Machine learning modeling and prediction on the seawater aging and creep behavior of GFRP based on XGBoost algorithm

Objectives To address the limitations of traditional theoretical methods, the XGBoost algorithm is employed to construct a machine learning model. The model aims to achieve efficient prediction of the long-term performance of Glass Fiber Reinforced Polymer (GFRP) in seawater environments and to elucidate the influence mechanisms of environmental/load factors on aging behavior and the flexural creep strain of GFRP. MethodsGFRP creep and aging data were collected and compiled, encompassing 244 data points for flexural strength after seawater aging (without external load) and 333 data points for flexural creep strain under water immersion conditions (with sustained load). The preprocessed datasets were randomly divided into training and test sets at a 4:1 ratio. Model hyperparameters were optimized and determined using grid search combined with k-fold cross-validation (k=10). Feature importance analysis was subsequently conducted using the SHapley Additive exPlanations (SHAP) method. Results The machine learning model based on the XGBoost algorithm demonstrated high accuracy on both the seawater aging and flexural creep test sets, with coefficient of determination (R²) of 0.9964 and 0.9941, respectively. It also provided reasonably accurate predictions for new, previously untested conditions used for validation. ConclusionsThe effect of temperature on the seawater aging behavior of GFRP is more pronounced than that of seawater concentration. Furthermore, dry-wet cycling accelerates the degradation of its flexural strength. Regarding the long-term flexural creep performance under sustained loading, however, the influence of temperature is inferior to that of the stress level. These quantitative relationships provide crucial references for the long-term performance assessment of GFRP in seawater environments.