From Degradation Prediction to Maintenance Decisions: A Data-Driven Framework for Railway Infrastructure
2026 (English)Doctoral thesis, comprehensive summary (Other academic)
Abstract [en]
Reliable railway infrastructure is fundamental to safe and sustainable transport. Maintaining track conditions requires timely interventions that balance infrastructure performance, maintenance costs and operational demands. Predictive maintenance offers opportunities to support these decisions, but its application is complicated by the characteristics of railway inspection data. Short and irregular measurement histories, missing observations and measurement noise can obscure degradation trends, while maintenance interventions introduce abrupt changes in track condition. These challenges affect both the reliability of forecasts and their use in maintenance planning and scheduling.
This thesis aims to develop an integrated, data-driven framework for railway maintenance planning and scheduling that incorporates degradation predictions and their associated uncertainty. The research combines a systematic literature review, semi-structured interviews with practitioners at the Swedish Transport Administration, and a longitudinal case study of the Swedish Iron Ore Line. The review and interviews establish the decision-making context and inform a taxonomy that clarifies the relationships between strategic, tactical and operational decisions. Track geometry measurements and maintenance records provide an empirical basis for investigating how degradation forecasts can support these decisions.
The identified decision requirements and data limitations guide the investigation of forecasting approaches. Stochastic models provide a probabilistic representation of degradation, while machine learning methods are explored for their potential to capture nonlinear relationships and temporal dependencies in track condition. Data preparation, model architecture and hyperparameter tuning are examined as ways of adapting forecasting methods to sparse and heterogeneous inspection histories. Evaluation across forecast horizons and inspection regimes clarifies the conditions under which model development can strengthen prediction and how the resulting forecasts should be interpreted for maintenance planning.
This thesis makes conceptual, empirical and methodological contributions to railway maintenance engineering. Conceptually, it develops a taxonomy that clarifies the interdependencies between maintenance planning and scheduling decisions across strategic, tactical and operational levels. Empirically, it advances understanding of how data characteristics, model configuration, hyperparameter tuning and forecast horizon influence degradation prediction. Methodologically, it connects degradation forecasts with maintenance decisions through probabilities of exceeding intervention limits and develops a robust maintenance optimisation formulation that incorporates disagreement between forecasting models. This formulation enables maintenance decisions to account for uncertainty associated with model choice while balancing preventive maintenance costs, the risk of delayed intervention and resource constraints. Together with the coordination of intervention timing and maintenance activities across neighbouring track segments, these contributions establish a structured connection between degradation modelling, uncertainty representation and maintenance optimisation. The thesis thereby advances the scientific basis for using imperfect condition information to support coordinated maintenance decisions under uncertainty.
Place, publisher, year, edition, pages
Luleå tekniska universitet, 2026.
Series
Doctoral thesis / Luleå University of Technology, ISSN 1402-1544
Keywords [en]
railway track maintenance, maintenance planning and scheduling, predictive maintenance, track geometry degradation, data quality, stochastic modelling, machine learning, hyperparameter tuning, uncertainty modelling, robust maintenance optimisation
National Category
Other Civil Engineering Reliability and Maintenance
Research subject
Quality Technology & Logistics
Identifiers
URN: urn:nbn:se:ltu:diva-120273ISBN: 978-91-8142-195-8 (print)ISBN: 978-91-8142-196-5 (electronic)OAI: oai:DiVA.org:ltu-120273DiVA, id: diva2:2112230
Public defence
2026-12-16, A109, Luleå University of Technology, Luleå, 09:00 (English)
Opponent
Supervisors
Funder
Luleå Railway Research Centre (JVTC), 473192026-10-262026-10-222026-10-26Bibliographically approved
List of papers