to schedule preventive maintenance tasks under application of prognostics, using a rolling-horizon scheduling approach to allocate tasks to appropriate maintenance blocks. The resulting maintenance schedule of a fleet of aircraft is subsequently used for the simulation of subsystem failures and the application of prognostics in order to anticipate them. Finally, the possibility of reducing maintenance cost, increasing fleet availability and improving operational reliability is investigated through a cost-benefit analysis. Results show significant improvements in terms of fleet availability and operational reliability, and minor reductions of maintenance cost. Moreover, the achieved benefits are shown to be in relation to the prognostic performance levels, and the scale to which condition-based maintenance is applied.

Condition-based maintenance (CBM) is emerging in the airline industry as a revolutionary concept that could potentially increase the efficiency of aircraft operations. The adoption of CBM strategies is not yet widespread due to the stringent requirements imposed by aviation authorities and the fact that the prognostic and health management (PHM) technology is still in its infancy. Along with the developing technology, CBM requires a performance evaluation in terms of aircraft maintenance schedule optimization. This paper proposes a novel reinforcement learning model to solve the airline maintenance scheduling problem subject to prognostics uncertainty. A Deep Q-Learning model is trained to schedule A-check routine tasks. The approach preserves the dichotomy of the maintenance problem by scheduling both interval-based clusters of tasks and condition-based tasks monitored with synthetic prognostics. Maintenance, repair, and overhaul (MRO) data from a major European airline was used to construct a realistic simulation and a model tailored to their operations. We explore a wide range of scenarios with varied numbers of tasks scheduled with a CBM policy, as well as different magnitudes of uncertainty in order to enable a viable maintenance strategy. Compared to traditional maintenance policies, the results demonstrate that the implementation of CBM reduces the fleet ground time and improves the task's interval utilization when assessing the uncertainty involved in prognostics.