Pattern Recognition on Railway Points with Machine Learning: A Real Case Study

Abstract

Railway points are crucial components to ensure the reliability of railway networks. Several types of condition monitoring systems are widely applied in the industry with the aim of early fault detection, since the presence of faults can cause reductions in operational safety, delays and increased maintenance costs. The application of fault detection systems and pattern recognition tools is essential to ensure new improvements in the industry. The novelty proposed in this work is the application of statistical analysis and Machine Learning techniques in power curves defined by the movement of the motors in the opening and closing movements. The Shapelets algorithm is selected for pattern recognition, analyzing curves with abnormal distribution that demonstrate the presence of faults. The results provide high accuracy with performance measures above 90%.