Optimisation of Rolling Stock Maintenance Operations using Autonomous Intelligent Systems

  • Edvin Foric

Student thesis: Doctoral Thesis

Abstract

Rolling stock maintenance is typically carried out using a combination of corrective and preventative maintenance. Corrective maintenance is unplanned reactive work undertaken when a fault is identified or following an in-service failure. Preventive maintenance is planned work carried out to avoid in-service failures through a range of inspections, checks and replacement of components as they degrade with use. Preventative maintenance is typically carried out using an interval-based strategy, however, there is a current trend towards condition based preventative maintenance. The selection of the optimum strategy or maintenance plan is often subjective and based on experience with little scientific proof. Advances in non-destructive testing (NDT) and Remote Condition Monitoring (RCM) technology and increases in the volume of data acquired to describe the condition of rolling stock components/systems, provides the opportunity to implement intelligent and autonomous systems (AIS) which collectively automise the planning and management of rolling stock maintenance operations. This has the potential to provide significant benefits to train operators/maintainers through improvements in efficiency and reduction in costs. The potential impacts and benefits of using AIS in rolling stock maintenance include improved utilisation of infrastructure and resources, enhanced reliability, maximisation of component lifespan, reduced unnecessary maintenance tasks, improved safety, more accurate and efficient maintenance planning, as well as an increased responsiveness to changes in factors like vehicle, staff availability, equipment readiness, safety regulations, and standards. AIS within the context of the research refers to artificial intelligence (AI) algorithms which, without direct human supervision, have the ability to make important maintenance decisions and manage a range of depot tasks.The research presented in this thesis investigated the potential of AIS for the control and planning of rolling stock maintenance operations to optimise maintenance decisions in a depot. The project was split into several stages: the first stage included an investigation into the basic concepts of rolling stock maintenance and a literature review which covered the current state-of-the-art of rolling stock maintenance and identification of emerging AIS techniques, which has looked at different methods to improve the efficiency of rolling stock maintenance and other systems, such as Genetic Algorithms, Particle Swarm Optimisation Algorithms, Agent-Based Modelling and various decision support models. There was an identification of a gap in research regarding the development of a hybrid approach combining optimisation techniques which focuses on optimising a combination of different maintenance techniques such as the use of predictive maintenance, scheduled maintenance, and corrective maintenance together in a singular environment, which incorporates many decision variables that are present in examples seen in older rolling stock and more modern rolling stock maintenance depots. This approach is necessary as it focuses on a more realistic setting which doesn’t necessarily use a fully modern approach. A model simulator of a typical maintenance depot was developed to baseline existing performance and investigate how AIS through a digital twin can be used to automate and optimise maintenance decisions within a rolling stock maintenance depot. Historical and simulated condition data was used in the development of a model simulator of a generic train maintenance depot, which would connect to the digital twin.The research has used key performance indicators such as depot resource allocation, number of maintenance activities, train out-of-service frequency and inspection intervals to evaluate the effectiveness of the digital twin implemented in the model simulator. There was an improvement in operational efficiency and resource management, as the digital twin demonstrated a reduction from 54 to 0 occurrences in the frequency of out-of-service of trains due to insufficient depot resources to action wheelset replacement operations. This shows the efficiency of the AIS technique within the digital twin to identify and resolve resource strains by optimally rescheduling maintenance tasks. AIS successfully reduced the frequency of out-of-service incidents by 88%, which shows a significantly1reduced train downtime. There was also an improved distribution of maintenance workload on the underfloor wheel lathe and maintenance road 3. However, the research also identified a compromise between the reduced lifespan of components and the increased optimisation of scheduling.
Date of Award6 Dec 2024
Original languageEnglish
SupervisorAdam Bevan (Main Supervisor), Gareth Tucker (Co-Supervisor) & Mauro Vallati (Co-Supervisor)

Cite this

'