Adapting and updating simulation based probabilistic analyses

Abstract

ABSTRACT: Adapting and updating simulation-based probabilistic analyses This work develops new simulation-based methods to account for uncertainties in the assumed form of input probability distributions and enable a single probabilistic analysis to be updated as input distributions are updated from, for example, Bayesian inference. The method enables existing samples to be probabilistically re-weighted according to a post-stratification of the probability space. With this re-weighting, it is possible to assign samples drawn from any given distribution to any other arbitrary distribution possessing the same sample space. By reweighting existing samples, the method enables a single probability study to produce an ensemble of probabilistic response quantities of interest (e.g. statistical estimates or empirical CDFs) that result from uncertainties in the distributions of the input quantities. Moreover, this process allows sensitivity analysis of the output distribution/estimate to the uncertain input parameters. This aids in identifying those parameters whose uncertainty should be reduced through additional testing and those whose uncertainties have little influence on variability in the resulting distributions. Lastly, upon collection of additional data, the re-weighting process allows the input distributions, and hence the ensemble of response distributions, to be updated directly without the need to conduct a new computationally expensive probability study. The proposed method, referred to as Probabilistically Re-weighted Simulation Sets, will be demonstrated using empirically derived analytical equations for plate buckling with uncertain material and geometric parameters with material data collected through extensive US Navy testing programs.

Document Details

Document Type
DoD Grant Award
Publication Date
May 22, 2016
Source ID
N000141512754

Entities

People

  • Michael D Shields

Organizations

  • Johns Hopkins University
  • Office of Naval Research
  • United States Navy

Tags

Fields of Study

  • Computer science

Readers

  • Computational Modeling and Simulation
  • Regression Analysis.
  • Statistical inference.

Technology Areas

  • AI & ML
  • AI & ML - Bayesian Inference
  • Space