Artificial Neural Network to Predict the Compressive Strength of Semilightweight Concrete Containing Ultrafine GGBS

Abstract

Abstract Design strength is usually determined after a 28-day curing period as per codal provisions. The prediction of compressive strength before curing reduces waiting time and expedites regular construction activity. The aim of this study is to develop a neural network model to predict the 28-day compressive strength of semilightweight concrete (sLWC) containing ultrafine ground granulated blast-furnace slag (UFGGBS). In this investigation, a novel lightweight coarse aggregate that is made up of wood ash was used to prepare sLWC. Six input parameters, such as cement, UFGGBS as cement replacement, lightweight wood ash pellets as coarse aggregate, fine aggregate, water content, and superplasticizer, were used to train the model. The 28-day compressive strength was taken as an output parameter. A total of 384 data was collected from 24 sLWC mixes, each containing 16 specimens, and trained in an artificial neural network (ANN) using a feedforward-backpropagation model. Trained data were validated with a set of tested data. The correlation coefficient R2 values for trained and tested data were 0.932 and 0.917, respectively, with least errors. The study concluded that ANN was a reliable and fast tool for predicting the compressive strength of sLWC. It also efficiently reduced cost and time.