Experimental and predictive analysis of pile raft foundation in clayey soils using machine learning and MATLAB tools

Area of Science:

• Geotechnical Engineering
• Soil Mechanics
• Foundation Design

Background:

• Clay soils in southern India exhibit high clay content (54-70%) and montmorillonite mineralogy.
• These soil properties result in high Free Swell Index (FSI: 75-200%) and low dry densities (13.5-16.9 g/cc).
• Piled raft foundations are a viable solution for structures on problematic clayey soils.

Purpose of the Study:

• To evaluate the performance of piled raft foundations in diverse Indian clay soils.
• To investigate the impact of varying pile configurations on load sharing and settlement.
• To develop and validate machine learning models for predicting piled raft performance.

Main Methods:

• Comprehensive physical, chemical, mineralogical, and geotechnical characterization of 12 clay soil samples.
• Experimental loading tests on piled raft configurations (R, R4, R6, R9).
• Development and validation of machine learning models using soil parameters (clay %, FSI, MDD) to predict Load Sharing Ratio (LSR) and Settlement Reduction Ratio (SRR).

Main Results:

• Increasing piles from 4 to 9 (R4 to R9) resulted in a 64% increase in LSR and a 75% reduction in settlement.
• Machine learning models accurately predicted LSR and SRR for higher pile configurations (R12-R21) with >90% accuracy.
• MATLAB verification confirmed >95% accuracy of the ML models for predicting piled raft performance.

Conclusions:

• Piled raft foundations offer a cost-effective design solution for various clayey soils in southern India.
• The number of piles significantly influences the load-sharing capacity and settlement reduction of raft foundations.
• Machine learning provides a reliable and accurate tool for optimizing piled raft foundation design.