A modular approach to linear uncertainty analysis

This paper introduces a methodology to simplify the uncertainty analysis of large-scale problems where many outputs and/or inputs are of interest. The modular uncertainty technique presented here can be utilized to analyze the results spanning a wide range of engineering problems with constant sensitivities within parameter uncertainty bounds. The proposed modular approach provides the same results as the traditional propagation of errors methodology with fewer conceptual steps allowing for a relatively straightforward implementation of a comprehensive uncertainty analysis effort. The structure of the modular technique allows easy integration into most experimental/modeling programs or data acquisition systems. The proposed methodology also provides correlation information between all outputs, thus providing information not easily obtained using the traditional uncertainty process based on analyzing one data reduction equation (DRE)/model at a time. Finally, the paper presents a straightforward methodology to obtain the covariance matrix for the input variables using uncorrelated elemental sources of systematic uncertainties along with uncorrelated sources corresponding to random uncertainties.