Industrial Reinforcement Learning for the Quality Control of Bulk Metal Forming Processes – IRLeQuM

Instabilities in bulk metal forming processes lead to rejects despite existing process controls due to external influencing variables and unknown interactions between process parameters and product quality characteristics. Current control concepts are based on implicit operator knowledge and rely on manual adjustment of the process parameters. It is often not possible to adjust the processes in time to meet the manufacturing tolerances of the products. Quality control loops (QCL) are a tool for compensating quality deviations. In combination with machine learning approaches, QCL offer the potential to reduce rejects.

The research project IRLeQuM aims to develop a method for implementing novel controllers based on reinforcement learning (RL) and transfer learning (TL) in QCL of bulk metal forming processes.
In order to enable RL-based control, the necessary IT infrastructure is first defined and implemented. RL control offers the advantages that, on the one hand, all quality-relevant information (e.g. process parameters, ambient conditions or raw material properties) can be included in the control. On the other hand, the implicit operator knowledge of the control becomes permanently usable.

In order to reduce the learning time of the RL algorithm and to save resources, the algorithm is not trained directly on the real process but on a stochastic process simulation. The knowledge gained from the simulation is then transferred to the QCL of the control loop using TL.
The result of the research project will be a quality control of bulk metal forming processes, which controls processes automatically, comprehensively and in real time and optimises the processes’ quality. The improved process quality increases the quality of the products and reduces rejects.