Abstract
This paper presents a thorough reliability assessment of cavity foundation systems involving the generation of 272 datasets using Plaxis 2D automation. The parameters were systematically varied across feasible ranges, and Sobol‐based sensitivity analysis identified the negligible influence of the soil modulus of elasticity (E) on subsequent reliability analyses. A robust 1D‐CNN surrogate model was developed to predict the critical foundation responses by integrating Gaussian white noise to simulate real‐world uncertainties. A log transformation with 1,000 bootstrap samples was chosen for resampling non‐normally distributed data. This study employed a novel approach utilising 1D‐CNN regressor models for bearing capacity (BC) prediction, achieving promising results with R2 values of 0.953 and 0.945 for BC in the training and testing phases, respectively. Bootstrapping resampling facilitates reliability analysis preparation and ensures robustness in handling complex data. Simulated noise varied with specific variance (p) from 0.01 to 0.5, allowing the examination of model efficacy under varying noise levels. Both the Monte Carlo Simulation (MCS) and first‐order reliability method (FORM) were employed, revealing a reliability index (β) of 2.046 for FORM and 2.066 for MCS. This indicates a 0.976% increase in β and a 75% increase in the probability of failure transitioning from FORM to MCS, underscoring the model’s sensitivity to analytical methods.