A Supervised Artificial Neural Network-Assisted Modeling of Magnetorheological Elastomers in Tension–Compression Mode

Modeling of highly sophisticated behavior of magnetorheological elastomers (MREs) is an essential step toward optimally designing and effectively controlling the smart material-based devices. While modeling MREs in shear mode has been widely carried out by employing continuum mechanics, mathematical techniques, and phenomenological approaches, the correct determination of dynamic behavior of MREs in tension–compression mode has been addressed in only a few studies due to inherent complexities mainly arising from the computational demandingness of the process. This article addresses the functionality of artificial neural network (ANN) for prediction of MRE’s dynamic behavior in tension–compression mode under different levels of strain, frequency, and magnetic flux density. A multilayer perceptron-based feed-forward neural network with backpropagation training technique was used with various structures to identify an optimal configuration. A neural network structure with 20 neurons in the hidden layer was adopted, which revealed the mean square error (MSE) magnitude of 7.1 kPa with $R^{2}$ values above 0.97. Afterward, the predicting capacity of the model was evaluated using experimental data sets. The obtained results are suggestive of reasonably acceptable performance of the proposed ANN model, which holds the capacity for a close mapping of the predicted tension–compression stress values to those of experimental ones. Further development of the proposed ANN model serves as a promising approach to deal with the modeling and controlling of engineering devices equipped with tension–compression MREs.