Maximum dry density estimation of stabilized soil via machine learning techniques in individual and hybrid approaches

• Published in: Journal of ambient intelligence and humanized computing Vol. 15; no. 11; pp. 3831 - 3846
• Main Authors: Zhao, Lianping, Guan, Guan Dashu
• Format: Journal Article
• Language: English
• Published: Berlin/Heidelberg Springer Berlin Heidelberg 01.11.2024; Springer Nature B.V
• Subjects: Algorithms; Artificial Intelligence; Cement; Civil engineering; Computational Intelligence; Density; Dry density; Engineering; Geotechnical engineering; Grain size distribution; Machine learning; Moisture content; Optimization; Original Research; Particle size distribution; Performance prediction; Regression analysis; Robotics and Automation; Soil compaction; Soil mixtures; Soil moisture; Soil shrinkage; Soil strength; Support vector machines; User Interfaces and Human Computer Interaction; Variables; Support vector regression; Maximum dry density; Artificial rabbits optimization; Manta ray foraging optimization; Improved Manta-Ray foraging optimizer
• ISSN: 1868-5137; 1868-5145
• DOI: 10.1007/s12652-024-04860-5

In geotechnical engineering, the maximum dry density (MDD) stands as an important parameter, denoting the utmost mass of soil achievable per unit volume when compacted to its maximum dry state. Its significance extends to the design of various earthworks like embankments, foundations, and pavements, influencing the soil’s strength, stiffness, and stability. The MDD is contingent on diverse elements like soil type, grain size distribution, moisture content and compaction effort. Generally, heightened compaction effort correlates with an increased MDD, while elevated moisture content corresponds to a reduced MDD. Accurate prediction of the MDD under specific conditions is imperative to uphold the quality and safety standards of earthworks. This research aims to introduce Support Vector Regression (SVR) as a modeling technique for predicting the MDD of soil-stabilizer mixtures. To establish an accurate and comprehensive model that can correlate the stabilized soil’s MDD with attributes of natural soil, consisting linear shrinkage, particle size distribution, plasticity, as well as the type and number of stabilizing additives, three optimization algorithms, namely Artificial Rabbits Optimization (ARO), Manta Ray Foraging Optimization (MRFO), and Improved Manta-Ray Foraging Optimizer (IMRFO), were employed in addition to SVR. Considering the results of evaluative metrics, the SVAR model (combination of SVR and ARO) experienced the highest predictive performance, registering an impressive value of R 2 in the training phase with 0.9948, as well as the lowest RMSE value of 19.1376.