Predicting failure envelopes of skirted spudcan footings under combined loads using ACOoptimized extremely randomized trees

Skirted spudcan foundations are widely employed in offshore geotechnical engineering due to their enhanced penetration capability and superior load resistance in soft clay soils. In this study, the undrained failure envelope of skirted spudcan subjected to combined vertical, horizontal, and moment (VHM) loading is investigated using Finite Element Limit Analysis (FELA). A total of 624 simulations are performed in OptumG3, systematically varying the embedment ratio (L/D), soil strength heterogeneity (κ), vertical load mobilization (V/V₀), and loading limit for the FELA (β) to construct the VHM failure envelope in non-homogeneous clay. To complement the numerical approach and mitigate computational intensity, a machine learning model is developed using Extremely Randomized Trees (ET) optimized via Ant Colony Optimization (ACO). The resulting ET-ACO model demonstrates excellent agreement with the FELA outcomes, achieving R² values exceeding 0.998. Feature importance analysis highlights the FELA loading limit (β) and embedment ratio (L/D) as the most influential parameters governing failure capacity. This data-driven methodology provides a reliable and effective alternative for evaluating offshore foundation performance, as it not only accelerates the prediction of failure envelopes but also significantly reduces computational costs.