Data Analysis Systems
The mission of the “Data Analysis Systems” group (DAS) is to advance methods for the analysis and modelling of complex and massive (sensor) data in its (industrial) application context. Applied research is carried out in the following areas:
- Big Data Processing (e.g. processing (big) complex data across multiple servers),
- Process Data Analysis (e.g. better understanding of single process steps; virtual sensors),
- Fault Detection and Diagnosis (e.g. detect faulty components, machines... through data analysis and identify causes),
- Predictive Analytics (e.g. data-driven fault prediction for Predictive Maintenance), and
- Control and Optimization (e.g. use of extracted knowledge from data analysis to optimize systems (production, machines...)).
The common task behind all these applications is the extraction of information and knowledge about the process, product or machinery from operational data and utilizing this knowledge to build robust computational / predictive models from data, a.k.a. machine learning, data mining and knowledge discovery.
The applied research in DAS concentrates on methods which contribute to answers for highly relevant data analysis questions like: How can mechanistic process understanding and causal correlation be inferred from data? How can we efficiently transfer models from one setting to another without having to gather another expensive set of data? How can knowledge from ad hoc data analysis sessions be modelled, consolidated and algorithmically be used in the next session? Answers to these questions cast into software allow us to more efficiently provide data-driven solutions to problems of our (industrial) partners. E.g. a tool for mechanistic model inference enables reasoning about causes and effects in industrial production processes for non-data-scientists, while generic transfer learning solutions would enable the deployment of large numbers of virtual sensors without expensive recalibration.