Efficient soft computing techniques for the prediction of compressive strength of geopolymer concrete

Abstract

In the recent year, extensive researches have been done on fly ash-based geopolymer concrete for its similar properties like Portland cement as well as its environmental sustainability. However, it is difficult to provide a consistent method for geopolymer mix design because of the complexity and uncertainty of its design parameters, such as the alkaline solution concentration, mole ratio, and liquid to fly ash mass ratio. These mix-design parameters, along with the curing time and temperature ominously affect the most significant properties of the geopolymer concrete, i.e., compressive strength. To overcome these difficulties, the paper aims to provide a simple mix-design tool using artificial intelligence (AI) models. Three well-established and efficient AI techniques namely, genetic programming, relevance vector machine, and Gaussian process regression are used. Based on the performance of the developed models, it is understood that all the models have the capability to deliver higher prediction accuracies in the range of 0.9362 to 0.9905 (based on R2 value). Among the employed models, RVM outperformed the other model with R2=0.9905 and RMSE=0.0218. Theodore, the developed RVM model is very potential to be a new alternative to assist engineers to save time and expenditure on account of the trial-and-error process in finding the correct design mix proportions.