Coupling of scales in a multiscale simulation using neural networks

• Published in: Computers & structures Vol. 86; no. 21; pp. 1994 - 2003
• Main Authors: Unger, Jörg F., Könke, Carsten
• Format: Journal Article
• Language: English
• Published: Kidlington Elsevier Ltd 01.11.2008; Elsevier
• Subjects: Applied sciences; Buildings. Public works; Computational techniques; Concretes. Mortars. Grouts; Constitutive modelling; Exact sciences and technology; Fundamental areas of phenomenology (including applications); Homogenization; Materials; Mathematical methods in physics; Multilayer perceptron; Neural network; Physics; Solid mechanics; Structural and continuum mechanics; Support vector machines; Support vector machines; Homogenization; Multilayer perceptron; Neural network; Constitutive modelling; Constitutive equation; Sensitivity analysis; Modeling; Mesoscale; Multiscale method; Vector support machine; Multilayer perceptrons; Concrete; Mesh generation
• ISSN: 0045-7949; 1879-2243
• DOI: 10.1016/j.compstruc.2008.05.004

Multiscale approaches require the coupling of models on different spatial scales. In this paper, a coupling using neural networks is proposed. Based on a set of mesoscale simulations of concrete a system of neural networks is trained to approximate the response. A macroscale constitutive model is obtained by homogenizing the mesoscale response. Special focus is put on the mesh sensitivity, since the mesoscale model includes softening and consequently the width of the localization zone compared to the dimension of the mesoscale model is an additional parameter in the model.