|About this Abstract
||NUMISHEET 2022: The 12th International Conference on Numerical Simulation of 3D Sheet Metal Forming Processes
||Learning Product Properties with Small Data in Forming Simulations
||Rodrigo Iza Teran, Jochen Garcke, Lukas Morand, Dirk Helm
|On-Site Speaker (Planned)
||Rodrigo Iza Teran
We present an approach that can learn from simulation data in the case where only small data sets are available. To some extent, the approach can predict product properties locally in such data sparse situations. For the approach, a representation of the underlying geometry is computed that stays invariant under certain type of transformations. Once found, the representation is used as a new orthogonal basis to represent all available time dependent deformations. The representation is based only on the geometry and is therefore independent of the simulation data. Here, we apply the approach to metal forming simulations. Machine-learning methods applied to simulation data can be used to get a better understanding of the relation between process parameters and process outcome. Nevertheless, the computational costs are an issue and limit the generation of numerical data especially for complex processes.
||Definite: At-meeting proceedings