Automatic Modal Parameter Extraction Of Engineering Structure And Separation Of Influence Of Temperature

Once the engineering structure is damaged during the service period,it will cause significant social and economic losses.Therefore,it is an urgent problem to realize the early warning of structural safety under operational conditions.At present,the commonly used method is continuous on-line dynamic monitoring of engineering structures,combined with real-time processing of collected data to identify modal parameters in order to detect the early deterioration of structural stiffness.Continuous on-line monitoring is bound to produce a large amount of data,so it is necessary to propose effective methods to automatically acquire and track structural modal parameters from monitoring data.Secondly,in normal operation environment long-term modal parameters of engineering structures will inevitably be affected by environmental factors,which will mask the early stage of engineering structures.It is necessary to separate the influence of environmental factors on modal parameters and propose only sensitive indicators for structural deterioration.Aiming at the problem of automatically obtaining and tracking modal parameters of engineering structures from a large amount of data,the Stochastic Subspace Identification(SSI)method is widely used at present,however,it often generates noise points in the construction of stable diagram,which leads to the inaccurate extraction and tracking of modal parameters.In this paper,the determination the order of the stable diagram is firstly discussed to select the stable point,and then the Density-Based Spatial Clustering of Applications with Noise(DBSCAN)method is to denoise the stable graph.The problem of noise interference in the acquisition and tracking of engineering structural modal parameters is solved by combining DBSCAN clustering algorithm analysis with SSI method.Meanwhile,a MATLAB program for automatic data processing is compiled according to this method.In addition,a numerical model of Rainbow Bridge in Shenzhen University Town is established by ANSYS software,and the modal parameters of the numerical simulation are obtained,which are compared with those obtained by the modal experiment.The results of the two methods are verified by each other,which proves that DBSCAN clustering algorithm is feasible to denoise the stable diagram constructed by SSI method.Then,the proposed method is used to process the long-term data of Rainbow Bridge,and the 320-day continuous variation of modal frequencies of Rainbow Bridge is obtained,through which the structural changes in the construction stage are detected.At last,aiming at the problem that environmental factors masking structural changes,this paper compares the relationship between fitting temperature and frequency between BP neural network and multiple linear regression model,and finds that the former method performs better in data fitting.Furthermore,based on the five-year modal frequency and temperature data of Pedro e Ines pedestrian bridge in Portugal,this paper chooses BP neural network model and Extreme Learning Machine(ELM)to fit the relationship between frequency and temperature in the first year.By comparison,it is found that ELM model performs better than BP neural network model.Furthermore,the frequency of Pedro e Ines pedestrian bridge in the following four years is predicted by using the fitted ELM model to remove the influence of environmental temperature and early structural changes are also found.