ICON Project for context-aware automated learning for anomaly detection
Just like your own car, industrial machines and vehicles can fail over time, resulting in breakdowns and unexpected downtime. In order to prevent reduced product quality, loss of operational efficiency, decreased performance or malfunctions, companies want to continuously monitor the drivelines of their rotating machines and vehicles to detect deviations.
Recognising deviations is not always easy. Changes in the environment or processes can cause measurements to malfunction. If your car makes a strange noise, is it because you are driving on a bumpy road (context), or is your tyre flat (anomaly)?
In the CAATS_ICON project we are working together with Bekaert, Oqton, Sentigrate, Vintecc and Waylay on robust and accurate detection of anomalous behaviour of machine and vehicle drives. These companies will also use the project results in their operational environment.
In this ICON project we are developing a framework using an AI model that detects anomalies based on historical data. To learn to distinguish between context and anomalies, we create a feedback loop to the operator of the machine for unknown cases. The operator gets to see the available information in an orderly fashion and can decide whether it is an error, normal behaviour or a change in context. The AI model is then automatically re-trained and will recognise similar behaviour in the future, eliminating the need for intervention by an AI specialist.
- Reduce the dependence or need for an AI expert/data scientist.
- Improve accuracy of deviation detection algorithms.
- Reduce implementation time by automating the maintenance of the AI model.