Prediction of failure envelopes of foundations using machine learning algorithm and finite element limit analysis

This thesis is concerned with the V-H-M failure envelopes for strip, ring, and conical foundations under combined loadings on anisotropic clay. The study employs the finite element limit analysis (FELA) technique and utilizes the well-established Anisotropic Undrained Shear (AUS) failure criteria. The focus of the study is on evaluating the influences of the geometric footing (ri/ro, β) and the anisotropic factor (re) on the bearing capacity of foundations subjected to external forces of vertical force (V), horizontal force (H), and moment (M). The combinations of V-H, V-M, and H-M load spaces are analyzed using dimensionless output parameters, and the various characteristics of failure mechanisms of the foundations are examined. Throughout this thesis, examining predicted foundation failure mechanisms is given importance. This research aims to take a holistic approach to foundation analysis, linking foundation capacity to the corresponding collapse mechanism in the soil to improve the understanding of foundation behavior. Alongside FELA, the study introduces an innovative machine learning approach using Artificial Neural Network (ANN) and hybrid ANN (ABC, ALO, ICA), Categorical Boosting (CATBoost), Extreme Gradient Boosting (XGBoost), and Gradient Boosting Machine (GBM) to evaluate the correlation between input parameters and their outcomes. The proposed machine learning models are rigorously verified and validated with the machine learning model, showing exceptional agreement with numerical results, as demonstrated by an impressive R2 value. The present study is a practical and efficient method for evaluating the 3D failure envelope of foundations on anisotropic clay under general loading conditions in (V-H-M) space.