Multivariable models including artificial neural network and M5P-tree to forecast the stress at the failure of alkali-activated concrete at ambient curing condition and various mixture proportions

Alkali-activated concrete (AAC) has emerged as a sustainable construction material due to the environmental issues associated with cement production. This type of concrete is cementless concrete that employs industrial or agro by-product ashes like fly ash (FA) and ground granulated blast furnace slag (GGBFS) in their mixture proportions as the primary binders instead of conventional Portland cement. All concrete composites, including AAC, rely on compressive strength. However, the 28-day compressive strength of concrete is critical in structural design. Therefore, developing an authoritative model for estimating AAC compressive strength saves time, energy, and money while guiding the construction and formwork removal. This study used artificial neural network (ANN), M5P-tree, linear regression, non-linear regression, and multi-logistic regression models to predict blended GGBFS/FA-based AAC's compressive strength at different mixture proportions curing ages. A comprehensive dataset consists of 469 samples collected in several academic research studies and analyzed to develop the models. In the modeling process, for the first time, twelve effective variable parameters on the compressive strength of the AAC, including the alkaline solution-to-binder ratio, FA content, SiO2/Al2O3 of FA, GGBFS content, SiO2/CaO of GGBFS, fine and coarse aggregate content, NaOH and Na2SiO3 content, Na2SiO3/NaOH ratio, molarity and age of concrete specimens were considered as the modeling input parameters. Various statistical assessment tools such as RMSE, MAE, SI, OBJ value, and R-2 were used to evaluate the efficiency of the developed models. The results indicated that the ANN model better predicted GGBFS/FA-based AAC mixtures' compressive strength than the other models. Moreover, the sensitivity analysis demonstrated that the alkaline liquid-to-binder ratio, NaOH content, and age of concrete specimens were those parameters that significantly influenced the compressive strength of the AAC.

Automated summary

This study developed various models to predict the compressive strength of blended GGBFS/FA-based AAC. The results showed that the artificial neural network (ANN) model outperformed the other models in predicting the compressive strength. Sensitivity analysis revealed that the alkaline liquid-to-binder ratio, NaOH content, and age of concrete specimens significantly influenced the compressive strength of AAC.