Antonio Squicciarini
Projects
Welcome to my portfolio of selected research and engineering projects. My work spans machine learning applications, deep learning optimization, time-series analysis, signal processing, and real-world diagnostics in high-voltage systems and mechanical components. Below you’ll find detailed descriptions of my key contributions, including methodologies, visuals, and links to related publications.
Multiscale Time-Series Feature Extraction Using Entropic & Informational Functionals
We developed a multiscale feature extraction methodology for time series using adaptive kernel density estimation and a library of entropic and informational measures, enabling interpretable anomaly detection and time-series characterization across various domains like fault monitoring and EEG analysis.
Jensen-Tsallis Divergence for Supervised Classification under Data Imbalance
We propose the Jensen-Tsallis Divergence (JTD) as a novel loss function for deep learning on imbalanced datasets, offering tunable regularization to improve generalization and outperforming traditional methods like cross-entropy, JSD, and focal loss.
Rub Detection in Aeroderivative Gas Turbines
We developed an intelligent fault detection system for early rub detection in aero-derivative gas turbines by training deep neural networks on synthetic vibration data from finite element simulations, achieving successful transfer learning to real experimental signals.
Partial Discharge Detection
We developed a partial discharge classification system for high-voltage DC environments using deep learning, combining PRPD-inspired preprocessing, phase-robust data augmentation, and model interpretability through Integrated Gradients.
Congress Posters and Presentations
Here you can find my posters and presentations displayed at various conferences and workshops.