Predicting the creep behavior of thermoset polymers using machine learning techniques

Creep, the time-dependent deformation of materials under constant stress, is a critical factor in assessing material performance under long-term mechanical loading. Accurate prediction of creep behavior is essential across fields such as structural engineering and materials science. This study explores the use of machine learning (ML) techniques—specifically Multilayer Perceptron (MLP) networks and regression methods—for predicting creep deformation, considering key variables like stress, temperature, and time. To enhance prediction accuracy, a hybrid model is proposed, combining nonlinear regression to capture the overall exponential trend in strain with an MLP network to model residual deviations. Experimental data on the creep behavior of epoxy resin (Araldite LY 564) at various stress levels and temperatures, provided by Bakbak et al. (Polym Bull 79:1–17, 2022) and Birkan et al. (J Compos Mater 57(22):3449–3462, 2023), were used, supplemented by interpolated artificial data to improve model training. Results show that both regression and MLP models yield satisfactory predictions, while the hybrid model offers improved accuracy and robustness in capturing creep behavior.