Analýza poruchových stavů výsuvných schůdků kolejového vozidla

Abstract

The thesis focuses on failure-state analysis of railway vehicle door systems and retractable steps using operational data with limited information quality. The main objective was to design and validate a methodology applicable to quantized, and event-driven data streams. The proposed approach combines a discrete-state Markov model with a Bayesian state observer that continuously refines the probability-state estimate for a specific component. The methodology includes data preprocessing, transition-matrix construction, state prediction, computation of supplementary health metrics like hazard function or remaining useful life, and statistical comparison of serviced and non-serviced components using the Kolmogorov--Smirnov test. Experiments were conducted on data from 31 trainsets, that coresponds to 496 door systems. For selected data streams, statistically significant differences between serviced and non-serviced cases were confirmed. The thesis provides a practically applicable foundation for further development of predictive diagnostics and a downstream classification layer for maintenance decision support.