ABSTRACT Stress relaxation, the gradual reduction of stress under constant strain, is a key parameter in assessing long‐term mechanical performance of polymers and polymer matrix composites. Accurate prediction of relaxation behavior is vital in structural engineering and materials science. This study applies machine learning (ML) methods, specifically multilayer perceptron (MLP) networks and nonlinear regression, to model stress relaxation behavior of epoxy resin (Araldite LY 564) as a function of strain, temperature, and time. Also, a hybrid model is introduced, combining nonlinear regression to describe the overall exponential decay with an MLP network to capture residual deviations. Experimental relaxation data were supplemented with interpolated artificial data to enhance model training. Results show that both regression and MLP approaches predict relaxation behavior effectively, while the hybrid model demonstrates superior accuracy and robustness, making it a promising tool for long‐term material performance prediction. Augmenting the training set with artificial data improves long‐term prediction performance and produces results more consistent with the experimental data.
Cakir et al. (Sat,) studied this question.