A machine learning regression approach for predicting uplift capacity of buried pipelines in anisotropic clays

The uplift capacity of pipeline systems in geotechnical engineering is influenced by internal loading and external factors, making it a significant consideration in pipeline design problems. Previous research has conducted experimental tests and numerical solutions to investigate the relationship between force and displacement or the resistance of pipelines in numerous soil media. This paper proposes a machine-learning regression technique to predict the uplift capacity of buried pipelines in anisotropic clays with parametric analysis. Specifically, the Multivariate Adaptive Regression Spline (MARS) is employed to establish the relationship between input parameters, namely the depth ratio (H/D), anisotropic strength ratio (r_e), load inclination (β), overburden factor (γH/S_uc), adhesion factor (α), and the output uplift resistance (N) obtained from the finite element limit analysis (FELA), utilizing the AUS material model integrated with the OptumG2 commercial program. Furthermore, the sensitivity analysis outcome shows the embedded depth ratio is the most critical parameter, followed by the anisotropic strength ratio, overburden factor, load inclination, and adhesion factor. Additionally, the shear velocity field contours show that when the depth ratio and the load inclination increase, the dissipation of shear changes. Design data visualizations, tables, graph contours, and empirical equations are created and can be utilized to aid in the development of practical designs.