Fraunhofer Institute for Industrial Mathematics ITWM
Fraunhofer Institute for Industrial Mathematics ITWM
Machine Learning for Production
© iStockphoto
Machine Learning for Production

ML4P

In the Fraunhofer Lighthouse Project, seven Fraunhofer Institutes bundle their extensive experience in the field of Machine Learning. This is because there is a need both in the process industry and in the general cargo manufacturing industry.

© iStockphoto

Machine Learning for Production

Fraunhofer Lighthouse Project ML4P

Modern production plants are by now highly complex. Processes are networked, machines, interfaces and components communicate with each other. Such industrial plants are predestined for optimization through methods of Machine Learning (ML). They make predictions on the basis of large amounts of data possible.

In the Fraunhofer Lighthouse Project, seven Fraunhofer Institutes bundle their extensive experience in the field of Machine Learning. This is because there is a need both in the process industry and in the general cargo manufacturing industry.

With the help of Machine Learning, we can:

  • Learn hidden relationships
  • Model processes
  • Implement adaptive machansims that make plants flexible and quickly convertible.
Logo ML4P
Machine Learning for Production gehört seit Ende 2017 zu den Fraunhofer-Leitprojekten.

Data Bundled with Expertise

In contrast to the application domains of Machine Learning, where gigantic amounts of data are available (Internet, social media, etc.), in an industrial context you have "only a lot" of data, but usually additional expert knowledge. In a consistent plant optimization, both must be used: all available data and the entire expert knowledge. In the industrial context, not only the current trend topic of Deep Learning is therefore interesting, but also a wide variety of other special ML methods which can also handle comparatively less data while at the same time using previous knowledge.

In the ML4P project, we are formulating intelligent methods for meeting the needs of industry and preparing the way for flexible, fast-learning systems. A "learning machine" could, for example, include the installation of intelligent components or the efficient, holistic handling of very large amounts of data.

Goals of the Project ML4P:

  • Development of a tool-supported process model
  • Realization of a software tool to record and analyse the current situation in order to identify possible optimization potentials
  • Selection of suitable methods of Machine Learning in production

Such a tool-based process model for the industrial use of ML does not yet exist, so that the project "ML4P" results in a unique position for Fraunhofer. The composition of the participating institutes is interdisciplinary. The competencies complement each other in a meaningful way and, in addition to process model, method and software development, allow the validation of project results in three sophisticated demonstrations of different production plants.

 

Tasks of Our Institute

Three departments of our institute work together in the ML4P project with the most varied areas of responsibility: »Optimization«, »Financial Mathematics« and »Image Processing«.

As a rule, production plants consist of many individual networked components. The aim is to create a model of the entire plant in order to propose improved plant designs or operating methods on the basis of this overall model using methods of mathematical optimization. Both physical model knowledge and methods of Machine Learning are used for this purpose, whereby these learning procedures are based on complex simulation data and on measured operating data of the production plants. The operating data can come from sensorial monitoring as well as from the documentation of the operating conditions. In particular, the application of ML methods in the statistical analysis of time series and the automatic analysis of image data are focal points.

 

Involved Fraunhofer Institutes:

 

Project Duration:

12/01/2017 – 11/30/2021