Numerical characterisation of uncured elastomers by a neural network based approach
•Model-free neural network approach captures large strain inelasicity.•Inelastic model-free approach represented by recurrent neural network.•Adaptive recurrent neural network represents different kinds of inelasticity (viscous, plastic and visco-plastic).•Novel combination of neural network based m...
Gespeichert in:
Veröffentlicht in: | Computers & structures 2017-04, Vol.182, p.504-525 |
---|---|
Hauptverfasser: | , |
Format: | Artikel |
Sprache: | eng |
Schlagworte: | |
Online-Zugang: | Volltext |
Tags: |
Tag hinzufügen
Keine Tags, Fügen Sie den ersten Tag hinzu!
|
Zusammenfassung: | •Model-free neural network approach captures large strain inelasicity.•Inelastic model-free approach represented by recurrent neural network.•Adaptive recurrent neural network represents different kinds of inelasticity (viscous, plastic and visco-plastic).•Novel combination of neural network based material formulation and micro-sphere framework.•Application of novel model-free approach to describe uncured rubber material.•Different strategies to identify the parameters of the neural network.•Performance demonstrated by numerical characterisation of uncured rubber.•Analysis of a rubber forming process.
This research paper contributes to the model-free characterisation of elastic and inelastic materials. The introduced material description is based on neural networks for the constitutive stress-strain-relationship and is an alternative to a classical constitutive material description. This approach is an efficient method to represent the material behaviour numerically. The performance of the novel model-free formulation is exemplary shown for uncured natural rubber. One major advantage of the presented approach is the capability to use it for the representation of a wide range of materials.
The novel model-free characterisation consists of a neural network that is coupled to the so-called micro-sphere approach. This formulation was developed for the characterisation of rubber like materials and takes the micro-structure of the material into account. As a benefit of this coupling, the neural networks, representing the stress-stretch-dependency, have to be exploited analysed only in onedimensional direction. In the first instance, the derivation is introduced for an Artificial Neural Network to yield a pure elastic description. Subsequently, the model-free approach is expanded in order to represent inelastic material behaviour as well. This extended formulation is obtained via a Recurrent Neural Network.
Finally, uncured natural rubber material is described by the derived numerical approach. The model-free characterisation is validated by the finite element simulation of material tests. A complex forming of a rubber block into a mould is basis for a final validation of the model-free description. |
---|---|
ISSN: | 0045-7949 1879-2243 |
DOI: | 10.1016/j.compstruc.2016.12.012 |