Development of a cost optimal predictive maintenance strategy

Maintenance costs account for a significant share of operating expenses. Selecting the optimal maintenance strategy for each application is crucial to optimize operational processes and minimize MRO spending. In recent years, Machine Learning has become popular for analyzing large amounts of data and improving decision-making in various industries. This yields great potential in the field of Predictive Maintenance. In this thesis, a methodology to determine and compare the average maintenance costs per cycle for Reactive, Preventive, and Predictive Maintenance, as well as a Reference Case is developed. This cost comparison methodology is then applied to a realistic example of a fleet of ten aircraft. Unlike previous research, this thesis combines all aspects in one approach, from Machine Learning algorithm selection and RUL prediction, to the maintenance cost comparison based on a fleet of aircraft. The NASA CMAPSS jet engine dataset is used as an example. Results suggest that maintenance costs per cycle for Predictive Maintenance are 36.0 % lower than for Preventive Maintenance and 88.3 % lower compared to Reactive Maintenance. In general, this thesis serves as a guideline that highlights the necessary steps to determine the cost-optimal maintenance strategy for an application.