Predictive maintenance of technical systems using data from condition monitoring systems is playing an increasingly important role in many industrial applications. Machine learning methods are the algorithmic core to achieve this goal. SCAI develops approaches which allow to determine a small number of significant indicators with which deviations from the standard state of a system can be easily measured, but which go beyond standard methods in recognition quality and robustness.
An application example is the predictive maintenance of wind turbines based on data from vibration sensors, which measure the vibration behavior of the rotor blades. Machine learning methods are investigated and developed in order to better detect anomalies, e.g. to enable ice detection or to detect possible damage in the data at an early stage. Further diagnoses will be made possible by the status data of the control station of a wind turbine, with the aim of developing semi-automatic control strategies, for example to reduce wear and tear and thus achieve a longer service life of the turbine.