The company contact Terminus7 to develop an advance signal processing solution through Machine Learning, looking for infrastructure monitoring. This company also contact Terminus7 to develop a train detection solution (see Train Detection use case).
First, we normalise noisy signal over the installation. The optical fiber demonstrates not to be regular through the rail and so it signals under the same situation. Therefore, we applied Machine Learning advanced techniques to find a reference pattern over which normalise every signal.
In the second step, we applied Machine Learning to detect signal patterns related to rails and different wheels defects and failures. Therefore, the model looks signal deviations over the expected signal when rails and wheels suffer or are going to suffer any damage.
This is a WIP project. Our results demonstrate the ability of Terminus7 to find rail and wheels patterns and patterns deviation that announce a potential future failure. This gives the company the ability to provide an infrastructure preventive maintenance with higher precision compared to costly inspection techniques or standard average lifespan.
Current work is focused on model fine tuning to speed the solution to its best accuracy. At the same time we are improving the model to detect patterns related to other rail events, like the crossing of people or cattles.