Application of model-based damage identification to a seismically loaded structure

Abstract

The aim of the presented model-based damage identification approach is to detect, localize and quantify changes in a mechanical structure due to damage by means of a computational model and measured changes of the structure's dynamic behavior. An inverse sensitivity problem is formulated, leading to a large number of damage parameters when the structure has many structural members. While the number of potential candidates for the damage locations is very large, usually there are only very few active parameters concentrating on the damaged areas. It turns out that the parameter subset selection is an essential step. The method is applied to a large scale structure, the so-called Steelquake structure, a two-storey building, which was subjected to a seismic loading. Cracks at different locations developed during this loading. All crack locations are successfully identified by the algorithm, but one undamaged position was also localized. Additional simulation studies show that errors in the measurement data can cause a false indication of damage.

Global Damage Identification of the ‘Steelquake’ Structure Using Modal Data

“…A detectability study pointed out, which damage locations can be found when using a certain selection of the overall available modal data information [6]. It turned out that, e.g.…”

“…Usually, the initial model has to be corrected by a model-updating procedure so that the model displays the same properties as the undamaged structure within a certain accuracy. This was done by minimising the deviations of the natural frequencies and maximising the MAC values, see reference [6] for the cost functions. Parameters which had to be adjusted where especially the orthotropic material data of the concrete-steel composite material of the two slabs (see Figs.…”

Improved iterative regularization for vibration-based damage detection

“…5,6 However, it is easy to see that the incremental update ∆a k can become small as the number of iterations increases and the penalty term in Equation (15) can disappear. Since only ∆a k is restricted but not a k+1 , the update parameters can grow without restriction, although only slowly.…”

“…In model updating and damage detection, most researchers just apply regularization methods to the linearized problem. 5,6 This strategy is characterized as "linearize, then regularize". However, as pointed out by Vogel,7 it has its shortcomings.…”

Structural damage detection using nonlinear parameter identification with Tikhonov regularization

“…After damage has occurred, the model is updated to match the new measurements and the change in parameters indicates the damage. While nonmodel-based methods have become very popular lately, it is believed that a model gives first one is an application to a highway bridge [12]; the second treats a composite steel-concrete building [13]. The same building is analysed in Reference [14], also using Tikhonov regularization.…”