Machine learning models to predict the residual tensile strength of glass fiber reinforced polymer bars in strong alkaline environments: A comparative study

Abstract

The long-term durability of glass fiber reinforced polymers (GFRPs) in strong alkaline environments is of utmost importance in marine infrastructure construction. The residual tensile strength, as one of the important durability indicators, can be characterized by the tensile strength retention (TSR). However, accurate prediction of the TSR is a challenging task. Therefore, the main objective of this study is to develop a generalized, accurate, and optimized TSR prediction model from machine learning (ML) perspective. To this aim, seven machine learning models were developed using the diameter of the bar (db), the volume fraction (Vf) of the E-glass fibers, the pH of the alkaline solution, the conditioning temperature (temp) and the duration of immersion (T) as input variables. A database containing 150 sets of samples, divided into training and testing sets, was created for model building and comparison. Evaluated against three performance evaluation metrics (including RMSE, R2, and VAF) and the Taylor diagram, the final generalization performance of the models was found to be the extreme gradient boosting (XGBoost), long short-term memory (LSTM), support vector regression (SVR), random forest (RF), extreme learning machine (ELM), backpropagation neural network (BPNN), and generalized additive model (GAM) from highest to lowest. In addition, the relative sensitivity of the five input variables was assessed by the one variable at a time (OVAAT) method, and pH and temp were identified as the top two most significant variables in TSR prediction. This study also explored the effect of the training set/test set division ratio on the model, an aspect that has not been investigated in previous studies, and identified 8:2 as the optimal division ratio. The findings of this study could be of real benefit to long-term durability studies of GFRP bars, and also expand the application of machine learning in civil engineering.