AI-Augmented Big Data Frameworks for Smart Urban Infrastructure Maintenance Prediction

Abstract

Innovations in artificial intelligence (AI) and predictive analytics are driving the convergence of big data, sensor networks, the Internet of Things, and cyber-physical systems into environments known as smart cities. By augmenting conventional city information systems with predictive modeling capabilities that exploit the wealth of high-frequency operational data generated by sensor-equipped urban infrastructure assets, it is possible to implement proactive preventive maintenance strategies that promise to enhance infrastructure robustness, safety, performance, sustainability, and cost-effectiveness. Such strategies, termed predictive maintenance, exploit predictive models of asset health (measured via condition indicators), typically learned using supervised learning techniques, and create actionable information to optimise maintenance scheduling and better allocate limited maintenance budgets.

Despite considerable interest in predictive maintenance research, most efforts have focused on specific domains (e.g., buildings, roads, bridges, tunnels) in isolation. Thus, a big-data framework architecture that provides for the integrated prediction of asset health indicators across different urban infrastructure systems and their underlying sub-systems is developed. In the proposed framework, an architecture for the ingestion and integration of diverse high-dimensional data types is presented, together with a scalable ensemble-learning-based approach for feature engineering that exploits the multi-variate time-series nature of the data generated by real-world monitoring systems. By providing a generic architecture for the predictive maintenance of infrastructure assets, the framework lays the foundations for effective predictive maintenance strategies in smart cities.