In the research focus area "Sensor-based situation and condition assessment", we develop learning methods for processing and analysing transient sensor data in an industrial environment. Our aim is to use the latest deep learning technologies to efficiently monitor complex, dynamic systems, recognise anomalies and make predictions.

Analysing sensor data opens up a wide range of possible applications: from anomaly detection and fault classification to the development of soft sensors and the prediction of future events. We use them for monitoring and quality control of production processes, predicting maintenance scenarios or forecasting critical events such as machine breakdowns or flooding.

Industrial sensor data is complex and multidimensional. We use and extend state-of-the-art model architectures such as transformer networks to analyse this data. Our focus is on automated learning of the best possible representations of sensor data (representation learning) and overcoming real-world challenges such as small amounts of data, complex temporal patterns and high-frequency data streams.