Study On Structural Damage Detection Using Intelligent Computing

The structural health monitoring and diagnosis is very important for keeping the structure safety and decreasing the cost of maintenance. Many techniques have been proposed and applied these days. Although the diagnosis techniques based on vibration have been used most widely, there are still many problems need to solve, for example, the incomplete of the measurement, the effect of noise, the FE model updating, structural online diagnosis and so on. The focus of this paper is the research and application of complex structural damage detection. Depending on the kind of measurement data, the damage detection techniques is divide into three methods, the method based on modal parameters, the method based on frequency region data and the method based on time region data. The paper does some research on these three kinds of data.The methods based on modal parameters are the most widely used for their complete theory and simple data form, but the identification using the methods is easily affected by the noisy measurement. In order to conquer this problem, a diagnosis model based on updated modal parameters is proposed. With the help of the diagnosis model, a much better identification can be deduced.Structural damage identification can be treated as an optimization problem. This problem is a complex nonlinear problem and hard to get a satisfied result using normal mathematic optimization methods. Genetic algorithm (GA) as a new and promising optimization method has been used to solve this problem, but GA still have some shortcoming, for example, weak local search ability, sensitive to local minimum point and so on. A new method named HOMGA is proposed to improve the ability of GA, which is the combination of micro-GA, orthogonal design and Solis-Wets, moreover the substructural identification is also used. Using the two methods can not only improved the ability and efficiency of algorithm greatly, but also makes it possible to identify the damage of complex structure.Frequency response functions (FRF) are the direct measurement data and include more information about the state of structure, so the method based on FRF should have more precise identification. For the amount of FRF is very large, it is impossible to use them directly. Principle component analysis (PCA) and Independent component analysis (ICA) are used differently to extract the feature of FRF in this paper, Latin hypercube sampling (LHS), neural network (NN), HOMGA and substructural identification are also used to detect structural damage. The simulation indicates that the method have a good ability to resist the noise, less the amount of sensors and good identification.When a structure is damaged, its response signals will be nonstationary. Wavelet transform (WT) is a mathematical tool that can decompose a temporal signal into a summation of time-domain basis functions of various frequency resolutions. This simultaneoustime-frequency decomposition gives the WT a special advantage over the traditional Fourier transform in analyzing nonstationary signals. Dynamic signals measured from a structure are first decomposed into wavelet packet components. The wavelet packet component energies were sensitive parameters and can be used as structural condition signatures. In this study, a novel structural condition index, wavelet packet signature (WPS), is proposed for locating and quantifying structure damage. After extracting the WPS from the response measurement at various locations, the similar coefficients of the spatial distribution curvature of the WPS between the original and the damaged structure are used for confirming the sensitive WT frequency domain, and the difference of the spatial distribution curvature of the WPS is used for locating damage. One special advantage of the proposed method is that it does not require an accurate analytical model of the structure been monitored and is very suitable for on-line diagnosis.