A robust carbonation depth model in recycled aggregate concrete (RAC) using neural network

Abstract

Carbonation depth involves complex physical process and interactions of multiple variables and is thus extremely complicated to predict in concrete structures. It is imperative to quantify this depth due to its vital role in the corrosion of rebars in recycled aggregate concrete (RAC). This paper developed a novel carbonation depth prediction model from a large database of 445 experimental results using artificial neural network (ANN). The relative importance and effect of the independent parameters in the carbonation depth are identified using Garson index and parametric analysis, respectively. Among all the architectures considered, the N 8-10-1 having 10 nodes in the hidden layer provided the best prediction in good agreement with experimental results. The model demonstrated superior performance relative to existing carbonation depth equations in the literature. Despite the presence of fuzziness in the data, the effect of each variable in the development of carbonation is explored in great detail. The model proposed here can provide a robust prediction of carbonation depth that can be used as a basis for assessing the structural health of recycled aggregate concrete structures.