摘要：Abstract Reduced models of manufacturing processes are essential for online control of these processes. By generating a low-dimensional, nonlinear process representation, we extract process state features, ordered by relevance, that can be used for the construction of a reduced model. In this paper, we propose new approaches for nonlinear dimensionality reduction based on sequential bottleneck neural networks. The extracted features represent the state in a low-dimensional nonlinear subspace of the original process representation space. The feasibility of these approaches is shown with data of a numerically simulated deep drawing process.
