Numerical And Experimental Studies Of Damage Detection For Shearing Buildings

It is very important to prevent catastrophic structural failures via health monitoring and damage detection, furthermore, the maintenance expense of the structure can be sharply decreased by early detecting structural damages. Therefore, structural health monitoring and damage detection have attracted wide attention in structural engineering field in recent years. Supported jointly by the National Natural Science Foundation of China (No. 50378041), the Specialized Research Fund for the Doctoral Program of Higher Education (No. 20030487016) and the Excellent PhD Dissertation Foundation of Huazhong University of Science and Technology (No. 2003-12), in this dissertation, a two-step damage detection method is adopted for shearing buildings, i.e., the sensitivity analysis of structural modal parameters with respect to damage is conducted firstly; and then, based on the results obtained from the sensitivity analysis, only changes in slopes of the first order mode shape are used to locate damage; finally, damage extents of the structure are identified based on the measured frequencies by using neural network algorithm. In this dissertation, following several aspects are studied theoretically, numerically and experimentally, and some results are achieved: 1) Utilizing the stiffness matrix characteristic of shearing buildings, sensitivity coefficients of modal parameters (natural frequency, mode shape and slope of mode shape) with respect to damage are derived from the free vibration equation of structures. Based on the sensitivity analysis, the variation rule of the sensitivities of modal parameters with respect to damage is found, and the results of sensitivity analysis can be considered as the basis of identifying damage locations and sizes. 2) The damage location method based on only changes in slopes of first order mode shape has been presented, that is to say, the change in slope of first order mode shape of the damaged storey must be greater than zero when only one storey is damaged. The above conclusion has been proved theoretically. Due to the coupling of changes in slopes of mode shapes induced by multiple damages, it maybe fails to locate damage according to those changes directly. Therefore, combined with results of sensitivity analysis, the coupling is eliminated via iterative computation. And then, the problem of locating multiple damages can be solved. 3) Once the damage is located, damage extents can be estimated using neural network algorithm. Because natural frequencies can be measured more accurately than other modal parameters, the change rates of some lower order natural frequencies are chosen as the input of neural network. Based on the frequency sensitivity analysis, those frequencies are selected which can reflect damages sensitively and completely. Thus, the selection of neural network input vectors is improved. 4) The sensitivity analysis of modal parameters of a 10-storey shearing building numerical model is conducted. For absolute sensitivity coefficients, upper modal parameters are more sensitive to damage than lower modal ones and slopes of mode shapes are more sensitive than mode shapes. For relative frequency sensitivity coefficients, sensitivity coefficients of a specific frequency with respect to different damage locations are different and those of different frequencies with respect to a specific damage location are also different. Through frequency sensitivity analysis, some lower frequencies can be found which reflect damages sensitively and completely. Changes in these frequencies can be acted as signal of damage occurrence or used to size damage. Change rates of mode shapes and slopes of mode shapes can not directly reflect damage locations. 5) Three damage cases of the 10-storey shearing building are considered. Locations of damages can be identified using the method based on changes in slopes of first order mode shape. Extents of damages are predicted accurately using neural network algorithm in which change rates of frequencies sensitive to damage are selected as neural network input. 6) The damage detection method is verified via a 3-storey experimental shearing model. Results of sensitivity analysis of modal parameters are similar to those of numerical example, i.e., upper modal parameters are more sensitive to damage than lower modal ones; slopes of mode shapes are more sensitive than mode shapes; sensitivity coefficients of a specific frequency with respect to different damage locations are different and those of different frequencies with respect to a specific damage location are also different. Through using the damage detection method of locating damage by changes in slopes of the first mode shape and sizing damage by neural network algorithm, it is demonstrated that damages are located correctly and errors of identifying damage extents are small.