Grey Box Model – Integrating Application Knowledge in the Learning Process

"Machine Learning processes require huge volumes of data if they are to work well. However, this is not always available in the manufacturing sector. Yet, we do have a great deal of knowledge about the processes," said project coordinator Michael Bortz. "If we take this into account and combine it with machine-learning methods, we can make model-based predictions and enable more reliable and substantially improved optimization designs. This is the basic concept explored in our project."

Sample Application Fields

We have selected three application fields to illustrate the benefits of model-based machine learning:

• Chemical engineering process
• Pattern recognition in process data
• Forming technology
  

By combining and further developing the methodological competence at the participating institutes, three software prototypes are being created to demonstrate the benefits of the grey box method in finding solutions to industrial challenges.

Advantages

Machine learning has achieved impressive successes over the years in various areas, to include pattern recognition in the broadest sense, for example, text and image recognition, object classification, and personal identification.

However, machine learning applications often exclusively use a black box approach, that is, they rely on a completely non-parametric approach while no simple parametric model exists. Mathematical modeling, simulation and optimization in industry, on the other hand, are usually based on white box approaches, meaning parametric models.

The grey box model combines qualitative prior knowledge with quantitative data. This approach uses all available information about a certain industrial process to determine the best possible process model. Compared to the black box and white box models, the grey box method has the advantage of using previous knowledge and the information available from existing data.

Steps in grey box modeling

• Apply white box and understand model boundaries
• Learn black box model and test extrapolation validity
• Integrate a grey box model
• Optimize the grey box model with simulation and actual data reconciliation
• Smart model implementation

Involved Institutes

• Fraunhofer-Institut für Techno- und Wirtschaftsmathematik ITWM
• Fraunhofer-Institut für Algorithmen und Wissenschaftliches Rechnen SCAI
• Fraunhofer-Institut für Optronik, Systemtechnik und Bildauswertung IOSB
• Fraunhofer-Institut für Intelligente Analyse- und Informationssysteme IAIS
• Fraunhofer-Institut für Werkstoffmechanik IWM

Project Coordinator:

• Dr. Michael Bortz (ITWM)
• Prof. Dr. Jochen Garcke (SCAI)

Project Duration:

The project will run for a period of two years.