Performance Prediction of Glass Fibre-Reinforced Concrete Using Machine-Learning Methods

Abstract

Fibre materials are now widely used in reinforcing the mechanical properties of cementitious materials. Glass fibre has high technical value due to its good tensile strength and low price. To achieve accurate prediction of the compressive and splitting tensile strengths of glass fibre reinforced concrete (GFRC), this research built machine-learning models based on six methods, namely Gaussian process regression (GPR), support vector machine regression (SVR), boosted regression tree (BRT), random forest regression (RFR), multilayer perceptron (MLP) and convolutional neural network (CNN). A 28-day compressive strength data set containing 191 samples and a 28-day splitting tensile strength data set containing 141 samples were used to train the machine-learning models. The MLP model was evaluated as the best-performed model to predict the strength of GFRC. Test specimens were designed using the predicted ingredient ratios of the MLP model. The predicted results of the machine-learning models were compared with those of the test results. It was found that all the machine-learning models predicted the results with good accuracy within the range of the dataset, among which the MLP model performed the best. For the groups that fell outside the upper and lower limits of the dataset with the extreme conditions of no coarse aggregate or no glass fibre, the MLP model still showed good prediction results although the overall performance of the six trained machine-learning models showed large variability. The MLP model showed bad prediction for the group with no coarse aggregate and no glass fibre.

Keywords: Glass fibre-reinforced concrete, Machine-learning models, Test