Artificial intelligence–enabled self-healing infrastructure systems

Lauren McMillan*

*Corresponding author for this work

Research output: ThesisDoctoral Thesis

Abstract

Modern infrastructure systems are grappling with increased complexity and interdependence, struggling to predict and manage failures amid factors like population growth, urbanisation, rapid climate change, and economic challenges. While management methods remain fragmented, the rise of digitalisation and artificial intelligence (AI) offers a chance to adapt complex software-based approaches for infrastructure applications. One such approach is 'self-healing,' which anticipates and autonomously responds to system failures. AI's characteristics align well with self-healing concepts, making it a pivotal enabler. However, AI's current status in infrastructure management is unclear and there is a need to explore its application, learning from best practices in various sectors. Hence, this work presents a framework for self-healing infrastructure systems and explores the key components and processes necessary for implementation. Furthermore, in order to explore practical implementation, the framework is applied to leakage management in a water distribution system. Intelligent, data-driven solutions are proposed for each of the processes – anticipation, detection, and restoration – required to manage leakage as a self healing system and these are trained and tested on a dataset of over 2,000 district metered areas (DMAs) managed by a UK water company. By offering a rapid and cost efficient method for the identification of potential leakage, the benefits of this approach include enhanced resilience, optimised repair strategies, and improved consumer confidence, fostering sustainable demand-side behaviours. The contribution is a self healing framework for management of leakage in water distribution systems, which demonstrates strong performance on the historical data provided and has the potential to be adapted to suit other contexts (including other types of infrastructure network). The findings of this research are of value to infrastructure owners and operators, regulators, and researchers, who see the potential in adopting a complex system perspective and recognise the role of AI in effectively applying this perspective to the management of real-world systems.
Original languageEnglish
QualificationDoctor of Philosophy
Awarding Institution
  • University College London
Supervisors/Advisors
  • Varga, Liz, Supervisor, External person
  • Fayaz, Jawad, Supervisor, External person
Award date28 Mar 2024
Place of PublicationLondon
Publisher
Publication statusUnpublished - 28 Mar 2024
Externally publishedYes

Cite this