ABSTRACT Reliable design of shallow foundations is critical for offshore and coastal infrastructure such as towers, tanks, and wind energy structures. This study presents a framework that combines finite element limit analysis (FELA), random field theory, and ensemble machine learning to evaluate the probability of failure ( PF ) of ring footings in spatially variable soils. Random fields of soil friction angle ( ϕ ) and unit weight ( γ ) are modelled with lognormal distributions and exponential autocorrelation, while Monte Carlo simulations with FELA generate a comprehensive reliability dataset under varying coefficients of variation, correlation lengths, and safety factors. Several surrogates, including support vector regression, multivariate adaptive regression splines, artificial neural networks, and Kolmogorov—Arnold networks, are benchmarked. A stacked model integrating KAN and ANN with a support vector meta‐learner achieves superior accuracy ( R 2 = 0.993) and consistent calibration. Parametric analyses show that PF is mainly controlled by safety factor and soil variability, while correlation length and ring slenderness exert secondary influences. The framework further provides reliability design charts that incorporate geotechnical variability and geometric effects. These results highlight the efficiency of surrogate‐assisted random field FELA for reliability‐based design, offering offshore engineers practical tools to assess foundation safety under uncertainty.
Vu‐Hoang et al. (Thu,) studied this question.