MULTISCALE MODELING APPROACH FOR PREDICTING THE HYPERELASTIC RESPONSE OF ACRYLONITRILE-BUTADIENE RUBBER COMPOSITES WITH ARTIFICIAL NEURAL NETWORK

Abstract

A novel framework is presented to predict the hyperelastic behavior of acrylonitrile-butadiene rubber (NBR) composites at continuous strain rates (SRs) using an artificial neural network (ANN) assisted by multiscale modeling. Initially, the macroscopic experimental stress–strain data of NBR gum rubber at various SRs were curve-fitted using an appropriate material model. The resulting smoothened curves were then piecewise linearized by computing local tangents at multiple strain intervals, yielding a discrete set of effective elastic moduli for the gum matrix. Assuming linear elastic behavior for both matrix and filler, these moduli were used in a microscale model based on the representative volume element (RVE). The discrete set of predictions by the RVE model across strain intervals was then used as additional input in the ANN model to enhance its prediction accuracy. When applied to NBR composites, the model achieved significantly lower root mean square error values (<0.071) and higher R 2 values (>0.999) across various SRs, even while trained with a considerably smaller dataset, requiring input from stress–strain curves at only three SRs. This integrated, data-efficient approach effectively bridges scales and offers a powerful tool for predicting the hyperelastic behavior of rubber composites at continuous SRs.