Estimating the creep rupture time of GFRP bars using machine learning

Abstract

Fiber-reinforced polymer (FRP) bars are increasingly utilized in civil structures due to their advantages in terms of corrosion resistance and a high strength-to-weight ratio. Current research on long-term durability, particularly under sustained loading (creep-rupture), has not yet fully explored the use of methods like machine learning to accurately predict the creep rupture time of FRP bars. This study seeks to address this gap by applying machine learning techniques to estimate the creep rupture time of glass fiber-reinforced polymer (GFRP) bars. The motivation for this research comes from the shortcomings of traditional models, which are often inadequate for capturing the complex nonlinear behavior of materials subjected to long-term stress. This research aims to evaluate the effectiveness of different machine learning models, including neural networks, support vector machines, and ensemble methods, in predicting the creep behavior of GFRP bars. Within the study context, a large dataset consisting of 435 experimental tests is collected from the literature. In the testing phase, the optimized neural network achieved an RMSE of 926.29 h and an R² of 0.99 on a heterogeneous dataset that also included bars tested under environmental conditioning reported in the source studies. Gaussian process regression and support vector machines also performed well, albeit with higher errors. Sensitivity analysis revealed that the level of sustained stress and bar diameter were the most critical factors for environmentally conditioned bars. Importantly, the predictors reflect standard design and material descriptors (diameter, fiber content, modulus, UTS, sustained stress) and, when reported, environmental conditioning, which together capture the primary sources of variability relevant to civil engineering practice. Overall, the findings suggest that machine learning, particularly through optimized neural networks, offers a powerful tool for predicting complex material behavior and improving the reliability of GFRP-reinforced structures. This study contributes to the field by highlighting the potential of machine learning to enhance the precision of long-term performance predictions for engineering materials, facilitating improved design and material selection in critical infrastructure.