The use of neural networks and nonlinear finite element models to simulate the temperature-dependent stress response of thermoplastic elastomers

Abstract

In this study, a methodology that combines artificial neural networks and nonlinear hyperelastic finite element modeling to simulate the temperature-dependent stress response of elastomer solids is presented. The methodology is verified by a discrete model of a tensile test specimen, which is used to generate stress–strain pairs of existent experimental data. The proposed method is also tested with a benchmark problem of a rubber-like cylinder under compression. Three grades of an elastomer used for diverse engineering applications are used throughout the study. On this basis, three neural network architecture with 10 hidden neurons are implemented as constitutive models to reproduce the experimental data of the materials. The validation results show that the proposed methodology can reproduce tensile tests with an error of 5% of less than regarding experimental data for elastomers that present no yielding point. The benchmark problem results were at the range expected for the elastomer materials with no yielding, where it was possible to derive force temperature-dependent responses. These results suggest that the methodology helps the prediction of the material response when only material stress–strain curves at different temperatures exist. Therefore, the presented approach in this contribution helps to simulate the temperature-dependent stress responses of elastomeric solids with no defined yielding point.

A machine learning approach to estimate the strain energy absorption in expanded polystyrene foams

“…Moreover, a well trained ANN can simulate material responses other than those used for its training, but within the limits of the extreme values of the labeled inputs. In this aspect, it has been argued in the past that this appeals to an" exploration" of the circumstances and conditions of the factors studying the phenomenon, which ultimately allows to understand why neural networks are more well-suited to study stress/strain responses than the formal descriptions given by hyperelastic models (for the ANNs argumentation of the exploration concept see the work by Cichy and Kaiser in 2019; 37 for an example of a comparison of ANNs and hyperelastic models, see the work by Rodr ıguez-Sa´nchez et al 38 )…”

Comparative Study of Machine Learning Approaches for Predicting Creep Behavior of Polyurethane Elastomer

“…In contrast, five hyper-elastic models, the neural network model had the best accuracy with a coefficient of determination R = 0.996 and 1% difference from the experimental data. The ANN was combined with the nonlinear hyper-elastic finite element model to simulate the temperature-dependent stress response of elastomer solids in their following research [ 20 ]. Stoffel et al [ 21 ] developed a series of ANN including a feedforward neural network, radial basis function neural network and a deep convolutional neural network to predict the structural deformations by comparison to experiments.…”

Experimental investigation and optimization of quality characteristics during friction stir welding of Al- and Zn-based magnesium alloy using artificial intelligence

“…Lakshminarayanan et al 22 compared the quadratic model of response surface methodology (RSM) and linear model of ANN to predict the TS of FS-welded Al 7039 alloy and reported better results with an ANN model. The ANN model was also used by many other researchers 23–26 in their works. They reported a good agreement between predicted and experimental results.…”