ReLkat is co-funded by the European Fund for Regional Development [EFRE]
Reinforcement Learning for Complex Applications of Automation Technology (ReLkat)
Saving energy plays a central role in all industries for ecological and economic reasons. With intelligent local control in buildings, infrastructure and production, efficiency can be significantly improved. Instead of elaborate manual configuration, their automation using machine learning (ML) is an option. An alternative to the currently popular cloud approaches is local learning, with reinforcement learning (RL) playing a key role.
The research project ReLkat examines the possibilities of introducing RL into locally operating systems. Along with the development of new mathematical processes, RL basic modules for the IoT and the use of RL in industrial control technology are created. A slim RL library is being newly developed for use on a large number of IoT devices.
With XONBOT®, Signal Cruncher already has a field-tested RL analysis core that is to be expanded for IoT applications. The core is intended to be applied to a broad class of control tasks. The Fraunhofer IPK researches the XONBOT developments and their possible uses in the field of smart factories. How can industrial controls be connected to RL to solve current industrial problems? The answer includes integration concepts, uniform optimization descriptions and integration into industrial communication technology. A testbed enables simulation-based testing of all developed modules. WIAS Berlin is involved in this project with a focus on optimization under complex constraints and TU Berlin for learning processes using hierarchical tensor networks.
Funding format: Programm zur Förderung der Forschung, Innovationen und Technologien – ProFIT
- Fraunhofer-Institut für Produktionsanlagen und Konstruktionstechnik (IPK), Process Automation and Robotics Department
- Signal Cruncher GmbH
- Weierstraß-Institut für Angewandte Analysis und Stochastik, Leibniz-Institut im Forschungsverbund Berlin e.V. (WIAS), (under contractor)
- Development of a real-time control module based on reinforcement learning (RL)
- Enhancement for Heating, Ventilation and Air Conditioning (HVAC) in order to derive a local solution minimizing consumption of energy
- Combination with learning of user’s behavior to add personalization
- Minimization of overall energy cost in complex industrial production systems, development of a robust integration framework based on industrial standards like AutomationML and OPC-UA
- Establishment of RL-based framework for real-time control of IoT devices
- Fully automatic solution for local control of HVAC tasks in smart buildings
- Toolkit for energy-efficient control in smart factories including co-simulation
- Validation of results in real-life smart building and smart factory applications, starting with laboratory tests