Artificial neural network (ANN) approach for modelling of pile settlement of open-ended steel piles subjected to compression load

Abstract

This study was devoted to examine pile bearing capacity and to provide a reliable model to simulate pile load-settlement behaviour using a new artificial neural network (ANN) method. To achieve the planned aim, experimental pile load test were carried out on model open-ended steel piles, with pile aspect ratios of 12, 17, and 25. An optimised second-order Levenberg-Marquardt (LM) training algorithm has been used in this process. The piles were driven in three sand densities; dense, medium, and loose. A statistical analysis test was conducted to explore the relative importance and the statistical contribution (Beta and Sig) values of the independent variables on the model output. Pile effective length, pile flexural rigidity, applied load, sand-pile friction angle and pile aspect ratio have been identified to be the most effective parameters on model output. To demonstrate the effectiveness of the proposed algorithm, a graphical comparison was performed between the implemented algorithm and the most conventional pile capacity design approaches. The proficiency metric indicators demonstrated an outstanding agreement between the measured and predicted pile-load settlement, thus yielding a correlation coefficient (R) and root mean square error (RMSE) of 0.99, 0.043 respectively, with a relatively insignificant mean square error level (MSE) of 0.0019.