Numerical investigation and ML-based formulation for conventional and modified suction caissons subjected to unidirectional and combined loadings

The modified suction caisson (MSC), incorporating an external short-skirted structure in addition to its inner caisson, represents an innovative variant of the conventional suction caisson (CSC) in offshore engineering. However, currently available studies lack a comprehensive comparative analysis of CSCs and MSCs and a unified prediction formulation for these foundations, which are essential for practical engineering design. To address this gap, this study systematically compares bearing capacities of CSCs and MSCs under unidirectional and combined loadings using finite element limit analysis (FELA), and proposes a unified failure envelope formulation via the evolutionary polynomial regression (EPR) machine learning technique. In particular, key influencing factors including caisson geometries, soil strength profiles, and interface adhesion are also systematically analyzed. FELA results reveal design recommendations: (i) the failure envelope of MSC and CSC is primarily influenced by the embedment ratio of the inner caisson, with minimal effects from interface adhesion or soil strength heterogeneity; (ii) a 20% increase in external skirt width results in improvements of at least 20% and 23% in vertical and lateral bearing capacities, respectively; and (iii) the lateral capacity becomes independent of the external skirt length once it exceeds 20% of the inner caisson length. Using the FELA dataset, the EPR technique can provide unified and effective bearing capacity formulations.