Optimization of marine clay stability treated by environmentally friendly materials using machine learning techniques

Marine clay is instable soil under loading due to its limited shear strength, leading to significant settlements. Such weakness poses a risk to structures, often causing collapses and disasters, especially on long–term of loading, impacting economy, society, and ecology. Consequently, geotechnical strategies for soil stabilization become crucial. However, conventional methods employing calcium–based binders exacerbate environmental degradation and contribute to climate change concerns such as cement which accounts for 30% of all CO2 emissions worldwide. Besides, controlling the ratio of used materials in reinforcing and stabilizing marine clay from the strengthening perspective is challenging. Therefore, this study focuses on reinforcing marine clay to resist short–term and long–term loading using ecofriendly materials such as natural or green pozzolanic materials, using machine learning algorithms. The used ecofriendly materials were palm oil fly ash (POFA) and calcined shale (CS) as supplementary materials for lime (L). The treated marine clay specimens were investigated using ratios of 1%, 3%, and 5% for lime, 1%, 5%, 10%, 15%, and 20% for POFA, and 5%, 10%, 15%, and 20% for CS. Various L–POFA–CS treated specimens were tested under various curing durations: 7, 14, 21, 28, 56, and 90 days. Promising enhancements in soil properties were observed, particularly in sample SL5P5C10, comprising 5% lime, 5% POFA, and 10% CS, where cu increased by 4 times, Øu increased by 50%, c' increased by 6 times, and Ø’ increased by 90% compared to the control specimen. The Atterberg enhanced significantly after 7 curing days from 75.20 to 34.50 for liquid limit WLL, from 48.95 to 32.01 for plastic limit WPL, and treated specimens resisted compressibility under loading, with compression index Cv reduced significantly from 147.5 to 3.15 m2/yr and permeability k from 3.25E–08 to 6.30E–09 cm/sec after 7 curing days for SL5P5C10 treated specimens. 90 finite element model (FEM) simulations of consolidated undrained (CU) triaxial tests were executed using PLAXIS 2D to compare and indicate the relation between the experimental and simulation results. The FEMs produced stress–strain relationships that were consistent with the experimental data, although the FEMs trended towards normal consolidated behavior, while the experimental results were more over consolidated. To enhance the accuracy of the results for final work without experimental work, machine learning techniques were utilized. Machine learning algorithms, including LR, PR, ANN, GBoost, xGBoost, and RF, were used in optimizing eight manual selected feature models of a total 48 models and hybrid algorithms incorporating genetic feature selection, forward feature selection, and backward feature elimination yielded a total of 52 models. Machine learning helped in generation 102 models and optimizing the ratios of used materials in limited time compared to the traditional analyzing methods using cross–validation process. The machine learning models could be used as a guideline and estimation and prediction for shear strength after treatment using mixture of lime, CS, and POFA for future research. Eventually, a further investigation was then made on the natural and optimum mixture (SL5P5C10) through FEM to indicate its deformation behavior using PLAXIS 2D program for both short–term and long and long term of loading of Serdang Landslide. The simulation shows a substantial increase in soil resistance to horizontal stresses post–treatment, with stress levels decreasing from 20 kN/m2 in natural soil to 2 kN/m2 in SL5P5C10 treated soil fill. This signifies a tenfold enhancement in soil's ability to withstand horizontal stresses compared to untreated soil under similar loading conditions. Utilizing SL5P5C10 treated soil as fill behind retaining walls holds potential for engineering, geotechnical, and economic advantages, facilitating taller building constructions and accommodating heavier traffic loads in adjacent parking areas.

Text FK 2023 20 - Declaration Form.pdf Restricted to Repository staff only Available under License Creative Commons Attribution Non-commercial No Derivatives. Download (1MB) Text FK 2023 20 - Full Text.pdf Available under License Creative Commons Attribution Non-commercial No Derivatives. Download (16MB) Text FK 2023 20.pdf Restricted to Repository staff only Available under License Creative Commons Attribution Non-commercial No Derivatives. Download (16MB) Official URL or Download Paper: http://ethesis.upm.edu.my/id/eprint/18937

Actions (login required)

View Item