The recent popularity of different Internet of Things (IoT) applications, such as Digital Twins and Precision Agriculture, has led to the development of different devices and wireless communications protocols to support the deployment of Wireless Sensor Networks (WSN). Unfortunately, WSNs are fault-prone systems, consisting of resource-constrained devices, and unreliable wireless communication links. Additionally, these devices are regularly exposed to the physical world, making WSN more vulnerable to malicious attacks, interference, and physical damage compared to other non-wireless networked systems. Identifying and localizing faults in WSN is relevant for the correct operation of these systems. However, it is a challenging task, and, therefore, there is a growing effort to develop effective fault localization methods.

Network tomography, the act of observing end-to-end in-network properties through special nodes known as monitors, is an emerging approach to localize faults in such networks.
This non-intrusive solution can be applied to different applications without requiring knowledge of the underlying system. The efficiency of this method depends on the number and position of the monitors, with recent research proposing algorithms to optimize the monitor placement while accurately localizing a number \emph{k} of simultaneous faults. The state-of-the-art solution solves this problem using a greedy heuristic named maximum node-identifiability monitor placement (MNMP). Initial experiments with different networks show, however, that this approach is too slow and does not scale for large systems.

This thesis develops solutions to make MNMP fast and adaptable to the topology changes common to WSN. For the fast computation of the monitor placement, we propose schemes for both reducing the size of the network analyzed and the number of redundant operations. For adapting this placement to topology changes, we develop strategies to minimize the search for new monitors by only analyzing the areas of interest. Our experiments show that our fast algorithms compute the best monitor placement 100x faster than MNMP, while the adaptive algorithm adjusts the initial monitor set 1000x more quickly.