ETD

• adaptive threshold
• anomaly detection
• aws
• predictive maintenance
• time series analysis

In the context of Industry 4.0, the widespread adoption of Industrial Internet of Things (IIoT) sensors has made real-time condition monitoring an increasingly important component of modern industrial systems.
Among the available monitoring techniques, lubricating oil analysis represents an effective approach to assess machinery health, as oil degradation and contamination provide early indications of component wear and failure progression.
This thesis proposes an adaptive thresholding approach for oil condition monitoring aimed at detecting anomalous behavior in time series data, and thus identifying the early onset of degradation mechanisms.
The methodology combines signal preprocessing, feature extraction and the definition of an adaptive thresholding algorithm.
The proposed solution has been validated on both controlled datasets and real industrial scenarios, showing promising results in detecting different types of anomalies.
Furthermore, the entire processing pipeline has been deployed on a cloud-based architecture using Amazon Web Services, enabling automated data processing, real-time monitoring and user notifications. The work was carried out in collaboration with the italian company SanChip S.r.l., demonstrating the practical applicability of the proposed approach in an industrial context.