Legitimacy of Model Calibration in Structural Dynamics (Manuscript).

摘要

In structural dynamics, a finite element model is often calibrated to better reproduce experimental measurements collected on the structure. Features commonly used for test-analysis correlation are the resonant frequencies, mode shapes and modal damping ratios of the linear response. When the agreement between measurements and predictions is unsatisfactory, the model is parameterized and calibrated to improve its overall accuracy. We argue that model calibration may not always be legitimate to improve test-analysis correlation. This is because a calibration study attempts to compensate for parametric errors when, in fact, the disagreement between measurements and predictions may originate from other sources. For example, the truncation error caused by a lack-of-resolution of the spatial discretization adds to the modeling uncertainty. In this work, the scaled model of a three-story frame structure that responds mostly in bending is tested experimentally and modeled with finite elements. The agreement between measurements and predictions is assessed relative to the overall level of experimental variability. Truncation error is quantified by performing mesh refinement studies. Guidance on the legitimacy of model calibration is formulated by comparing the overall levels of truncation error, parametric uncertainty, and experimental variability. It is concluded that, while useful, model calibration is a technique that should be deployed only after other sources of modeling error have been rigorously quantified and adjusted for.