QUICKER: Quantifying Uncertainty In Computational Knowledge Engineering Rapidly—A rapid methodology for uncertainty analysis

Abstract

Most engineering systems have some degree of uncertainty in their input parameters, either of a stochastic nature or on account of a lack of complete information. The interaction of these uncertain input parameters, and the propagation of uncertainty through engineering systems, lead to the stochastic nature of the system performance and outputs. Quantifying the uncertainty in an experiment or computational simulation requires sampling over the uncertain range of input parameters and propagating the uncertainty through a computational model or experiment to quantify the output parameter uncertainty. Conventional direct sampling methods for input uncertainty propagation, such as Monte Carlo sampling or Latin Hypercube sampling, require a very large number of samples for convergence of the statistical parameters, such as mean and standard deviation, and can be prohibitively time-consuming. This computational tedium has been partially eliminated through the use of meta-models, which approximate a computational simulation or experiment via a response surface, but the computational time savings from these models are limited to systems with a small number of uncertain input parameters. Toward addressing the challenge of input uncertainty propagation, this paper presents a new uncertainty analysis methodology, QUICKER: Quantifying Uncertainty In Computational Knowledge Engineering Rapidly, that can reduce sample sizes by orders of magnitude while still maintaining comparable accuracy to direct sampling methods. In this paper, the QUICKER methodology is described and demonstrated with both analytical and computational scenarios.