Testing the Generalizability of Deep Learning Based Plausibility Detection with Unknown Finite Element Simulations
DS 130: Proceedings of NordDesign 2024, Reykjavik, Iceland, 12th - 14th August 2024
Year: 2024
Editor: Malmqvist, J.; Candi, M.; Saemundsson, R. J.; Bystrom, F. and Isaksson, O.
Author: Bickel, Sebastian; Stefan, Goetz; Sandro, Wartzack
Series: NordDESIGN
Institution: Friedrich-Alexander-Universitat Erlangen-Nurnberg, Engineering Design, Germany
Page(s): 51-60
DOI number: 10.35199/NORDDESIGN2024.6
ISBN: 978-1-912254-21-7
Abstract
Shorter development times and greater product variety are necessary in the current business world. These challenges can be managed with virtual testing methods, but these require experienced engineers. To counteract the shortage, a data-driven plausibility detection was developed. The plausibility is determined with Deep Learning models that are trained on existing simulations. The practical application requires the models to work with simulations not included in the training data. Therefore, this paper analyzes the transferability of a plausibility detection model to new, unknown instances.
Keywords: Artificial Intelligence (AI), Data Driven Design, Structural Analysis, Deep Learning, Finite Element Simulation