An Identification Method For Bridge Dynamic Signal Noise Mode Based On Multi-scale Local Pattern Filtering

As the basis of important scientific problems in the field of bridge health monitoring,such as structural modal identification,finite element model modification,and damage identification,the quality of bridge dynamic signal will directly affect the solving effect of these problems.The filtering for dynamic signal has very important research significance in this field.Aiming at the shortcomings of traditional Hilbert-Huang theory in filtering,they are mainly appeared as the mode mixing phenomenon in the adaptive decomposition stage and the components selection in reconstruction stage.For the shortcomings of wavelet threshold de-noising method,they are mainly appeared as the down-sampling and randomness of threshold determination.These problems will lead to the performance degradation of the final filtering,resulting in incomplete de-noising and the loss of effective modes.In this regard,a large number of scholars had carried out the algorithm improvement research and made different degrees of progress.In this paper,aiming at the shortcomings of the above two theories in signal filtering,the two theories were combined and improved on each other.An identification method for noise mode based on multi-scale local pattern filtering was proposed which was applied in the real bridge engineering,the proposed method achieved an excellent effect.This paper relies on the National Natural Science Foundation of China(41430642,51108207),and Science and Technology Development Plan of Jilin Province(20180201083SF),the following researches have been done:(1)For the shortcomings of EEMD(ensemble empirical mode decomposition)and CEEMD(complementary ensemble empirical mode decomposition)filtering,this paper proposed an optimization filtering method with adaptive decomposition and reconstruction based on multi-scale local pattern filtering.Firstly,for the deficiency of completeness and endpoint effect problem,the CEEMDAN(complete ensemble empirical mode decomposition with adaptive noise)was used in initial decomposition stage and endpoint mirroring method was used in the process of decomposition.Secondly,for the mode mixing phenomenon in decomposition,the MPE(multi-scale permutation entropy)algorithm was used to find the IMFs(intrinsic mode functions)with high randomness and these IMFs were integrated to form the noise integration signal.The multi-scale local pattern filtering was used to separate the effective modes and noise modes in it to achieve the purpose of anti-mode mixing.And the filtering result was reconstructed with residual components,the endpoint mirroring-EMD(empirical mode decomposition)was used again to complete the final adaptive decomposition.Finally,an optimization balance model for similarity and smoothness was established for optimization reconstruction to complete the final filtering.(2)For the problems of wavelet threshold de-noising method,the multi-scale local pattern filtering for noise integration on the basis of CEEMDAN uses SWT(stationary wavelet transform)and SVM(support vector machine)as the core algorithm,considering local pattern difference of different frequency bands.The noise modes identification was regarded as the problem of pattern classification.Firstly,the multi-scale decomposition in time-domain based on SWT for noise integration was carried out,and local windows of each sub-band signal were used as the test set.Secondly,the features approximation method was used to establish the training sample signals of noise and effective patterms.Meanwhile,the idea of data localization based on the f-fuzzy granulation was used to extract multiple local features of sample signals.And dimensionality reduction analysis for features was carried out.Finally,the SVM as the core of machine learning algorithm was applied in the local noise pattern recognition to complete the multi-scale local pattern filtering.(3)The simulation model for bridge dynamic signal was established for experiment,and the EEMD,CEEMD and wavelet threshold de-noising method were applied in the simulation experiment to contrast with the identification method for noise mode based on multi-scale local pattern filtering.The research showed that the adaptive decomposition stage on the basis of the proposed method has more excellent ability of anti-mode mixing and suppressing endpoint effect than traditional Hilbert-Huang theory.Meanwhile,the proposed method has better completeness.Compared with the above 3 traditional methods,the final filtering result based on the proposed method has more optimized similarity,smoothness and power reserve characteristics.And the filtering result has higher correlation degree with the real mode in simulation signal,the filtering performance is excellent.(4)The proposed method was applied in the dynamic signal of Changtai bridge for theoretical verification.By the comparative analysis with EEMD,CEEMD and wavelet threshold de-noising method,from the perspective of real application,it was further indicated that the adaptive decomposition based on the proposed method has more excellent decomposition ability than EEMD and CEEMD.Meanwhile,the optimization filtering of the proposed method has more optimized time-,frequency-domain characteristics and filtering performance than traditional filtering method on the basis of excellent decomposition performance.The filtering result of the identification method for noise mode based on multi-scale local pattern filtering has higher signal-to-noise ratio and reliability.It can be used as the basis of bridge health monitoring technology and has very important scientific and engineering significance.