Workflow concepts to model nonlinear mechanics with computational intelligence. Issue 1 (14th December 2021)

• Record Type: Journal Article
• Authors: Koeppe, Arnd; Bamer, Franz; Selzer, Michael; Nestler, Britta; Markert, Bernd
• Editors: Kaliske, M.
• Is Part Of: Proceedings in applied mathematics and mechanics. Volume 21:Issue 1(2021)
• Page Start: n/a
• Page End: n/a
• Publication Date: 2021-12-14
• Subjects: Applied mathematics -- Periodicals; Engineering mathematics -- Periodicals; Mathematical physics -- Periodicals; 519
• DOI: 10.1002/pamm.202100238
• Languages: English
• ISSNs: 1617-7061

Abstract

Abstract: Methods from computational intelligence, such as (artificial) neural networks, have become an active research direction in mechanics, leading to the development of intelligent constitutive models, surrogate models, and meta elements. Therein, many neural network architectures are inspired by mechanical domain knowledge in the form of physics‐informed or physics‐guided neural networks. Complementary approaches that systematically analyze and compare neural networks trained on mechanical data, i.e., physics‐informing neural networks, have not yet been established. As a step in that direction, we propose a workflow concept to describe neural networks in mechanics, as well as a workflow concept to systematically search and train neural network architectures on mechanical data. The workflow concepts will be presented in the scope of Kadi4Mat. Following these workflow concepts, neural network design can be unified, compared, and interpreted, which enables explainable artificial intelligence for mechanics in future works.