Methods for QMU Using Computational Simulation When Resources Are Limited

• Activity Number: 331
• Type: Invited
• Date/Time: Tuesday, August 6, 2013 : 10:30 AM to 12:20 PM
• Sponsor: Section on Statistics in Defense and National Security
• Author(s): Brian Milne Rutherford*+
• Companies: Sandia National Laboratories
• Keywords: QMU ; Computational Simulation ; Reliability ; Response-Modeling ; Epistemic Uncertainty

Abstract

Computational modeling often plays a major role in QMU analyses that address reliability requirements -- computational simulation results provide a basis for the required inferences. The 'probability-of-frequency' (PoF) approach provides a method for uncertainty propagation that maintains a careful distinction and separation between the effects and implications of aleatory uncertainty and the effects and implications of epistemic uncertainty. If simulations are expensive and consequently limited, additional uncertainty can be introduced into the analysis, particularly when the predicted response has to be estimated over a range of input parameter values. This presentation describes a 'response-modeling' approach to uncertainty quantification and propagation that utilizes the PoF approach and captures the additional uncertainty that results from limited simulations and incorporates it into the analysis. The response-model consists of an ensemble of response surfaces over the parameter space that is treated as a discrete probability measure and used to formulate an uncertainty distribution for the response.