An innovative intelligent design method of alkali-activated foamed geopolymer: Mixture optimization and performance prediction

Abstract

The complex design parameters and lack of standard mix design methods generate a difficulty in the high efficiently preparation of alkali-activated foamed geopolymer (AAFG). Therefore, this paper proposes an innovative intelligent design method of AAFG based on machine learning (ML) and Metaheuristic Optimization Algorithms (MOA). The effects of twelve factors on drying density (DD) and compressive strength (CS) of AAFG are revealed. The DD and CS predictive models of AAFG with high accuracy (R2 = 0.96 and 0.84) and strong generalization are trained through Random Forest (RF). The enlargements of proportions of granulated blast furnace slag (GBFS) and metakaolin (MK) in precursors can improve the CS but are not conducive to reduction of DD. Higher proportion of fly ash (FA) in precursors can efficiently reduce DD but can weaken the CS. Higher or lower Na2O content, silicate modulus (Ms) and water to binder (W/B) can reduce DD and CS with the optimal foaming conditions of 20 %, 1.0 and 0.75. More additions of foaming agents and foam stabilizers can reduce DD, while the CS has a remarkable degradation. Higher or lower foaming temperature (FT) is both not beneficial for the reduction of DD and enhancement of CS with the optimal FT of 60 °C. The intelligent design system of AAFG developed by RF and Slime Mould Algorithm (SMA) has easy operability, high efficiency, and good stability. The optimal preparation parameters can be efficiently acquired and the corresponding performances can be quickly and accurately predicted.