UC Irvine

Modern community infrastructures are increasingly instrumented with Internet of Things (IoT) sensors and actuators, which enable many essential infrastructure monitoring applications. These applications are now ubiquitous across domains such as smart transportation, power grid, ambient environment sensing, smart buildings, among others, and provide important real-time information about the infrastructure and enable the accurate detection of critical events. A typical infrastructure monitoring framework involves data collection from distributed deployments of sensors, transmission over communication networks, and analysis using analytical models at edge and cloud servers to generate meaningful and actionable information.

However, communities exhibit heterogeneity in their structure and behavioral patterns. Structural heterogeneity can manifest through differences in topography, infrastructure scale and layout, community demographics, and available monitoring resources, while behavioral diversity can occur due to differences in weather phenomena, spatio-temporal patterns like vehicular movement, and other infrastructure activities. Current monitoring approaches are limited by this heterogeneity, and only work for specific communities and applications.

In this thesis, we propose solutions to leverage this heterogeneity to build effective, efficient, and adaptive infrastructure monitoring applications that can be deployed and shared across communities. Our proposed techniques leverage the unique structural and behavioral characteristics of communities, while also balancing monitoring requirements of applications with infrastructure resource availability. We explore our approach within the context of several real-world infrastructure monitoring applications and address three research problems across sensor deployment, operation of monitoring applications, and the generalization of monitoring solutions across communities.

First, we propose an impact-driven approach to IoT sensor placement that leverages community characteristics to determine vulnerable regions and measures the potential impact of events which is used to prioritize deployment locations. Second, we design an operational monitoring framework that handles heterogeneity in devices, communication networks, and analytical models and develop an adaptive decision making approach to determine the optimal choices for monitoring while balancing performance, resource consumption and current community conditions. Finally, we present an approach that enables training robust infrastructure monitoring models from multiple data sources in a distributed and bias-agnostic manner, that can then generalize or be reused across communities without a loss in performance. Together, the proposed techniques provide a comprehensive approach for infrastructure monitoring that can exploit and adapt to structural and behavioral characteristics of communities. We validate our approach using prototype implementations on several real-world infrastructure testbeds.