The Method Combining Wavelet And Neural Network To Identify Structure Damage

Recent years large and complex structures are used more and more widely, what's more, existing structures also become old, all this things result in great concern on the identification of structural damage. Form this view, the text concentrates on the identification of structure damage using neural network technique by two-step method.(1) Putting all the signals into the neural network for one time would lead to some bad results, such as unshrinking, long training time, low training-exaction and so on. To solve this question, complex structures is to be decomposed to several substructures in according to certain regular criterion, and then constitute calculating modal, where the substructure is considered as a unit.(2)Firstly, put a set of data including substructures to the self-organizing competitive neural network and train it, this neural network can local damage to the substructure. Secondly, put a set of data including poles of the substructure to the BP neural network, set the number of pole which is damaged as its target, then this neural network cans local damage to the pole which is damaged. Through the two steps, two kind of network are gained, and can be used to identify structure damage. Besides, the impact of network parameters such as studying ratio, training time to the neural network performance are also discussed, and by trial calculating method the suitable value to enhance the capability of neural network is chose.(3) The signals captained in actual engineering always contains much noise ,which will not reflect the dynamic character of structure preciously, and will reduce the exactness of damage identification system, so signal de-noise is necessary before it is used. In the text, signals which contains noise are put into neural network, then this net work become sensitive to signals containing noise, and its capacity is also improved, Wavelet analysis method is used to eliminate noise mixed in the structure frequency, After the operation, the signal/noise is elevation to some extent.