Time Series Symbolization Based Bridge Structural Damage Status Multi-step Identification Method

With the increasing number of large-scale bridge structures and the widely application of structural health monitoring system on large-scale bridge,the monitoring data present significant characteristics of “big data” such as huge data volume,variety of data type,and increase rapidly.The data storage and transmission cost,computational efficiency and accuracy are the key problems that need to be solved in order to effectively analyze the “bridge structural big data”.In this paper,an effective and flexible bridge structural damage status multi-step identification method framework based on time series analysis via studying the characteristics of bridge structural health monitoring big data.At first,a bridge structural damage early warning method based on training set optimization is puts forward.Clustering method is applied to optimization the original training set so as to obtain a refine training set containing most of the original training set's information,which improves the training set's quality,reduces its scale and promotes the efficiency of time series data mining.Then the refined training set is used to classify the test set's data and assess the bridge structural health condition.Afterwards,experiments on the ASCE Benchmark structure dataset is carried out to demonstrate the feasibility and effectiveness of this method.Then,a bridge structural damage status recognition method based on symbolic approximate time series model is proposed.The symbolic aggregation approximate time series model is used to segment and symbolize the original bridge structural health monitoring data.Compared with traditional data compression techniques such as discrete wavelet transforms and discrete Fourier transform,SAX has the advantages of small storage capacity and can express raw data with different granularity.Therefore,the compression ratio can be selected according to specific requirements while compress the bridge monitoring data.For the monitoring data of health,a low-granularity data model can be used to achieve high compression ratio data representation and transmitted to the data processing center for storage.And the damage status monitoring data adopts the high granularity data model to represent the monitoring data in more detail,and is stored in the local area for convenient real-time processing and analysis.Thus the problem of bridge structural monitoring data compression under the big data environment is solved.Through the good performance on the ASCE Benchmark structure dataset,it is proved that this method can effectively improve the efficiency of damage identification of the bridge structure.Finally,this paper integrates the previous two methods and proposes a multi-step method for identifying the bridge structural damage status based on the training set optimization and the time-series symbolization.The proposal is validated by the experiments using the real bridge data set provided by Harbin Institute of Technology.Experiment results demonstrate the effectiveness of the proposal and its feasibility in engineering applications.Further experiments prove that the framework has higher efficiency and accuracy than traditional structural damage identification methods.The bridge structural damage status multi-step identification method based on training set optimization and symbolization of time series proposed in this paper is conducive to improving the real-time of data processing and accuracy of data analysis,which also provides a theoretical basis and technical support for the analysis and assessment of bridge structural health status under big data environment.