Artificial Intelligence-powered Devices and Sensors

Permanent URI for this collection

Browse

Recent Submissions

Now showing 1 - 3 of 3
  • Item
    Graph Neural Network Solutions for Anomaly Detection in Time Series
    (2025-01-07) Cornali, Roberto; Ronchieri, Elisabetta; Cesini, Daniele; Dell’Agnello, Luca
    Anomaly detection is an essential task for many firms and organizations. Identifying unusual patterns in messy multivariate time series can prevent catastrophic events and optimize operations. Traditional statistical methods struggle with high-dimensional data and complex temporal dependencies. In this paper, we propose a novel approach, combining Generative Adversarial Networks (GANs), a reconstruction-based framework, and Graph Neural Networks (GNNs) for effective and interpretable anomaly detection. Our method involves representing multivariate times series as graphs and training two interconnected GANs and an Autoencoder to capture the normal behavior of the networks. Anomalies are detected by measuring the reconstruction error and the discriminator's score. Our case study on the signals of the INFN CNAF Tier-1 data center demonstrates the effectiveness of our approach in terms of robustness and interpretability. We validate our model on NASA's public space telemetry signals, a popular multivariate time series benchmark and observe cutting-edge performance. This work highlights the potential of GNNs in developing interpretable deep learning solutions for real-world applications.
  • Item
    Predictive Maintenance Study for High-Pressure Industrial Compressors: Hybrid Clustering Models
    (2025-01-07) Costa, Alessandro; Mastriani, Emilio; Incardona, Federico; Munari, Kevin; Spinello, Sebastiano
    This innovative approach presents a predictive maintenance strategy for high-pressure industrial compressors based on sensor data. In the context of accurately classifying potential compressor failures, this study investigates whether and how much features from upstream unsupervised clustering enhance clustering models in terms of classification accuracy and training performance. The methodology integrates time series analysis, advanced clustering techniques, and hybrid clustering modeling, including feature engineering using auto-correlation analysis and ANOVA, time-series-aware clustering, and six clustering models (Logistic Regression, SVC, GaussianNB, Gradient Boosting, KNC, and RFC). This hybrid clustering sets our work apart from traditional solutions. The approach is validated using cross-validation and key metrics such as accuracy tests. The final results indicate that using the top features identified through unsupervised pre-clustering improves the accuracy of the test set in detecting non operating conditions by an average of 4.87%. Additionally, the training time for clustering models is reduced by an average of 22.96%.
  • Item
    Introduction to the Minitrack on Artificial Intelligence-powered Devices and Sensors
    (2025-01-07) Dell’Agnello, Luca; Cesini, Daniele; Ronchieri, Elisabetta