Fitting parameters in the failure envelope expression are influenced by the complex boundary conditions. This paper carried out a comparative study on the performance of five machine learning (ML) algorithms, including decision tree (DT), support vector machine (SVM), random forest (RF), multilayer perceptron neural networks (MLPNN) and gradient boosting machine (GBM), in the prediction of fitting parameters. Results show that the GBM algorithm has the best performance in predicting fitting parameters among all considered ML algorithms. The relative importance of variables influencing fitting parameters was also investigated.
It is usual to consider a shallow foundation as having an embedment depth to foundation diameter ratio less than one. Shallow foundations are widely used in offshore engineering due to their reliable capacity, installation convenience and cost-effectiveness (Barari & Ibsen, 2012; Kourkoulis et al., 2014). They are generally subjected to combined vertical (V), horizontal (H) and moment (M) loading resulting from the harsh environmental conditions (Shen et al., 2017; Fu et al., 2017). The failure envelope approach has been increasing popularly employed to define the load-carrying capacity of shallow foundations under combined V-H-M loading conditions, and its advantages are well-acknowledged (Gottardi & Butterfield, 1993; Gourvenec & Randolph, 2003; Gourvenec, 2011; Vuple et al., 2014), such as explicit consideration of (H, M) interaction and concurrent consideration of foundation geometry, embedment and soil strength profile. An algebraic expression is commonly introduced to describe the failure envelope (Gourvenec & Barnett, 2011; Feng et al., 2014) with fitting parameters for different boundary conditions (e.g., different combinations of embedment ratio d/D and soil strength heterogeneity index κ) provided in a table. Other boundary conditions that are not available in the table can be obtained through interpolation (Gourvenec, 2007; Vulpe et al., 2014; Vuple, 2015).
Compared to simple linear interpolation of fitting parameters, the machine learning (ML) algorithm provides a potential solution to understand the complex interaction among several variables that influence failure envelopes (Zhang et al., 2020). The application of ML algorithms in engineering has recently proliferated such as offshore and geotechnical engineering because of their high accuracy and computational efficiency (Chen et al., 2019; Zhang et al., 2019).
