Abstract Metal forming processes like open-die forging or hot rolling are well-established for the production of key components in various industries. Nevertheless, the control of the final microstructure and hence mechanical properties is not yet common. To achieve this, the authors propose and discuss a control concept based on reinforcement learning, fast process models (FPM) and an “operator in the loop” approach. The concept is explained and tested using deviating initial ingot temperatures as idealized process disruptions. RL algorithms are trained for both processes and transferred into the controllers that are connected to a simulative environment based on FPM. Within this framework, the online adaption is possible in ∼2 s in rolling and 4–6 s in forging. This highlights the concepts suitability to be used for property control in hot metal forming.
Read full abstract