Analýza poruchových stavů dveří kolejového vozidla založená na analýze provozních dat

Abstract

Railway transport is a major mode of transportation in much of the modern world. Its reliability is a key factor influencing the smooth flow of traffic, safety, and passenger comfort. Door system malfunctions are a significant source of operational complications, as they can cause delays and increase maintenance costs. Opportunities are emerging for the application of machine learning methods to predict these failures with increasing digitalization and the availability of operational and service data. This can lead to reduce maintenance costs and increase safety and reliability of railway transport.

This thesis focuses on prediction of door failures in trains operated in the Czech Republic. The goal is to design and test a model capable of predicting and alerting to failures in advance. The work is based on an analysis of provided operational data and maintenance records, which are used to train several machine learning methods. The results showed that door faults on this trains can be predicted only partially due to limitations in both datasets. The highest prediction rate was 51.3 % of faults with a higher rate (8.4 %) of false alerts, or possibly another model with 29.9 % and only 0.2 % of false alerts.

The contribution of this work is the discovery of a method to predict failures partially at least. Another contribution, particularly for practical applications, is the identification and demonstration of the impact of limitations, mainly in the methodology of operational data measurement. As a result, it is possible to learn from these errors that affect the information carried by the data signal.