Metakaolin as a soil stabilizing admixture: A comprehensive analysis of California bearing ratio and consolidation behavior using experimental and machine learning approaches
摘要
Soil stabilization using pozzolanic admixtures, particularly metakaolin, has gained significant attention in geotechnical engineering for enhancing soil properties. This study uniquely investigates the effects of metakaolin, a highly reactive pozzolanic material, on the engineering properties of soil, specifically focusing on bearing capacity and consolidation behavior while incorporating advanced machine learning techniques. The study employs a systematic experimental approach that varying metakaolin contents (0%, 2.5%, 7.5%, 12.5%, 15%, and 25%) by dry weight of soil for traditional tests such as Atterberg limits, specific gravity, compaction characteristics, California bearing ratio (CBR), and consolidation tests, in combination with sophisticated statistical analyses, including Pearson correlation analysis (PCA), gradient boosting machines (GBM), and partial dependence plots (PDPs). Results showed significant improvements in soil properties with metakaolin addition. Unsoaked CBR increased from 41.56% to a maximum of 75.85%, while soaked CBR rose from 15.36 to 39.89%. Maximum dry density increased by 6.98–0.98%, and optimum moisture content decreased by 34.91–11.05%. Consolidation properties improved, with the coefficient of consolidation increasing from 0.1 cm²/min to 0.3075 cm²/min, and the time consolidation decreased to a minimum of 30 min. The GBM sensitivity model demonstrated high accuracy for unsoaked CBR (MSE = 418.9944, R² = 0.819), soaked CBR (MSE = 228.1761, R² = 0.860), and consolidation parameters (MSE = 2.965 × 10−9, R² = 0.9993). Integrating advanced machine learning models to predict complex soil-admixture interactions offers a novel, data-driven approach to soil stabilization. The findings contribute to optimizing soil stabilization strategies, enabling more sustainable and efficient construction practices.