ResilAI: Exploring the Applications of AI Techniques in Supporting Infrastructure Resilience

During the past 40-years, natural hazards led to 15 billion euros per year in economic losses in Europe. Infrastructure is vital for a safe society, but it is also prone to fail during disasters. Infrastructure system resilience has been a cutting-edge research topic, where AI offers new insights. However, the utilisation of AI in this field remains inadequate due to limited data availability.

It is imperative to explore synthetic approaches for data collection and investigate the principles governing data requirements and AI technique selection.

Furthermore, current studies mainly focus on the assessment and enhancement of resilience, while there is still a dearth of data-driven investigations that elucidate the mechanisms underlying the resilience of interdependent social-technical infrastructure systems (ISTIS).

Considering these gaps, the ResilAI is dedicated to investigating how AI techniques reinforce the research on infrastructure system resilience to contribute to more resilient and sustainable infrastructure.

The objectives are: (1) to devise a synthetic approach to collecting data for studying the resilience of ISTIS; (2) to propose the principles of data requirements for resilience prediction using machine learning (ML) approaches; (3) to identify the key drivers and reveal the mechanisms of the resilience of ISTIS.

Correspondingly, the ResilAI will: (1) combine conventional data collection and data generation methods to comprehensively gather data on ISTIS; (2) examine the performance of different ML approaches to predict the resilience of ISTIS with different availability of data and establish fundamental principles for data requirements; (3) find the key factors that contribute to the resilience of ISTIS and the formation mechanisms of resilience based on explainable AI.

The outcomes will provide academia and industry with a database on infrastructure system resilience as well as the foundation for further data-driven research and practice in this field.