Infrastructure owners monitor changes in the track geometry to safeguard operation and to plan maintenance activities. Usually, track geometry is monitored using specialized measurement vehicles assessing the track several times a year to establish information on the development of a specific locations along the track. In this paper, a prediction tool is proposed and described that combines information from the measurement vehicles with measurements of onboard monitoring systems on regular trains to estimate and predict the properties longitudinal level and twist. Further, static asset configuration and information on the infrastructure is used in the decision making to provide actionable insights. The tool provides an improved resolution in time for track geometry properties and predictions on how these properties develop in the future including information on the uncertainties.

It will be discussed how data from different sources with irregular sampling need to be preprocessed to be combined and harmonized. Moreover, in what way different principles from data science, machine learning and estimation theory can be combined with domain knowledge to enable a better analytics and decision making. To support engineers, it is shown how a decision support tool for maintenance can be tailored and used. Finally, the tool and approach are showcased and benchmarked on track systems in Europe, where both onboard monitoring data and data from measurement trains is available. The results indicate that such a tool provide improved actionable insights to practitioners.