Assessing the Porosity-Binder Ratio and Machine Learning Models for Predicting the Strength and Durability of...

Area of Science:

• Geotechnical Engineering
• Materials Science
• Sustainable Construction

Background:

• Silty soils often require stabilization for engineering applications.
• Recycled ground glass powder (GGP) offers a sustainable alternative binder in soil stabilization.
• Assessing mechanical behavior and durability is crucial for infrastructure development.

Purpose of the Study:

• To evaluate the mechanical properties (unconfined compressive strength, splitting tensile strength) and durability (accumulated mass loss) of silty soil stabilized with Portland cement and GGP.
• To validate the porosity-cement index (η/C_iv) as a predictor for strength and durability.
• To explore the application of machine learning models for predicting soil stabilization performance.

Main Methods:

• Preparation of soil mixtures with varying cement (3-9%), GGP (5-30%), and dry unit weights (13.5-15.5 kN/m³).
• Mechanical testing (486 tests) and durability testing (81 tests) after curing periods of 7, 28, and 90 days.
• Empirical modeling using the porosity-cement index and machine learning analysis with 28 presets.

Main Results:

• A strong power-law relationship (R² > 0.98) was found between the η/C_iv index and both unconfined compressive strength (qu) and splitting tensile strength (qt).
• Strength coefficient (A) increased with curing time and GGP content, with qu increasing by over 250% and qt by nearly 700% after 90 days.
• Durability improved significantly with higher density and binder content, showing exponential reductions in accumulated mass loss (ALM) to below 0.5% for mixtures with 30% GGP.
• Machine learning models, particularly Matern 5/2 Gaussian Process Regression and trilayered neural networks, achieved R² > 0.987 in predicting qu, qt, and ALM.

Conclusions:

• The η/C_iv index is a reliable predictor of strength and durability for soil-cement-GGP geomaterials.
• Incorporating GGP enhances the mechanical performance and durability of stabilized silty soils.
• Machine learning models show high potential for predicting the behavior of these stabilized soils, facilitating optimized mix design.