Time series are ubiquitous in aerospace engineering and represent a large part of the data with highest business value potential. In this talk we focus on automatic anomaly detection tasks for aircraft sensors. We assess the industrial viability of a semi-supervised anomaly detection system based on Deep Learning for automatic discovery of point, contextual and collective anomalies on a large dataset with little prior knowledge.
- Set up of detection approach with unlabelled data
- Understanding human-level detection performance
- Deep Learning architectures for anomaly detection