Linear/Nonlinear Reduced-Order Substructuring for Uncertainty Quantification and Predictive Accuracy Assessment

Abstract

Modal testing is routinely performed on space craft and launch vehicles to “verify” (calibrate and validate) analytical models. Large multi-component structures such as the Space Transportation System (STS) or “Space Shuttle,” and the newly proposed NASA Space Launch System (SLS), are impractical to test in their assembled configurations. Alternatively, substructure testing of these multi-component structures has been performed and used to calibrate/validate analytical models of the substructures, which are then assembled to analyze system response to applied loads. This paper describes a methodology applicable to uncertainty quantification (UQ) of reduced (modal) models of linear (or linearized) finite element models of substructures, as well as reduced (stochastic neural net (SNN)) models of nonlinear substructures, such as joints. The UQ at reduced substructure levels is propagated to higher levels of assembly by efficient means, enabling predictive accuracy assessment at the coupled system level. UQ is based entirely on comparisons on analysis and test data at the substructure level so that recourse to the specification and quantification of element-level random variables is not necessary. Examples are presented to illustrate the methodology.