Sustainable stabilization of clay soils using fly ash and blast furnace slag-based geopolymers: Comprehensive experimental evaluation and artificial neural network modelling

In this study, clay soils were stabilized using fly ash (FA), ground blast furnace slag (GGBS) and alkaline activators in order to increase their mechanical and durability properties. The effects of different additive ratios on the soil were evaluated by unconfined compressive strength (UCS), and California Bearing Ratio (CBR) tests. The results showed that the combination of alkaline activators and FA significantly increased the strength of clay soils and reached a UCS value of 600 kPa at the end of 28 days. On the other hand, mixtures containing GGBS showed the highest strength and reached a value of 2470 kPa. Field Emission Scanning Electron Microscope (FE-SEM) analyses have confirmed that these mixtures have been geopolymerized successfully and sodium-alumino-silicate-hydrate (N-A-S-H) gel has been formed. Furthermore, the analysis carried out using the Artificial Neural Network (ANN) model showed a high level of predictive accuracy, with an R2 value of 0.86 and a Root Mean Square Error (RMSE) of 198.73, confirming the model's reliability for practical applications in soil stabilization. Overall, this study reveals that environmentally friendly binders offer a viable alternative in clay floor stabilization and have a wide potential in the field of civil engineering. The innovative aspect of the study is the comprehensive experimental investigation of the mechanical performance of clay soils stabilized with sustainable and environmentally friendly geopolymer binders using different proportions of fly ash and blast furnace slag and the use of these experimental data to make highly accurate predictions using artificial neural network (ANN)-based advanced modeling techniques as a practical and reliable forecasting tool in soil stabilization applications. Thus, both the reduction of environmental impacts and the integration of artificial intelligencesupported optimization in engineering applications have been achieved.