Experimental and machine learning-based performance evaluation of ambient-cured geopolymer mortar using microstructural characterisation and ABAQUS numerical analysis

Abstract

The geopolymer mortar has become a viable substitute for the normal cement mortar due to its light carbon footprint, effective use of industrial wastes and its high durability. The effects of different precursor materials and alkaline activator systems on the fresh, mechanical, microstructural, and structural performance of the ambient-cured geopolymer mortar are systematically studied in this work. Alkaline activators based on potassium and sodium were used to evaluate fly ash (FA), ground granulated blast furnace slag (GGBFS), metakaolin (MK), and steel slag (SS) separately and in hybrid combinations. The developed geopolymer mortars were tested for flowability and compressive strength at 7, 14, and 28 days. The mixes were further characterised using scanning electron microscopy (SEM), X-ray diffraction (XRD), and thermogravimetric analysis (TGA-DTG), which were optimised to ascertain the relationship between microstructural development, gel formation, thermal stability, and mechanical performance. To rapidly predict the strength, four machine learning models (Artificial Neural Network (ANN), Gaussian Process Regression (GPR), Random Forest (RF), and Support Vector Machine (SVM)) were developed and compared to the experimental data. The Shapley Additive Explanations (SHAP) analysis was utilised to explain the ANN network and determine the most important input variables. The 100% GGBFS mixture exhibited the highest 28-day compressive strength of 68.51 MPa, whereas the 50% FA–50% GGBFS blend achieved 61.44 MPa and was selected as a balanced hybrid precursor system for further microstructural investigation. The NaOH-Na2SiO3 activator system exhibited the best geopolymerization and mechanical performance in comparison to the other combinations of activators. The preliminary ABAQUS predicted ultimate load analysis predicted an increase in numerical load-carrying capacity from 48 kN for the BBM reference configuration to 74 kN for the BGRM-3 configuration, corresponding to a 54.2% increase relative to the numerical reference model. These numerical results are considered preliminary because the structural models were not experimentally validated. The ANN model showed the greatest prediction among the models compared, and SHAP analysis found that GGBFS content, sodium hydroxide concentration, sodium silicate dose, and additional water were the most influential factors in controlling the compressive strength. Moreover, nonlinear finite element analysis at ABAQUS revealed that geopolymer mortar jacketing, which is reinforced by three layers of hexagonal wire mesh, enhanced the load-bearing capacity, stiffness, crack resistance and structural behaviour of reinforced concrete beams. The combined experimental, microstructural, machine learning, and numerical study validates the high potential of geopolymer mortar as a high-performance, low-carbon material to be used as a sustainable construction and structural rehabilitation tool.