The complexity of modern production plants opens up optimization potential that can be tapped using machine learning (ML) methods. With ML, unknown relationships can be identified with the help of data, processes can be modelled in a data-driven manner and self-learning mechanisms can be realised, which make plants flexible and quickly changeable.

In order to make the broad ML method spectrum available for complex production processes across the board and in the long term, the Fraunhofer lead project »Machine Learning for Production (ML4P)«, which was launched at the beginning of 2018 and is scheduled to run for four years, is developing a tool-supported process model and corresponding interoperable software tools. In response to the challenges in the industrial environment, the frequently unstructured plant and process data, which are often available in an unstructured form, are continuously and systematically collected together with relevant expert knowledge, formalized and prepared in line with ML-specific requirements. Existing optimization potentials in production can thus be identified, evaluated and addressed with suitable, application-specific ML processes. Furthermore, it is being investigated how in the future intelligent plant components can be systematically considered already in the engineering process of new plants.

The consortium consisting of the Fraunhofer Institutes IAIS, IFF, ITWM, IWM and IWU under the leadership of the IOSB is interdisciplinary and cross-connected (ICT, production and material) and enables, in addition to process model, method and software development, the validation of the project results in three demonstrators of different production plants.