Abstract
A wireless sensor network (WSN) for Structural Health Monitoring (SHM) can be described as a network with autonomous, spatially distributed sensor nodes that communicate wirelessly in a cooperative way to monitor physical or environmental conditions. Sensor nodes generally consist of a sensor, processor, power unit, and wireless communication device(s). Using WSNs to monitor a structure’s health has gained a lot of interest because of their ability to collect data in real-time without having to be physically present. WSN for SHM has garnered interest for protecting transportation infrastructure especially for the safe operation and maintenance of bridges. Two concerns that arise when designing and deploying these systems are energy consumption and information security. Limited battery capacity on hard-to-reach sensor nodes, especially on bridges, can significantly shorten WSNs with no easy way to replace them. Deployment conditions also leave these WSNs vulnerable to harsh environmental conditions as well as attacks on data integrity, confidentiality and availability from malicious actors masquerading as sensor nodes. This thesis proposes a scheme to protect data transmissions in WSNs for SHM without sacrificing energy consumption. The scheme solves the aforementioned problems by combining state-of-the-art technologies in deep learning, radio frequency (RF) fingerprinting and RF energy harvesting. RF Fingerprinting leverages process imperfections in transceivers that can be used in a deep neural network to authenticate known sensor nodes from unknown. Deep learning is also far less computationally intensive than more common forms of data security like encryption and decryption. RF energy harvesting can usurp the need for batteries by harnessing electromagnetic waves to convert to electrical energy that powers sensor nodes wirelessly. Deep learning requires a dataset to train the model. Additionally for authentication at the physical layer, each device needs its own dataset generation just like collecting fingerprints to establish a directory. This unique feature due to WSN for SHM of transportation infrastructure calls for the need for a framework to systematically generate datasets from individual sensor nodes. This brings out a novel approach of common applications in deep learning. The work shown acts as a proof of concept for this framework of data generation by building a prototype to present its feasibility through experimentation with using RF energy harvesting. This work also provides a framework for generating a dataset of device RF fingerprint to be used in a deep learning network to authenticate each sensor node.