Constitutive models for predicting field-dependent viscoelastic behavior of magnetorheological elastomer using machine learning

Abstract

This work presents new constitutive models of a magnetorheological (MR) elastomer viscoelastic behavior using a machine learning method to predict the magnetic field dependent-stiffness and damping properties. The multiple output-models are formulated using two basic neural network models, which are artificial neural network (ANN) and extreme learning machine (ELM). These models are intended to capture the non-linear relationship between the inputs consisting of shear strain and magnetic flux density and outputs, which are storage modulus and loss factor. The optimized model is firstly identified by varying the model parameters, such as the number of hidden nodes and activation functions for both proposed prediction models. Then, the model performances were evaluated for training and testing data sets. The results showed that ANN and ELM prediction models had performed differently on two different outputs. The performance of the ANN prediction model was significant in predicting storage modulus where the root mean square error (RMSE) and coefficient of determination (R2) of testing data out of modeling data sets were 0.012 MPa and 0.984 respectively. Meanwhile, the ELM model shows good agreement in predicting loss factor where the RMSE and R2 were 0.007 MPa and 0.989, respectively. These machine learning-based models have successfully proved its high accuracy prediction that can be further applied to distinguish the linear viscoelastic (LVE) region and predict the damping properties of MR elastomer.