• anomaly detection
• deep-learning
• encoder-decoder
• time-series
• transformer

Anomaly detection on multivariate time-series is critical for enhancing monitoring systems in various industrial applications, ensuring efficiency and reliability in productive processes. Recent advancements in deep learning have led to significant progress in this field, with diverse approaches spanning different domains. This thesis presents a review of relevant results from the literature and introduces two novel architectures: a transformer-based, free-attention model, and an encoder-decoder architecture utilizing a spectral approach.

Integration of anomaly detection with existing monitoring tools, particularly business intelligence platforms, is another crucial aspect addressed in this work. Leveraging the Qlik platform as a case study, the thesis proposes a Qlik plugin to extend its functionalities with PyTorch library powered architectures.

Anomaly detection on multivariate time-series is critical for enhancing monitoring systems in various industrial applications, ensuring efficiency and reliability in productive processes. Recent advancements in deep learning have led to significant progress in this field, with diverse approaches spanning different domains. This thesis presents a review of relevant results from the literature and introduces two novel architectures: a transformer-based, free-attention model, and an encoder-decoder architecture utilizing a spectral approach. Integration of anomaly detection with existing monitoring tools, particularly business intelligence platforms, is another crucial aspect addressed in this work. Leveraging the Qlik platform as a case study, the thesis proposes a Qlik plugin to extend its functionalities with PyTorch library powered architectures.