Stiemer, M.Nezhi, Z.Rathjen, K.Zazai, F.Hagel, M.Rozgic, M.2022-01-112022-01-112021-10-14http://hdl.handle.net/2003/40660http://dx.doi.org/10.17877/DE290R-22518In this work, approaches to the identification of high speed forming processes, whose simu lation requires models from different parts of physics are discussed. Particularly emphasis is laid on situations in which it is possible to break off the coupling and to profit from partial solutions for the design of the whole process. Such situations arise if it is possible to select relevant features that allow for a stable transfer of information between the different models. Creating situations in which a sequential approach to a coupled problem is favourably pos sible requires a profound process understanding. As an example, an electromagnetic form ing process is considered here. Approaches at identifying a coil geometry for electromag netic forming are discussed in case of an exemplary case involving the definition of a suitable feature-list and the study of several methods to tackle the electromagnetic subproblem, in cluding Nelder Mead Simplex Search, a combination of it with a neural network as surrogate model, and optimization via a neural network. These approaches are compared to each other, and quantitative results are given.enmetal formingmachine learningfinite element methoddesign optimization620670Text