Concept for machine learning and mixed reality in digital twins of production machines

Virtual machine and system models (digital twins) enable the development, optimization, commissioning and expansion of systems, production sequences or processes and support predictive planning or maintenance without real hardware components.

The prerequisite for this is the most accurate possible virtualization of the physical systems and the controllers used. At present, the virtual behavior models have to be specially developed (usually manually). SCADA systems, which are usually present in the real scenarios and have all the data on the system states, are not or only insufficiently networked with the digital twins and therefore cannot be used for model development and optimization, or the digital twins are not available for prediction in the control systems of the real systems.

In this project, the following methods and interfaces are therefore to be developed and tested on a component basis:

- Method development and module for machine learning of system properties or controller behavior for digital twins based on the coupling with SCADA systems.
- Methodology and interface for connecting a mixed reality module for testing the correctly learned behavior and for human-centered testing of human-machine interaction. It can also be used to carry out virtual commissioning scenarios and maintenance procedures in the iterative development process.
- Development of a standardized, real-time-capable interface between digital twins and SCADA or control systems including an engineering interface to reduce the development effort using a standard technology such as OPC UA or MQTT.
- Investigation of control loops from SCADA and digital twin to optimize processes or parameters. Based on a real-time interface, the aim is to investigate whether optimizations can be made while the process is still running in order to improve the process result.

The results will be demonstrated in a prototype implementation in the project partners' laboratories using defined use cases.