Many industries and manufacturing processes rely on fixed maintenance schedules. Given the abundance of data available nowadays, it is promising to closely monitor such processes, timely detect degradation, and intervene precisely when necessary.

The aim is to automatically detect such anomalies, i.e., whenever the underlying data-generating process changes its properties such that the behavior does not conform with normal behavior anymore. As accurate models for the normal behavior are rarely available, one must use the available data to either refine existing models or learn them altogether.

Since anomaly detection is important throughout different industries, there are multiple opportunities to conduct this master thesis with one of our cooperation partners as e.g., CERN, RHI Magnesita, or Siemens Mobility.

Your Tasks :

• Literature survey for unsupervised anomaly detection in multivariate time-series
• Implement and compare different algorithms from the literature
• Learn a model for the normal behavior
• Perform anomaly detection on practical data

Your Profile :

• Motivation and interest in the topic
• Background in machine learning
• Experience in python programming

Contact : Christian Knoll (christian.knoll@tugraz.at)