Artificial Intelligence Methods for Distributed and Embedded Sensing Systems
This doctoral research focuses on the development of artificial intelligence methods for distributed and embedded sensing systems in industrial applications. The objective of this work is to design robust and efficient approaches for monitoring, prediction, and adaptive inference in real-world scenarios. ; ; The research is articulated along two complementary directions. The first concerns AI-based gas leak detection in urban gas distribution networks, within a project funded by ARERA and conducted in collaboration with A2A/RetiPiù and Onyax. The study combines laboratory experiments on real pipeline sections and in-field campaigns to investigate gas propagation dynamics, including attenuation, delay, and signal behavior. A distributed infrastructure of more than sixty sensing devices collects daily recordings from the network, supporting the construction of a large-scale dataset. Ground-truth labels are obtained through periodic inspections performed with Picarro instrumentation. This dataset is used to develop and assess deep learning approaches, including autoencoders and convolutional neural networks, for the early identification of leak-related anomalies. ; ; The second direction, carried out with STMicroelectronics, investigates Edge AI solutions for smart sensing platforms. The objective is to design neural models that can operate directly on embedded devices, enabling inference and, in selected cases, on-device learning within severe resource constraints. The considered applications include sensor calibration, human activity recognition, and lifetime estimation for sensing components, including IGBT-related prognostic tasks. Particular attention is devoted to efficient and innovative training paradigms, such as forward-only neural approaches that avoid standard backpropagation and are therefore promising for deployment in highly constrained hardware environments. ; ; These research activities contribute to the advancement of intelligent sensing systems by addressing both distributed infrastructure monitoring and embedded on-sensor intelligence. The overall goal is to provide scalable, reliable, and resource-efficient AI methodologies for next-generation industrial sensing applications.
Back to Current Students