An Enhanced Neural Network Scheme to Model Pile Load-Deformation Under Uplift Loading

Abstract

This study designed to explore load displacement of steel open-ended model piles driven in cohesionless soil and subjected to axial uplift loads. The feasibility of a novel computational intelligence (CI) scheme to correlate the full behavior of the pile load-deformation has also been examined. Self-tuning Levenberg-Marquardt (LM) training algorithms, enhanced by the null-hypothesis tests (T-tests and F-tests), have been implemented in this process. The pile aspect ratios were varied from 12, 17, and 25. The piles were tested using an innovative pile-testing chamber in three relative densities of noncohesive soil, ranging from dense, medium and loose sand. The prediction metrics indictors demonstrate an excellent performance of the adopted modelling approach in capturing the full behavior of the pile load-displacement, thus yielding a Root Mean Square Error, Determination Coefficient, and Mean Absolute Error of 0.14, 0.96, and 6.8x10^3, respectively.