Comparison Resrerch Of Stochastic Subspace Method For Automatic Identification Of Bridge Modal Parameters Based On Clustering Algorithms

In the vibration test signal analysis of large and complex structures,achieving automatic identification of modal parameters is a key issue in the field of bridge health monitoring.Among the methods for automatic identification of modal parameters,the methods based on clustering algorithms are more common,but few people compare the identification effects of different clustering algorithms,there is a problem of initial value sensitivity among commonly-used clustering algorithms,and the influence of whether to reject spurious modal parameters on the clustering results is rarely studied.Based on the stochastic subspace method and clustering algorithms commonly used in the field of bridge modal parameter identification,the density peak clustering algorithm and Gaussian mixture model are introduced to compare and study the identification effects of different algorithms and whether to reject the influence of spurious modal parameters.The main contents include:(1)The stochastic subspace method was reviewed,the automatic identification method of modal parameters was introduced,the clustering algorithms commonly used in the field of bridge modal parameters identification were summarized,the Gaussian mixture model was introduced to analyze the stabilization diagrams,the density peak clustering algorithm was introduced to solve the sensitive initial value problem of clustering algorithms,the principle and identification process of clustering algorithms mentioned are clarified.(2)The exploratory data analysis method was used to screen the measured data of a large-span suspension bridge and a cable-stayed bridge,the adaptive variational modal decomposition method was used to complete the noise reduction and reconstruction of the measured signals.(3)Based on the identification results of the data-driven stochastic subspace method,the density peak clustering algorithm was used to determine the initial clustering centers and numbers,and different clustering algorithms were used to analyze the stabilization diagram,studied the influence of whether to reject the spurious modes on the clustering results.(4)The following conclusions can be drawn: the density peak clustering algorithm can effectively reduce the result fluctuation of the clustering algorithms;the results identified without rejecting spurious modal parameters are more than the results identified after rejecting spurious modal parameters;the fuzzy C-mean algorithm has the best identification effect and identifies the most modal parameters;the Gaussian mixture model algorithm has higher requirements for the initial clustering centers,it is not suitable for automatic identification.