Mechanical characterization of marl soil treated by cement and lignosulfonate under freeze-thaw cycles: experimental studies and machine-learning modeling

Calcium carbonate-enriched marl is supposed to lose the bearing capacity while subjected to an increase/decrease in the moisture content and under the effect of multiple freeze-thaw (F-T) cycles. Investigating different weight percentages and curing periods, combination of ordinary Portland cement (OPC) and lignosulfonate is introduced as an efficient method to improve the mechanical strength of the marl soil under the effect of F-T cycles. In this regard, compaction, unconfined compressive strength, and direct shear tests were conducted. It was observed that although the addition of OPC (as the sole binder) can enhance the shear strength, the performance of samples under the action of F-T was not significantly improved. However, samples containing lignosulfonate showed a rectified behavior. Microstructural investigations exhibited the development of new intensity peaks for the calcium-aluminate-silicate-hydrate (C-A-S-H) products and elaborated a denser structure with a lower porosity, keeping the soil particles closer. Next, different intelligent approaches of machine learning (ML) were employed to provide cost-effective and accurate speedy tools. Among eight machine learning algorithms and advanced ensemble models, comparative study revealed the efficiency of gradient boosting model with coefficients of determinations of up to 98.5% for the prediction of UCS. Feature importance analysis suggested the duration of treatment and the cement content as the main contributing factors to the UCS. Highly accurate and efficient EPR-based models with coefficients of determination of higher than 99% were also proposed for the prediction of shear strength and shear stress parameters.