Crack Spacing Prediction of Fibre-Reinforced Concrete Beams with Lap-Spliced Bars by Machine Learning Models

Abstract

Reduction of spaces between flexural cracks is one of the crucial factors in protecting reinforcements from corrosion. Different parameters, such as concrete cover, concrete compressive strength, specimens’ dimension and spacing of longitudinal and transverse reinforcements, influence the crack spacing. Insufficient lap-spliced length reduces the strength and increases the crack spacing. In this study, the effects of lap-sliced length, a cyclic number, steel fibre percentage, longitudinal rebar diameter, concrete compressive strength, cross-section height and width on flexural crack spacing are investigated. Multilayer perceptron (MLP) neural network, adaptive neuro-fuzzy inference system, support vector regression and least mean squares regression are utilized to predict the flexural crack spacing. The obtained results illustrate that the MLP model yields better accuracy compared to other ones in the prediction of cracking spacing. The proper features for predicting crack spacing are prioritized based on mutual information between features. Furthermore, it is found that the available codes are unable to determine the crack spacing within reinforced concrete beams with lap-spliced bars.