Plastic Properties and Collapse Investigation of Fine-grained Soil Rehabilitated with Styrene Butadiene Rubber: A Case Study in Kerman, Iran

Collapsible soils pose significant challenges due to their open structure, which causes settlement when exposed to moisture. Failure to identify these soil types can lead to structural damage when they become saturated or experience changes in moisture content. The presence of such soils in various regions, including Iran, necessitates greater attention and investigation into their behavior and properties. This study examines the impact of butadiene rubber on the stabilization of these soils. Fine-grained soil samples were collected from two different sites in Kerman province (Kerman City). The samples were injected with 2%, 3%, 4%, 5%, 6%, and 7% butadiene rubber for stabilization periods of 4, 7, 14, and 28 days, resulting in a total of 72 tests. The stabilized soils were evaluated using a double consolidation test (ASTM D5333) on intact soil samples. The penetration of butadiene rubber and the resulting rubber columns reduced the degree of collapse. In all cases, the collapse was reduced by more than 88%. The highest reduction was observed with a 7% additive after 28 days of stabilization. Given the increasing use of intelligent systems in predicting the behavior of stabilized collapsible soils, a model was developed to predict the degree of collapse for samples stabilized with butadiene rubber using an adaptive network fuzzy inference system (ANFIS). The accuracy of the model was evaluated, and it successfully predicted the collapse degree. Addition of styrene butadiene rubber additive in the tested soils led to a decrease in the plasticity index of clays with high liquid limits and an increase in the plasticity index of silts with low liquid limits. These changes varied depending on the mineral type. Subsequently, a model was developed to predict the plastic properties of the soil using a fuzzy inference system. The results demonstrate acceptable consistency between the training and prediction data (R2=0.93).