Artificial Neural Network Approach to Predict Carbonation Depth in Metakaolin, Brick Powder and Calcined Sediments-Modified Mortars

Abstract

This research uses Artificial Neural Network (ANN) as a soft computing technique to predict the carbonation depth and service life of cementitious materials with low clinker content. For this purpose, different mortars were prepared with 0, 10, 15, 20, 25 and 30% replacement levels of cement by metakaolin (MK), brick powder (BP) and calcined sediments (CS). The experimental results of the carbonation depth were obtained under natural and accelerated carbonation conditions for exposure periods of 12 months and 28 days respectively. ANN was utilized taking into account the main influential factors on mortars carbonation, including mix proportions and environmental conditions. For the ANN model, seven datasets were considered as inputs, covering mineral admixture content, cement content, curing time, CO2 concentration, relative humidity, temperature and CO2 exposure time, in addition to one output parameter which is the carbonation depth. The results show that the resistance to carbonation of the mortars decreases with the increase of cement substitution by MK, BP or CS. The network model gives good performance values in the validation and testing set with a lower mean square error (MSE) and a higher determination coefficient (R). The predicted carbonation depths are in good agreement with the experimental measurements of carbonation depths, confirming the efficiency of the developed ANN model to be applied to correctly estimate the carbonation depth of cementitious materials with low clinker content.

Keywords

• mortar
• carbonation depth prediction
• artificial neural network
• metakaolin
• brick powder
• calcined sediments