Prediction of compression capacity of under-reamed piles in sand and clay

This study aimed to develop a predictive model for estimating the compressive load-carrying capacity of single and double under-reamed piles in clay and sand based on numerical data generated by conducting finite-element analysis. The pile dimensions, soil characteristics, and properties were selected using relevant literature. The soil was assumed to follow Mohr-Coulomb's yield criteria in conjunction with an associated flow rule for clay and a non-associated flow rule for sand. The study outcomes revealed that incorporating an under-ream section was particularly advantageous for piles in stiff clay and loose to moderately dense sand. The increase in load capacity for under-reamed piles was substantial, reaching up to 570% for piles in clay and 297% for piles in sand. The results obtained from the numerical simulations were used to develop a regression equation for the bearing capacity factor Nc or piles in clay. In addition, predictive models were generated using random forest regression and artificial neural networks for piles embedded in sand. The derived Nc equation may prove helpful and effective in predicting the compressive load-bearing capacity of piles in clay. Meanwhile, the artificial neural network and random forest regression models successfully established a strong relationship between the numerical findings and available data, achieving an R2 value greater than 0.9. The sensitivity analysis results indicated that the friction angle of the sand had a significant impact on the compressive capacity of the piles, while the other variables contributed almost equally.