Application of artificial neural networks for predicting the bearing capacity of the tip of a pile embedded in a rock mass

Abstract

Addressing the problem of the bearing capacity of the tip of a pile in rock needs to consider a full-nonlinear behavior of the rock, as the Hoek and Brown failure criterion, and including the various geometric parameters defining the system geometry. It usually needs the use of complex numerical models. They require a high level of training and expertise and are commonly out of the reach of engineers on the day-to-day calculation procedures. Technical codes usually recommend simpler formulae based on one or two rock parameters, their accuracy relying on particular and local empirical testing. This research develops an artificial neural network (ANN) that solves the previous difficulties. The ANN is based on a group of 1440 calculations using the novel numerical procedure Discontinuity Layout Optimization (DLO). This numerical method can efficiently reproduce the non-linear behavior of the rock (including rock type, uniaxial compressive strength, and geological strength index) for every configuration of the model (foundation width, pile embedment, and height of the overlying soil). Once the ANN is trained and optimized, it can easily predict any result within its range of applicability. Furthermore, it can be reduced to a simple set of recursive equations to be implemented in a spreadsheet. The proposed ANN has only one hidden layer besides the input and output layers, with (6)-(8)-(1) neurons, respectively. It gives highly accurate results with minimum cost and can be a useful tool for engineers and scientists in the foundation field.