Prediction of the Compressive Strength of Fly Ash Geopolymer Concrete by an Optimised Neural Network Model

Abstract

This article presents a regression tool for predicting the compressive strength of fly ash (FA) geopolymer concrete based on a process of optimising the Matlab code of a feedforward layered neural network (FLNN). From the literature, 189 samples of different FA geopolymer concrete mix-designs were collected and analysed according to ten input variables (all relevant mix-design parameters) and the output variable (cylindrical compressive strength). The developed optimal FLNN model proved to be a powerful tool for predicting the compressive strength of FA geopolymer concrete with a small range of mean squared error (MSE = 10.4 and 15.0), a high correlation coefficient with the actual values (R = 96.0 and 97.5) and a relatively small root mean squared error (RMSE = 3.22 and 3.87 MPa) for the training and testing data, respectively. Based on the optimised model, a powerful design chart for determining the mix-design parameters of FA geopolymer concretes was generated. It is applicable for both one- and two-part geopolymer concretes, as it takes a wide range of mix-design parameters into account. The design chart (with its relatively small error) will ensure cost- and time-efficient geopolymer production in future applications.

Keywords

• fly ash
• geopolymer concrete
• alkali-activated binder
• compressive strength
• feedforward layered neural network
• Matlab code