Fraunhofer-Gesellschaft

Publica

Hier finden Sie wissenschaftliche Publikationen aus den Fraunhofer-Instituten.

Application and evaluation of meta-model assisted optimisation strategies for gripper assisted fabric draping in composite manufacturing

 
: Zimmerling, Clemens; Pfrommer, Julius; Liu, Jinzhao; Beyerer, Jürgen; Henning, Frank; Kärger, Luise

:
Volltext (PDF; )

European Society for Composite Materials -ESCM-:
18th European Conference on Composite Materials, ECCM 2018. Online resource : Athens, Greece, June 24-28, 2018
Athens, 2018
https://pcoconvin.eventsair.com/QuickEventWebsitePortal/eccm/program/Agenda
ISBN: 978-151089693-2
8 S.
European Conference on Composite Materials (ECCM) <18, 2018, Athens>
Englisch
Konferenzbeitrag, Elektronische Publikation
Fraunhofer ICT ()
Fraunhofer IOSB ()
artificial intelligence; deep learning; Deep neural network; Manufacturing Optimisation; meta-modelling

Abstract
With respect to their extraordinary weight-specific mechanical properties, continuous Fibre Reinforced Plastics (CoFRP) have drawn increasing attention for use in load bearing structures. Contrasting metals, manufacturing of CoFRPs components requires multiple steps, often including a draping process of textiles. To predict and optimise the manufacturing process, Finite-Element (FE) simulation methods are being developed along virtual process chains. For maximum part quality, draping process parameters need to be optimised, which requires numerous computationally expensive iterations. While efforts have been made for time-efficient process optimisation in metal forming, composite draping optimisation has is a comparably young discipline and still lacks time-efficient optimisation strategies. In this work, modelling strategies for time-efficient optimisation using computationally inexpensive meta-models are examined, which are used to guide the search for optima in the parameter space. The meta-models are trained by observations of FE-based draping simulations of an automotive part, thereby learning the relationship between variable gripper forces (input) and the resulting shear angles (output). Parametric model functions are compared against deep neural networks (DNN) as non-parametric models with respect to prediction accuracy. Best results are achieved using a DNN that predicts the shear angles of more than 24 000 fabric shell elements.

: http://publica.fraunhofer.de/dokumente/N-582975.html