Performance of different machine learning techniques in predicting the flexural capacity of concrete beams reinforced with FRP rods: Fid-Bau Portal

Performance of different machine learning techniques in predicting the flexural capacity of concrete beams reinforced with FRP rods

Ngamkhanong, Chayut / Alzabeebee, Saif / Keawsawasvong, Suraparb / Thongchom, Chanachai

Abstract

Nowadays, fiber-reinforced polymer (FRP) rods have become an optional material for reinforced concrete structures instead of steel. This material has high corrosion resistance, high chemical resistance, and thermal properties. It can maintain high strength while being lightweight material is easier to install. However, predicting the flexural capacity of concrete beams reinforced with FRP rods cannot be directly calculated due to the lack of robust correlation. Thus, this paper assesses the capabilities of the machine learning techniques to predict this flexural capacity. One hundred and one results of bending flexural capacity of concrete beams reinforced with FRP rods have been collected and used in this paper. In addition, the machine learning (ML) techniques used are the multiple linear regression analysis, the support vector machines, the k-nearest neighbor, the multi-layer perceptron, the decision tree, and the random forest. Overall, all of the aforementioned techniques provided reasonable accuracy as the obtained coefficient of correlation ranged between 0.82 and 0.96, the mean absolute error ranged between 9.85 and 20.42, and the mean absolute mean error ranged between 14.82 and 28.92. The ranking of the ML techniques based on the accuracy are: random forest, k-nearest neighbor, multi-layer perceptron, decision tree, support vector machines, and multiple linear regression. Importantly, two explicit models have been proposed based on the multiple linear regression and the support vector machines. These equations could be used by designers and future studies to predict the flexural capacity of concrete beams reinforced with FPR rods.