Methods Of Structural Damage Identification From Sparse Modal Response

Damage identification and health monitoring of large-scale structures are important challenges to engineering research. One common approach is to employ the vibration characteristics of a structure to predict the damage locations and to estimate the amount. For damage identification from sparse modal response, the main problem is the multiplicity of parameter estimation solutions arising from using spatially sparse and noised-polluted data. It is investigated espcially in this research. Structural damage identification of engineering structures can be achieved by estimating the physical design parameters from the measured dynamic characteristics. Damage is usually characterized by a reduction in a stiffness property of a parameterized finite-element model of the structure. The process is often posed as an optimization problem based on modal data matching. The methods attempt to minimize a no-linear error function between the analytical and measured modal parameters.The initial effort of this research investigates the theory of dynamic sensitivity. The theory formulated the first and second derivative of some types of error functions. It supports various applications including structural dynamic sensitivity analysis, parameter selection, damage location and damage assessment. For determining statistical distribution of the solutions according to different types of error objective functions, a united approach is presented based on the random starting point scheme and the optimum sensitivity method. The applications are studied and demonstrated using simulation data.A damage identification method for the continuum structures with making use of the blanketing effect is proposed in this paper. The presented method takes the change rates according to certain diagonal elements of the damaged structure flexibility matrix as identification indicator function. The "blanketing effect" of the identification indicator function with respect to the damage factors makes it more suitable to identify the damage of multi-span continuum structures. By combining natural frequencies and mode shape data at a few selected freedoms and the mode weight coefficients, the improved damage indicator function is obtained with better localization property to the damage factor distribution. The sensitivity matrixes of the indicator functions are studied and the strategies to analyze the localization property of the indicator function quantitatively are given. Then the damage indicator functionis is investigated as input to neural networks for the continuum structures damage identification. The numericalsimulations are performed and show the damage indicators are especially suitable for the damage identification of the multi-span continuum structures such as beam and plate.For the damage identification using only a few testing points information, the application of Genetic Algorithm in structural damage detection is investigated in this paper. The method of pile integrity detection is studied using GA and regularization method. The inverse-problem is translated into an optimization problem which minimizes the error function of frequency response in certain constraint conditions and is solved using genetic algorithm. Some numeric simulations and engineering examples are given and all show good results. The presented method can determine the parameters of piles such as cross area distribution, length, wave velocity and so on in a wide range. It is a type of test-analysis integration which makes the work of pile testing easy to do and the result more accurate.A new parameter group-updating scheme is proposed to localize damage in a structural system taking into account the multiplicity of parameter estimation solutions arising from using spatially sparse and noised-polluted data. The proposed scheme detects the dominance element parameters and updates the reference parameter values sequentially. Then the parameter group is expanded and updated. In the certain updating step, only a small scale optimization solution procedure is involved. Based on the proposed group-updating scheme, the damage identifications on a freely supported beam and the IASC-ASCE benchmark structure are studied. The simulation results of the model update strategy show robust property to processing spatially sparse and noised-polluted data.In the last of this dissertation, the research is summarized and the future extensions of the relevant study are discussed.