Quantification and Error Source Classification of Uncertainty in Experimental Modal Data Supported by Stochastic Finite Element Analysis

Abstract

The use of uncertainty propagation methods in finite element analysis has become more and more feasible in recent years due to increasing computer power. Among other things, uncertainty propagation is utilized to ensure the robustness of the computational model, or respectively, of the structural design. The determination of the stochastic re-sponse relies strongly on the assumptions made for the variability of the uncertain parameters. Mainly because it is im-possible to make reliable assumptions for the scatter of uncertain modeling parameters, like for example joint stiffness parameters. Such analyses are still prone to prediction errors. Consequently, an inverse approach based on experimen-tal modal data is required to quantify parameter uncertainties in a finite element model. Computational Model Updating (CMU) techniques offer the possibility to identify model parameters from measured response data. However, statistical test data are required for uncertainty identification by using finite element models and computational model updating techniques. Thus, it is necessary to analyze the impact of different possible sources of uncertainty on the test data vari-ability prior to CMU. This paper comprises the results gained from multiple modal survey tests performed on a replica of the well known GARTEUR-AG19 benchmark structure at the German Aerospace Center (DLR) and subsequent multiple extractions of modal parameters. In particular, the different sources and the corresponding magnitudes of uncertainty are illustrated. Comparison are made to express the variances resulting from different sources in the experimental modal analysis proc-ess which allows for a classification of the error sources (ranking with respect to the relative impact on the variability of the test data). The major error source will be analyzed in detail such that a stochastic finite element approach can be used to reproduce the scatter observed in the test data.