Research On Frame Structure Damage Identification Based On Empirical Mode Decomposition And Wavelet-neural Network

Frame structure is a common stressed structure in life.It is widely used in houses,factories and bridges.However,due to the influence of natural environment,external load,self-defects and other factors,the structure will be damaged to varying degrees.Therefore,it is of great significance to conduct damage identification research on structures.Based on the existing research results,this paper proposes a frame structure damage identification method based on empirical mode decomposition(EMD)and wavelet-neural network.The main research contents and work are as follows:(1)The research compared the noise reduction effects of three noise reduction methods on noisy signals.The three methods include: wavelet threshold noise reduction method,EMD-wavelet threshold joint noise reduction method and removing high frequency signal noise reduction method.The research results show that the EMD-wavelet threshold joint noise reduction method has the best noise reduction effect.(2)A new damage index is proposed and its damage identification effect is studied:a two-story,two-span×two-span frame structure model is established as the research object,and the acceleration response signal of the structure under the impact load is obtained,Constructed a new damage identification index named node comprehensive acceleration square difference.The damage index combined with wavelet transform is used to identify the damage location,and the effect is compared with the node comprehensive acceleration difference of the existing index.It is found that the damage recognition effect of the comprehensive acceleration square difference index of the node is better than the comprehensive acceleration difference index of the node,it can better identify the damage location of various working conditions.(3)A damage identification method based on EMD-wavelet threshold joint denoising and wavelet transform is proposed: considering the influence of noise on the recognition result,perform wavelet transform on the noisy signal,wavelet threshold denoising signal,and EMD-wavelet threshold joint denoising signal respectively,identify the location of structural damage according to the height of the peak of the detail signal after transformation.The results show that wavelet transform of noisy signal and wavelet threshold denoising signal wavelet transform cannot judge the damage location,while the wavelet transform of the signal after EMD-wavelet threshold de-noising can better judge the damage location,however,it is difficult to correctly identify the damage location when the noise component is large and the damage is small,indicate that this method has a certain feasibility,but it still needs further improvement.(4)The non-linear mapping relationship between damage identification index and damage state is established,and the damage degree of the frame structure is recognized by BP neural network and RBF neural network.The research results show that these two neural network models can better identify the degree of damage and meet the error requirements,which proves that it is feasible to use artificial neural networks to determine the degree of structural damage.