Research on the efficiency of machine learning in signal point monitoring

Abstract

Abstract. Railway signalling automation devices have been around for quite some time. Many systems in operation are considered obsolete. At the same time, modernization does not cover the whole Russian railway system. The deployment of monitoring systems since the beginning of the 2000s made the maintenance of bulky signalling systems more comfortable, reduced the time of fault detection. The second decade of the 21-st century saw a widespread deployment of information technologies in various spheres of life, including industry, yet at a slower pace, especially in railway transportation. One of the innovations was the emergence of the artificial intelligence, which enabled more progressive maintenance of devices through prediction of pre-failure states. The latter allows notification of technical personnel by an intelligent system that processes significant parameters of the observed facility or process, thereby replacing manual monitoring that requires time and professional experience.

Aim. To suggest the use of artificial intelligence-based methods in railway signalling devices based on the existing technical diagnostics and monitoring systems. Methods. Python-based unsupervised machine  learning methods are used to create, process and visualise data.

Results. The AI models showed a reaction to anomalous changes in the temporal characteristics of code generators.

Conclusion. An AI-enabled program can serve as the core for processing data related to the monitoring of railway signalling devices and requires careful research in predicting their known failures at the signal point.