Coupled Finite-Element-Method-Simulations for Real-Time-Process Monitoring in Metal Forming Digital-Twins

In many industrial processes it is not possible to measure variables that are crucial to the output quality of the manufactured goods due to manufacturing constraints. One possible approach to overcome this lack of information is using a digital-twin of the manufacturing plant. A digital-twin can make use of data generated by numerical simulation techniques to infer the real time state of the machine. Digital-twins are an established technology and will play a crucial role in optimizing manufacturing plants in the near future. The digital real time state of the manufacturing plant can be used to warn machine operators in case a predefined process window is not reached and the output is expected to violate Original Equipment Manufacturer (OEM) requirements. This work shows a developed control system and how using a digital twin improves the predictive power of the system. Machine learning models are used to infer the cooling curves of metal sheets in a press hardening process. They are trained on coupled Finite-Element-Method (FEM) Simulation data. We validate the performance of our models on the simulation data and find accurate estimations of the metal blanks temperature with a root mean squared error of 1.6±0.5 K, with this not being possible to measure beforehand.