Optimizing uncertainty estimation in Enhanced Monte Carlo methods

Abstract

The probability of failure serves as a key metric in a structural reliability analysis, but its accurate estimation remains computationally demanding, particularly for low-probability failure events. The Enhanced Monte Carlo (EMC) method has been developed in order to alleviate from inefficiencies due to the high number of required simulations. Recent advancements integrate Machine Learning techniques with the EMC to further accelerate the estimation process. However, a critical limitation of EMC lies in its fitted confidence interval (CI) estimation, which tends to overestimate uncertainty, leading to unnecessary computational overhead. This study proposes a new prescriptive CI formulation constructed from the method’s hyperparameters, offering a more accurate and computationally efficient approach to uncertainty quantification. The method is general and can be applied to any reliability problem that can be described by a probability curve. The effectiveness of the proposed method is demonstrated through a benchmark reliability problem and a real-world marine structural application. The results indicate significant improvements in efficiency without compromising accuracy, paving the way for enhanced structural reliability assessments.