Application of artificial neural networks to map the mechanical response of a thermoplastic elastomer

Thermoplastic elastomers are materials widely used in engineering applications due to their excellent performance to absorb mechanical vibrations and to reduce impact forces. However, their mechanical response is non-linear, which prevents linear models from predicting stresses reliably in the design and analyses of mechanical parts. This study presents a feedforward artificial neural network that was trained with stress/strain data of a thermoplastic elastomer. Such data come from a database specialized in materials from which ten curves were obtained to train and to develop an artificial neural network model. Additionally, five hyperelastic models and two probabilistic neural networks were used and compared to the proposed model. The simulation results show that the feedforward artificial neural network model is the most accurate to predict the non-linear thermoplastic elastomer response because it presented a coefficient of determination (R2) of 0.996 0 and differences of 1% with respect to the experimental data. The artificial neural network model also served to map the stress response for a temperature range between −20 °C to 160 °C for the thermoplastic elastomer material. On this basis, the presented feedforward neural network approach was tested by predicting the response of seven additional thermoplastic elastomers. The results showed that such an approach can attain thermoplastic elastomers responses with differences of 4% respect to experimental data. Consequently, the proposed approach simplifies the prediction of stress/strain curves for thermoplastic elastomer materials.