EMD Method,Neural Network Theroy And Its Application In Structural Health Monitoring

In this thesis, first, some Time-Frequency analyzing methods are reviewed, analyzing their advantages and disadvantages. Then a new analyzing nonlinear and non-stationary signal method is introduced. And a comparison between it and the other Time-Frequency analyzing methods is done, pointing out the new method's advantages. On this basis, using it associated with RBF neural network in structural damage detection is investigated thoroughly.Most of the former Time-Frequency analyzing methods are derived from the Fourier analysis. So they more or less have the problems exiting in Fourier analysis, for example: the discontinuity expressed with harmonic components, being integral mean of a time interval;and the problems derived from the conventional definition of frequency, that is, the contradiction between the time and frequency resolution. Improving the performance of some parameter is at the cost of sacrificing the other one.Hilbert-Huang transform (HHT) is a new technology for the analysis of the nonlinear and non-stationary signals, which was intrduced by prof. Norden E. Huang of NASA in 1998 on the foundation of classic Hilbert transform. This method consists of two successive parts, i. e., the empirical mode decomposition(EMD) and the Hilbert spectral analysis (HSA) : firstly, an arbitrary nonlinear and non-stationary signal is decomposed into a number of intrinsic mode functions (IMF) by EMD;then, HSA is performed on each IMF, and the Hilbert spectrum of the corresponding IMF is obtained;at last, the Hilbert spectrum of all IMFs are grouped to get the Hilbert spectrum of the original signal. The Hilbert spectrum obtained by this way describes the original signal in the joint time-frequency domain, and possesses time-frequency resolution. The problems existing in sevaral signal processing methods which are based on Fourier analysis are totally overcomed, furthermore, the IMF components decomposed by EMD method have distinct physical senses.In the aspect of the study of HHT method, aiming at the boundary problem inEMD, this text introduces a new algorithm based on RBF neural network. This algorithm uses the RBF neural network to predict the two additional maxima and the two additional minima at both ends of the signal respectively, and then all of the signal's original extrema and the estimated extrema are connected by cubic spline function method to construct the upper and the lower envelopes of the signal, at last the original signal is decomposed into a series of IMF components by the sifting process. The calculating result shows that this algorithm is very available for EMD's exact decomposition, especially for the low frequency IMF components which get very good results.In the aspect of the study of the application of HHT method, This dissertation has done three part of jobs: first, the Elcentro strong ground motion which is in common use in earhtquake engineering is analyzed by HHT method, getting its amplitude-frequency-time distribution and the comparison between the HHT marginal spectrum and Fourier spectrum is also made;second, the typical nonlinear system defined by the Duffing equation is analyzed by HHT method, getting its numerical value's amplitude-frequency-time distribution and the comparison between the HHT marginal spectrum and Fourier spectrum is also made;third, define the stationary degree of non-stationary signals making use of 3-D time-frequency spectrum and marginal spectrum, the stationary degree curve of Elcentro seismic recording is protracted according to its defining formula, the stationary degree curve of Elcentro seismic recording which change with time is protracted according to similiar definition.Structural damage detection using neural network is a problem of mode matching sustantially, including training phase and testing phase, training phase is a process of founding damage mode database, testing phase is a process of damage mode matching, its emphasis and difficulty lies in how to extract characteristic of damage signal efficientfy, that is damage gene. The appearance of EMD method provides a new way to solve this problem. This dissertation puts the IMFs of structural responding signal which are decomposed by EMD method into a designed RBF neural network to train the network in order to detect structural damage, emulational experiment has validated that the combination between EMD method andRBF neural network can not only improve the effiency of learning and training, but also enhance the distinguishing accuracy.At the end of the thesis, the application prospect of EMD method in structural damage detection is revealed, pointing out several problems which aren't considered when researching and putting forward some ideas in dire need of perfecting.