Assessment of relative impact of model uncertainty on the accuracy of nondestructive damage detection in structures

Vita.This dissertation deals with the general problem of nondestructively detecting, locating, and sizing damage in structures via changes in modal responses of the structures. The objective of this study is to assess the relative impact of model uncertainty on the accuracy of nondestructive damage detection (NDD) in structures. In order to achieve this goal, four major tasks are performed: (1) NDD methods to locate and size damage in structures are selected; (2) model uncertainty and NDD accuracy in structures are quantified; (3) a methodology to assess the relative impact of model uncertainty on the NDD accuracy in structures is developed; and (4) the relative impact of model uncertainty on the NDD accuracy in real structures is assessed. In this study, the following four major results are obtained. Firstly, a NDD scheme is selected to locate and size damage in structures with high complexity, with few modes measured, and without baseline modal information. Secondly, model uncertainty measures and NDD accuracy measures are selected. Thirdly, a methodology is developed to estimate the relative impact of model uncertainty on the NDD accuracy in real structures. Finally, the methodology is applied to two real structures and relationships between model uncertainty and NDD accuracy in those structures are produced as follows: (1) for a model plate-girder bridge it is shown that uncertainty in the selection of damage detection model and uncertainty in mode shapes due to errors in modal amplitudes have relatively higher impacts on NDD accuracy than uncertainty in stiffness parameters, uncertainty in the number of modes, and uncertainty in frequencies; and (2) for a complicated three-dimensional truss-type bridge it is shown that all three selected uncertainty types which are uncertainty in the selection of damage detection model, uncertainty in baseline mode shapes, and uncertainty due to missing sensor readings have significant influences on NDD accuracy