Abstract
The continued operation and reliability of critical systems like railways hinge upon effective maintenance practices. Various issues, ranging from missing fasteners and deformed track geometry to structural health insufficiencies and overgrown vegetation, can significantly impact system performance and lead to non-negligible downtimes. Traditional inspection methods, such as foot patrols, trolleys, and measurement vehicles, are comprehensive but often resource-intensive, time-consuming, and disruptive to routine traffic.This research proposes a paradigm shift by exploring the application of autonomously operated Unmanned Aerial Vehicles (UAVs) for detecting issues requiring maintenance in railway systems. While UAVs have proven successful in other industries like bridge maintenance and building inspection, their application in railway systems spanning long distances remains largely unexplored. The investigation specifically focuses on inexpensive and agile UAVs operating autonomously, a pioneering approach in the railway maintenance domain.
The aim of this research is to identify application-based improvements and develop solutions enabled by autonomous UAVs for detecting maintenance needs. By leveraging technologies such as computer vision and machine learning, the project involves designing and implementing an autonomous UAV system, integrating the required sensors, developing novel and efficient algorithms for performing autonomous navigation and maintenance needs detection from data gathered onboard the UAV, and bridging any simulation to reality gap that may exist in the solution stack.
The outcomes of this research includes a framework for autonomous UAV navigation along railway corridors, plus techniques for performing maintenance needs detection using data gathered from the UAV itself, reducing the time and personnel manpower required for comprehensive inspections. This innovative approach not only aligns with current trends in engineering but also addresses a pressing challenge in railway maintenance. It is expected to impact society by minimising train service downtime, reducing costs, and enhancing the reliability, safety, and efficiency of both maintenance and operation of railway systems. At a more fundamental level, the contributions of this work include a hackable UAV simulation software named PyFlyt; Critic Confidence Guided Exploration — a reinforcement learning algorithm for allowing agents to bootstrap off oracle policies instead of learning from scratch; and algorithms and insights for performing anomaly detection from image data for railway systems.
| Date of Award | Jan 2025 |
|---|---|
| Original language | English |
| Awarding Institution |
|
| Supervisor | Mauro Innocente (Supervisor), Nadjim Horri (Supervisor) & James Brusey (Supervisor) |
Keywords
- Autonomous UAVs
- Railway maintenance
- Infrastructure inspection
- Computer vision
- Machine learning
- Anomaly detection
- Autonomous navigation
- Reinforcement learning
- PyFlyt simulation
- Structural health monitoring
Cite this
- Standard