Speaker
Description
Computational models play a crucial role in decision-making across various disciplines, including engineering, science, economics, policy-making, and beyond, especially when input randomness introduces stochasticity into the outputs. Sensitivity estimation (SE) is essential for understanding how input variability influences these stochastic outputs. Unlike traditional methods that focus directly on input values, this study addresses a critical gap by examining the impact of input distribution hyperparameters on output uncertainty.
We present an SE methodology that is designed to evaluate how the parameters of input distributions influence the moments and cumulative distributions of the outputs. Novel sensitivity indices (SIs) are proposed, grounded in the first three moments and cumulative distribution function of the outputs. This approach can as such be naturally extended towards exceedance probabilities A numerical approach is developed to compute these SIs as part of the post-processing phase of uncertainty quantification, utilizing a moment-based model to approximate the output distribution. With the developed approach, there is no extra model evaluations required for SE.
Three illustrative examples are investigated, including nonlinear expressions and finite element models, which demonstrate the efficiency and versatility of the proposed SE method. The results highlight its potential to provide a more comprehensive understanding of the interplay between input distribution parameters and stochastic outputs, offering valuable insights for more informed decision-making in computational modeling.
