Can AI methods, based on the combination of new external data sources such as weather data and known internal data, give better predictions and lay the groundwork for a more dynamic and state-based maintenance program in rail traffic?
In short, this was the question that was asked in this project, and the answer to the question was in brief yes.
By collecting, integrating, cleaning and processing information from seven different data sources, a set of measures was created that captures 286 factors with potential impact on faults in the switches along the Malmbanan railroad. The data sources ranged from fault reporting data, changeover temperature and snow logs to pure weather data. Based on the data, machine learning models were evaluated to see which model had the best ability to determine the explanatory value of each factor for reported errors.
The result of the feasibility study was a rich data source for further analysis and a compilation of the identified power relationships. In addition, the AI model that was developed proved to have the potential to identify vulnerable switches, be able to provide early warnings and thus create a basis for risk planning based on different scenarios. All in all, the results of the project represent a step towards a more state-based and proactive maintenance program.