Predictive Maintenance Framework to Identify and Analyze Railway Failures

Abstract

With the continued growth of railway networks globally and more specifically within the Gulf Cooperation Council (GCC) region, it has become evident that traditional time-based maintenance strategies are no longer sufficient as these existing methods can compromise overall safety, reliability, and service continuity. This thesis presents a comprehensive investigation into common railway failure patterns and maintenance challenges for the development of a predictive modelling maintenance framework to analyze and forecast failures across interconnected railway systems. The primary focus of this thesis is on the on the Doha Metro network which is operated by Qatar Railways Company (Qatar Rail). A systematic review of existing literature on railway maintenance challenges and failure patterns is conducted which identified a significant transition in the overall industry from traditional preventive maintenance approaches toward data-driven and condition-based strategies. The literature review also further identified failures across specific disciplines and interconnected systems frequently propagate through the network which amplifies overall operational and safety impact. Digital technologies including the Internet of Things (IoT), machine learning, and advanced condition monitoring systems are identified as key enablers for improving maintenance effectiveness and system reliability. Building upon the comprehensive analysis of the Doha Metro Network datasets, a predictive modelling maintenance framework was developed to evaluate and enhance the maintenance strategies. The results of this approach are as follows: The Support Vector Machine (SVM) failure mode classifier achieved 91.5% accuracy and a macro-F1 of 0.903 across eight of the major subsystem categories. The climate-aware modeling revealed that energy related failures escalate above 40°C while track related failures follow an inverse seasonal pattern which confirms that some failure dynamics are environmentally dependent. The overall findings further demonstrate that effective analysis and modeling based on the relevant operational and maintenance datasets enable a scalable predictive maintenance framework which can be further applied across the broader GCC railway community.

Date of Award	2026
Original language	American English
Awarding Institution	HBKU College of Science and Engineering