Reliability analysis of performance functions via adaptive sequential sampling with detection of failure surfaces

Reliability analysis of performance functions via adaptive sequential sampling with detection of failure surfaces

Stať ve sborníku mimo WoS a Scopus

We propose an improved method for estimating rare event probabilities in computational models with smooth performance functions. Building on a previously developed robust strategy for generally non-smooth or discrete-state performance functions, we enhance scalability and efficiency by replacing the original nearest-neighbor surrogate with a Gaussian process regression model. This surrogate leverages numerical limit state values to preselect potential candidates in an active learning scheme that balances exploration and exploitation. The resulting method significantly reduces the number of required evaluations, particularly in low-dimensional problems, while extending applicability to higher-dimensional settings.

We propose an improved method for estimating rare event probabilities in computational models with smooth performance functions. Building on a previously developed robust strategy for generally non-smooth or discrete-state performance functions, we enhance scalability and efficiency by replacing the original nearest-neighbor surrogate with a Gaussian process regression model. This surrogate leverages numerical limit state values to preselect potential candidates in an active learning scheme that balances exploration and exploitation. The resulting method significantly reduces the number of required evaluations, particularly in low-dimensional problems, while extending applicability to higher-dimensional settings.

ψ criterion, failure probability, active learning, gradient-free optimization, Kriging, im portance sampling

ψ criterion, failure probability, active learning, gradient-free optimization, Kriging, im portance sampling

VOŘECHOVSKÝ, M.

2026

17.05.2025

CIMNE

14th International Conference on Structural Safety and Reliability

1

11

11

https://www.scipedia.com/public/Vorechovsky_2025a