Bauinformant bloggt Beton

BETON aus Leidenschaft und Profession

Neural networks in a multiscale approach for concrete

leave a comment »

von: Jörg F. Unger
Institut: Institut für Strukturmechanik
Fakultät: Fakultät Bauingenieurwesen
Dokumentart: Dissertation
Betreuer/Doktorvater: Könke, Carsten, Prof. Dr.-Ing. habil.
Sprache: Englisch
Erstellungsjahr: 2009
Publikationsdatum: 26.06.2009

Kurzfassung in Englisch:

From a macroscopic point of view, failure within concrete structures is characterized by the initiation and propagation of cracks. In the first part of the thesis, a methodology for macroscopic crack growth simulations for concrete structures using a cohesive discrete crack approach based on the extended finite element method is introduced. Particular attention is turned to the investigation of criteria for crack initiation and crack growth.

A drawback of the macroscopic simulation is that the real physical phenomena leading to the nonlinear behavior are only modeled phenomenologically. For concrete, the nonlinear behavior is characterized by the initiation of microcracks which coalesce into macroscopic cracks. In order to obtain a higher resolution of this failure zones, a mesoscale model for concrete is developed that models particles, mortar matrix and the interfacial transition zone (ITZ) explicitly. The essential features are a representation of particles using a prescribed grading curve, a material formulation based on a cohesive approach for the ITZ and a combined model with damage and plasticity for the mortar matrix.

Compared to numerical simulations, the response of real structures exhibits a stochastic scatter. This is e.g. due to the intrinsic heterogeneities of the structure. For mesoscale models, these intrinsic heterogeneities are simulated by using a random distribution of particles and by a simulation of spatially variable material parameters using random fields.

There are two major problems related to numerical simulations on the mesoscale. First of all, the material parameters for the constitutive description of the materials are often difficult to measure directly. In order to estimate material parameters from macroscopic experiments, a parameter identification procedure based on Bayesian neural networks is developed which is universally applicable to any parameter identification problem in numerical simulations based on experimental results. This approach offers information about the most probable set of material parameters based on experimental data and information about the accuracy of the estimate. Consequently, this approach can be used a priori to determine a set of experiments to be carried out in order to fit the parameters of a numerical model to experimental data.

The second problem is the computational effort required for mesoscale simulations of a full macroscopic structure.
For this purpose, a coupling between mesoscale and macroscale model is developed. Representative mesoscale simulations are used to train a metamodel that is finally used as a constitutive model in a macroscopic simulation. Special focus is placed on the ability of appropriately simulating unloading.

Quelle Abstract: Uni Weimar, externer Link

>>Download Volltext

Bauinformant ist nicht der Autor der eingestellten Information. Das Urheberrecht ist durch einen Direktlink und/oder die Quellenangabe gekennzeichnet und/oder vom Urheber zur Veröffentlichung genehmigt. Falls Sie einen Verstoß gegen das Urheberrecht vermuten, bitten wir Sie um eine persönliche Mitteilung, bevor unnötige rechtliche Auseinandersetzungen entstehen.


Kommentar verfassen

Trage deine Daten unten ein oder klicke ein Icon um dich einzuloggen:

Du kommentierst mit Deinem Abmelden / Ändern )


Du kommentierst mit Deinem Twitter-Konto. Abmelden / Ändern )


Du kommentierst mit Deinem Facebook-Konto. Abmelden / Ändern )

Google+ Foto

Du kommentierst mit Deinem Google+-Konto. Abmelden / Ändern )

Verbinde mit %s

%d Bloggern gefällt das: