Keywords: predictive maintenance, autocoupler, support vector machine, Markov chain, failure mode and effects analysis, railway reliability.

The automated coupler is a safety-critical interface in Electric Multiple Units (EMUs), where failures can lead to significant operational disruptions and safety risks. This study proposes a hybrid data-driven framework for the predictive maintenance of EMU autocouplers, analysing 4.5 years of operational data involving over 17 million fleet miles. The methodology integrates Failure Mode and Effects Analysis (FMEA) with machine learning to enhance risk assessment. A critical step in this framework involves converting text-based failure logs into a structured numerical format using text mining methods to extract meaningful failure patterns. A cost-sensitive Support Vector Machine (SVM) is then employed to classify failure modes into distinct Low, Medium, and High-risk categories, achieving 97% overall accuracy and a critical 100% recall for High-risk failures. Discrete-Time Markov Chain (DTMC) model is used to forecasts the short-term evolution of these risk states over a 7-day horizon. Results indicate that electrical faults contribute most to high-risk predictions, while mechanical and pneumatic categories show more stable, wear-driven risk behaviour. The proposed framework enables maintenance teams to transition from reactive repairs to predictive, risk-prioritised interventions with clear component-level insights.

download the full-text of this paper (PDF, 12 pages, 602 Kb)

go to the previous paper
go to the next paper
return to the table of contents
return to the volume description