Speaker
Description
The growing demand for high-performance materials has led to significant advancements in engineering composite structures, such as laminates, and fiber-reinforced materials, and novel developments such as voided slabs for reinforced concrete. While these materials optimize the structural performance by taking advantage of the unique properties of the different constituents, developing effective material models based on classical phenomenological approaches remains challenging. Full 3D solutions on the other hand, under consideration of the complex microstructures, quickly encounter the bottleneck of a high computational effort. Alternative approaches, as developed in, e.g., [1] for shell structures, employ multiscale techniques, based on consistent homogenization schemes, which effectively capture the microstructural constitutive response at each macroscopic integration point. In this contribution, we seek to further reduce the computational cost by taking advantage of the consistent homogenization scheme in order to train an artificial neural network (ANN) material model for arbitrary history-dependent stress-strain paths. As ANNs are capable of capturing inelastic material behavior, see. e.g. [2], this work focuses on the application to high dimensional strain-stress relations for shell structures with underlying viscoelastic microstructures. We demonstrate, that a small database comprising uniaxial synthetic material tests on a representative microstructure in combination with derivative information is sufficient to train a feasible ANN material model [3]. Furthermore, we explore the limits of the approximation capabilities of the ANN by including non-physical parameters of the microstructure, e.g. volume fractions, as inputs to the material model. Studies include comparisons with full-scale and multiscale models, highlighting computational efficiency and practical feasibility in application to real-world engineering problems involving complex microstructures.
[1] Gruttmann, F. and Wagner., W.: A coupled two-scale shell model with applications to layered structures. International Journal for Numerical Methods in Engineering 94(13), pp. 1233-1254, 2013.
[2] Rosenkranz, M, Kalina, K.A., Brummund, J., Kästner, M.: A comparative study on different neural network architectures to model inelasticity. International Journal for Numerical Methods in Engineering 124(21), pp. 4802-4840, 2023.
[3] Geiger, J., Wagner, W., Freitag, S.: Multiscale modeling of viscoelastic shell structures with artificial neural networks. Computational Mechanics 2025, under review
