Comparison between Multiple Regression Analysis, Polynomial Regression Analysis, and an Artificial Neural Network for Tensile Strength Prediction of BFRP and GFRP

In this study, multiple regression analysis (MRA) and polynomial regression analysis (PRA), which are traditional statistical methods, were applied to analyze factors affecting the tensile strength of basalt and glass fiber-reinforced polymers (FRPs) exposed to alkaline environments and predict the tensile strength degradation. The MRA and PRA are methods of estimating functions using statistical techniques, but there are disadvantages in the scalability of the model because they are limited by experimental results. Therefore, recently, highly scalable artificial neural networks (ANN) have been studied to analyze complex relationships. In this study, the prediction performance was evaluated in comparison to the MRA, PRA, and ANN. Tensile strength tests were conducted after exposure for 50, 100, and 200 days in alkaline environments at 20, 40, and 60 degrees C. The tensile strength was set as the dependent variable, with the temperature (TP), the exposure day (ED), and the diameter (D) as independent variables. The MRA and PRA results showed that the TP was the most influential factor in the tensile strength degradation of FRPs, followed by the exposure time (ED) and diameter (D). The ANN method provided the best correlation between predictions and experimental values, with the lowest error and error rate. The PRA method applied to the response surface method outperformed the MRA method, which is most commonly used. These results demonstrate that ANN can be the most efficient model for predicting the durability of FRPs.