Identification in Rotordynamics: Uncertainty Analysis and Quality Estimation

Abstract

Critical to the result value of an identification process is establishment of the reliability or accuracy of the identified parameters. Uncertainty in the identification process can stem both from uncertainty in the analytical model and from uncertainty in the test data. The uncertainty propagation turns out to be difficult to estimate due to rather complicated identification process and the dimension of the analytical model. Currently, there is no uncertainty analysis and quality estimation available in the literature to the author’s knowledge for model-based identification in rotordynamics. This paper borrows linear fractional transformation (LFT) and μ-analysis from the controls community to perform this job. The basic idea is that the uncertainty of the identified result can be expressed as a system with uncertainties, and therefore quality estimation is equal to bounding the gain of this system. This system is built in two steps: first, different types of source uncertainties are expressed as LFT format, and second, the whole identification process with uncertainties is transformed into a single LFT format. μ-analysis is then used to bound the gain of this LFT. The uncertainty analysis and bounding algorithm are illustrated with the same experimental data used in the last paper, for both model-based and direct measurement methods.