Computational Homogenization Using Convolutional Neural Networks

authored by: Henning Wessels, Christoph Böhm, Fadi Aldakheel, Markus Hüpgen, Michael Haist, Ludger Lohaus, Peter Wriggers
Abstract: The classic tasks of computational engineers are to investigate and optimize structures in terms of their mechanical behavior. This iterative process usually requires a large number of calculations of different macroscopic structures of the same material. The computational time in this design-loop directly affects the time to market. Depending on the model complexity, describing the interaction between micro- and macro-scale can be computationally expensive and even prohibitive for engineering practice. This holds especially true if the physics on the micro-scale is complex involving inelastic behavior, fracture and/or phase change. In this paper, recent trends in Scientific Machine Learning (SciML), which may advance computational homogenization in the sense of the digital twin paradigm, are reviewed. We believe that SciML techniques for computational homogenization will make micro-macro simulations become applicable at lowextra cost in engineering practice. This work is partially funded by the DFG Priority Program SPP 2020 Experimental- Virtual-Lab and the DFG Collaborative Research Center SFB 1153 Tailored Forming.
Organisation(s): Institute of Continuum Mechanics
Institute of Building Materials Science
External Organisation(s): Technische Universität Braunschweig
Type: Contribution to book/anthology
Pages: 569-579
No. of pages: 11
Publication date: 13.03.2022
Publication status: Published
Peer reviewed: Yes
ASJC Scopus subject areas: Engineering(all), Computer Science(all)
Electronic version(s): https://doi.org/10.1007/978-3-030-87312-7_55 (Access: Closed)
: Details in the research portal "Research@Leibniz University"

BibTeX

@inbook{1c1f997cbadf4720b5cc3e53427e6ea7,
title = "Computational Homogenization Using Convolutional Neural Networks",
abstract = "The classic tasks of computational engineers are to investigate and optimize structures in terms of their mechanical behavior. This iterative process usually requires a large number of calculations of different macroscopic structures of the same material. The computational time in this design-loop directly affects the time to market. Depending on the model complexity, describing the interaction between micro- and macro-scale can be computationally expensive and even prohibitive for engineering practice. This holds especially true if the physics on the micro-scale is complex involving inelastic behavior, fracture and/or phase change. In this paper, recent trends in Scientific Machine Learning (SciML), which may advance computational homogenization in the sense of the digital twin paradigm, are reviewed. We believe that SciML techniques for computational homogenization will make micro-macro simulations become applicable at lowextra cost in engineering practice. This work is partially funded by the DFG Priority Program SPP 2020 Experimental- Virtual-Lab and the DFG Collaborative Research Center SFB 1153 Tailored Forming.",
author = "Henning Wessels and Christoph B{\"o}hm and Fadi Aldakheel and Markus H{\"u}pgen and Michael Haist and Ludger Lohaus and Peter Wriggers",
note = "FA, MH, MH, LL and PW acknowledge funding through the DFG Priority Program SPP 2020 Experimental-Virtual-Lab under the grants number 373757395; project WR 19/58-2 and GZ-LO 751/22-2 (668132). CB, FA and PW thank the German Research Foundation (DFG) for financial support to this work in the Collaborative Research Center SFB 1153 Process chain for the production of hybrid high-performance components through tailored forming with the subproject C04 Modelling and simulation of the joining zone, project number 252662854.",
year = "2022",
month = mar,
day = "13",
doi = "10.1007/978-3-030-87312-7_55",
language = "English",
isbn = "9783030873110",
pages = "569--579",
booktitle = "Current Trends and Open Problems in Computational Mechanics",
publisher = "Springer International Publishing AG",
address = "Switzerland",
}