A machine learning regression approach for predicting the bearing capacity of a strip footing on rock mass under inclined and eccentric load

Abstract

In this study, the Multivariate Adaptive Regression Splines (MARS) model is employed to create a data-driven prediction for the bearing capacity of a strip footing on rock mass subjected to an inclined and eccentric load. The strengths of rock masses are based on the Hoek-Brown failure criterion. To develop the set of training data in MARS, the lower and upper bound finite element limit analysis (FELA) is carried out to obtain the numerical results of the bearing capacity of a strip footing with the width of B. There are six considered dimensionless variables, including the geological strength index (GSI), the rock constant/yield parameter (mi), the dimensionless strength (γB/σci), the adhesion factor (α), load inclined angle from the vertical axis (β), and the eccentricity of load (e/B). A total of 5,120 FELA solutions of the bearing capacity factor (P/σciB) are obtained and used as a training data set. The influences of all dimensionless variables on the bearing capacity factors and the failure mechanisms are investigated and discussed in detail. The sensitivity analysis of these dimensionless variables is also examined.