Research On Structural Damage Identification Based On Wavelet Analysis And Neutral Network

During the actual operation process, the structures may appear the damage because of many reasons. In order to safeguard the security, integrity and the durability of the structure, it's necessary to use the some effective methods to monitor and to appraise the existing structures' healthy condition. In other words the structures must carry on the health monitoring, and the most coral content is the structural damage identification.This paper introduces the wavelet analysis to the structural damage identification, and takes the advantage of its superiority on the processing unstable signal, thus obtains the damage information which can't gain through traditional Fourier transformation , at the same time unifies the ability of the artificial neural network ,handling the non-line information, to fuse the structural damage information and to processes damages information accurately, visually and fast.In the first chapter the goal and significance of this topic research is discussed with the review to latest structural damage identification methods based on the structures' dynamic performance, as well as the domestic and foreign research presentation, and the research work in the paper has been cleared about.In the second chapter the wavelet analysis theory is introduced. Contrasted the difference between the wavelet analysis and the Fourier analysis and the Short-Time Fourier analysis, its characteristic as one kind of outstanding time-frequency analyzing tools is elaborated. It may adjust the scale in the time range and the frequency range, extremely suits to analyze the signals which include the singular point. At the same time the algorithm structures of the multi-scale wavelet analysis and the wavelet packet analysis are presented.In the third chapter the simulating system of the structural damage is simply introduced at first, and then the principle, which the multi-scale wavelet analysis carries on the damage identification, is analyzed and a new real-time structural damage index based on multi-scale wavelet analysis is proposed. This index takes advantage of its superiority on the processing unstable signal, achieves the damageinformation from the acceleration response of the real-time structure, and can identify and locate both the damage time and the damage position. At last, a numerical example is presented to prove this method.In the fourth chapter through the theoretical analysis and the numerical examples it is proved that the relative change index dE'j (n)and the square ratio index cc'j(n) of the wavelet packet component energy and their slope index and the curvature index can localize the structural damage without the mathematics models. This is the supersity that the other present methods can't compare with. At the same time this method not only can accurately recognize structure sole-damage, moreover can also describe the structure multi-damages accurately, visually and fast. Meanwhile, the deviation of the slope index and the curvature index, based on the wavelet packet component energy relative change, proposed a early structural damage warning index, which can accurately judge structural damage and the different parts respond to different sensitivity of the structural damage at the same time.In the fifth chapter the wavelet packet component energy is taken as the training samples of the BP neural network. Two different neural networks are established separately, and they accurately identify the damage location and the damage severity.In the sixth chapter, conclusions of this paper are systematically made. Also, a guideline of further research on this topic is summarized.