Study Of Structural Damage Identification Based On Wavelet Analysis

Because the real-time features and complexity in structural health monitoring (SHM) system, it is difficult to apply damage identification by the general method presently. In the thesis, based on wavelet analysis, multi-scale damage information analysis and multi-sensor information fusion are proposed to finish damage identification, in which the ideas of system identification and pattern recognition are introduced. Combining numerical simulation with model experiment, a rather whole study is conducted by four steps of structural damage identification.To acquire more information, the multi-scale analysis theory is firstly introduced in SHM. After a few of critical issues is investigated in SHM, it is found out that analysis of measure signal of structure is crux of the overall problem. Based on wavelet, a damage identification method with four steps is proposed in SHM, and a flow chart which includes damage alarming, damage validity, damage location and damage qualification is given.The properties of dynamic parameters are investigated when damage occurs in structure. Applying wavelet transform, differential equation of structural dynamic system are decomposed and dynamic parameters are described on different scales. It is proved that more information about structural damage can be acquired by multi-scale analysis based on wavelet. According to the state space equation, the program SDSP is developed to simulate structural damage based on MATLAB. Finishing numerical simulation and temporal location of structural damage under different excitation load, two conclusions that are theoretically significant to select sampling frequency and scale of decomposing in SHM are obtained to be as follow: Given a structural damage, the component of signal on different scales from different sensors contains different information of structural damage; there is better damage index on the scale which is further from frequency band of the excitement loads.The characteristic and essentiality of structural damage alarming are discussed. A method which applied two different wavelet bases to decompose and recompose measure signals is proposed to detect structural damage. By a coefficient of damage alarming which is defined as λk and a threshold which is specified as u, information of emergence and damage time are obtained. According to Lipschitz index of the white noise and the signal of damaged structure, the variation and singularity of signal contaminated by white noise are analyzed on time-scale phase plane, and the method of wavelet coefficient modulus maxima which can realize damage alarming and anti-noise is discussed. A numerical simulation for a continue beam bridge with damage is conducted and the methods of wavelet coefficient modulus maxima and multi-scale signal decompose are applied to realize damage alarming on the condition with and without noises. The effectiveness of damage alarming and ability of anti-noise of the two methods are studied.The feature extraction and pattern recognition are introduced in structural damage identification. The wavelet packet transform is investigated as a novel means of extracting time-frequency information and reducing dimensions of feature from vibration signal in the infinite time domain. Neural network is discussed as a pattern classifier to finish damage pattern classification by inputting node energy feature vector.A numerical simulation of simply dynamic system with and without damage under random forces is conducted to obtain acceleration data, and feature extraction based on wavelet packet and damage pattern classification based on BP neural network are finished. Comparing node energy feature of wavelet packet coefficient with node energy feature of wavelet packet component signal, node energy of wavelet packet coefficient provides more feature information and causes shorter computation time.Based on the ideas of pattern classification and local decision referring whole information, a method of coupling neural network of multi-sensor information fusion is propos...