Experimental and Machine Learning-Based Investigation of Cyclic Thermal Resilience of Geopolymer Concrete with Slag and Glass Powders

Understanding the cyclic loading behaviour of geopolymer concrete (GPC) after exposure to elevated temperatures is vital in judging the damage to structures and fire-resistant applications. However, effectively forecasting the compressive strength of GPC under cyclic loading circumstances following exposure to elevated temperatures is complex. This paper presents the performance of GPC blended with glass powder (GP) or steel slag (SS) under cyclic thermal loading, simulating real-world fire conditions. Furthermore, micro-structural analysis and machine learning modelling were adopted to understand the morphology of GPC and optimize the desirable mix of GPC, respectively. The experimental results reveal substantial strength reductions, up to 65.83% for GP and 67.91% for SS, after ten thermal cycles at 800 °C. This decline is caused by silicate phase expansion at higher temperatures, exacerbating strength loss and highlighting the geopolymers' vulnerability under fire-like conditions. In addition, findings from nine machine-learning models, aside from Lasso regression, performed admirably regarding compressive strength prediction. The results indicate that machine learning models with a high coefficient of determination (R²) value of 0.90 will further enhance geopolymer block development. Finally, the study demonstrates that machine learning techniques may successfully enhance the monitoring of GPC following exposure to cyclic thermal loading and offers crucial insights into improving the fire resistance of GPC behavior under extreme thermal loads.