Damage Diagnosis Of Vibration Structure Based On Wavelet Neural Networks

With rapid development of high building, the research detectin structural damage of high building becomes a hot spot now. Accordingly, it is of practical significance to study the damage diagnosis problem of them.Damage diagnosis technique is an active research domain in current structure engineering and has strong engineering background. The correlative theory and technology are developing continuousl. Wavelet analysis is a new time-frequency analysis method internationally in engineering application domain in recent years. It is regarded as a breakthrough of Fourier Analysis. Since the frequency component of the high-frequency region is thicker than that of the low-frequency region, and the structural damage is typically a local phenomenon captured most likely by high frequency modes. The wavelet packet transform (WPT) adopts redundant basis functions and hence can provide arbitrary time-frequency resolution. It districts the frequency in many different bands and further decomposes the high-frequency region. In addition, WPT also can choose the frequency band based on the character of the signal to make it fit for the frequency spectrum and the time-frequency classification efficiency is improved thereby. As a result, the WPT is of great value in practice.This paper investigate the existing method of damage diagnosis. Base Wavelet transform has a good localization character both in time-domain and frequency-domain, and neural network has a good property of nonlinear mapping. and an approach based on energy theory for structural damage diagnosis is presented by combing with advantages of wavelet analysis and neural network. Component energies at different resolution of the wavelet transform are extracted by using the properties of its better multi-resolution analysis. These features are fed into the wavelet neural network as the input patterns for training and classification. Then the occurrence, location and severity of the damage are diagnosed successfully. The results of in-situ test are in good agreement with numerical simulation and it show that this method can successfully be applied to the identification and diagnosis of structural damage with the consummately trained wavelet neural network as anintelligentzed classifier.The component energy based on WPT can perfectly reflect the damage features of the signal in both the frequency change in time domain and the time change in frequency domain, so it is a good candidate indix, it is much more sensitive than the existing methods, and it can detect extremely small cracks in vibrating structural. It is sensitive to structural damage. Results show that it is applicable to use the WPT and the Neural Network in structural damage assessment and the diagnosis result is excellent.The research of this paper is subsidized by Doctor Special Research Fund of Ministry of Education ( 200403 86004).