NASA Uncertainty Quantification Challenge and Treatment of Multiple Uncertainty Types, 18-R8406
Inclusive Dates: 07/01/13 – 06/30/14
Background — Modeling and simulation activities play a key role in decisions associated with design, cost, maintenance, and reliability across a wide range of industries. Nevertheless, the predictions that these decisions are based on are subject to inherent variations in material properties, loading conditions, and other variables, as well as reducible uncertainties attributable to having only limited information or data. The use of probabilistic approaches has seen an increasing level of acceptance as a means for quantifying the impact of such variations and uncertainties on model predictions. However, use of these methods also brings to light new and more challenging questions associated with how different types of uncertainties should be treated. To assess the state of the art and collect proposed approaches in a common problem setting, NASA Langley released a "Multidisciplinary Uncertainty Quantification Challenge," and organized a corresponding special conference session and journal issue to solicit proposed approaches.
Approach — The goal of this research is to develop new methods, build experience, and enhance capabilities for treating multiple uncertainty types in modeling and simulation activities. The approach is to leverage existing SwRI expertise and software tools to develop new capabilities in the areas of sensitivity analysis and model calibration, which are capable of distinguishing between reducible and irreducible uncertainties. The NASA Challenge Problem definition is used both to guide methodology development and also as a testbed to demonstrate the proposed approaches on a realistic engineering application.
Accomplishments — Under this research, new capabilities were developed for calibrating computer simulations using test data, which expand the scope of problems that can be addressed to include additional types of uncertainty. Also, a new approach for sensitivity analysis, referred to as "variance decomposition for statistical quantities of interest," was developed. This approach allows for identification of elements in a model that have the most potential for uncertainty reduction, while accounting for inherent variations that are not reducible. The results were presented at a special session for the Challenge Problem at the AIAA Science and Technology Forum and Exposition, and they were published in a special issue of the Journal of Aerospace Information Systems.