Analysis of Unconfined Compressive Strength of Rammed Earth Mixes Based on Artificial Neural Network and Statistical...

Area of Science:

• Geotechnical Engineering
• Sustainable Construction Materials
• Computational Modeling

Background:

• Earth materials offer sustainable, healthy, and safe construction options.
• Optimizing soil mixes is challenging due to inherent soil property variations.
• Predicting soil strength is crucial for effective construction applications.

Purpose of the Study:

• To compare the efficacy of multilinear regression (MLR) and artificial neural networks (ANNs) in predicting soil compressive strength.
• To analyze key soil properties influencing construction material performance.
• To provide recommendations for optimal earth material mixes.

Main Methods:

• Analysis of 488 historical soil datasets (stabilized and unstabilized).
• Estimation of missing data using established correlations.
• Development and validation of MLR and ANN models, including Levenberg-Marquardt algorithms.
• Experimental validation and sensitivity analysis of ANN models.

Main Results:

• ANN models achieved a significantly higher predictive accuracy (R² up to 0.9883) compared to MLR (R² up to 0.766 for stabilized soils).
• Sensitivity analysis identified key soil properties affecting unconfined compressive strength (UCS).
• Experimental validation confirmed the predictive capabilities of the developed models.

Conclusions:

• ANNs provide a superior method for predicting the compressive strength of earth construction materials.
• Accurate prediction models facilitate the development of optimized and sustainable soil mixes.
• The study offers valuable insights and recommendations for utilizing earth materials in construction.