|
Authors
|
Gahlen, P. ; Mainka, R. ; Stommel, M.
|
|
Title
|
Prediction of anisotropic foam stiffness properties by a Neural Network
|
|
Date
|
10.03.2023
|
|
Number
|
61521
|
|
Abstract
|
In this work, an Artificial Neural Network is developed to predict the orthotropic stiffness tensor of anisotropic foam structures utilizing a tessellation-based foam RVE database. Based on a validated mesoscale simulation model, which is able to predict the mechanical properties of foams including the consideration of the cell structure, in a first step a simulative database is created. Variable parameters are: Foam density, Young’s modulus of foam base material, cell size, cell orientation, cell shape, etc. In total, more than 2000 simulations were performed leading to a novel RVE database. These simulations are subsequently used to train and compare different networks in a supervised manner. It can be seen that with suitable network settings a Mean Relative Error of max 2% compared to the RVE simulations occurs. As a result, the anisotropic mechanical stiffness values of complex foam structures can be determined via the Artificial Neural Network within seconds instead of performing time-consuming simulations (up to hours). In a final study, the effect of the database size on the prediction accuracy is examined. It can be observed that at least 500 training datapoints are required to obtain sufficient accuracy.
|
|
Publisher
|
International Journal of Mechanical Science
|
|
Wikidata
|
|
|
Citation
|
International Journal of Mechanical Science 249 (2023) 108245
|
|
DOI
|
https://doi.org/10.1016/J.IJMECSCI.2023.108245
|
|
Tags
|
|
