| As safety requirements increase,there is a growing concern about the service life and health of structures.In order to achieve real-time monitoring and damage diagnosis of structures,health monitoring has become a hot research topic in the last three decades.With the development of science and technology,intelligent monitoring guided by machine learning algorithms is also flourishing,among which cluster analysis is widely used as a powerful tool for data mining,such as personalised recommendation,abnormality monitoring,feature learning,etc.In this paper,the Gaussian mixture model in cluster analysis will be applied to the health monitoring system to detect and classify abnormalities in the structural state vectors obtained from vibration response signal analysis.In this paper,two types of structural state vectors are established,one is the structural state vector constructed from frequency and vibration components;firstly,the frequency decomposition method and the random subspace method are jointly used to identify the inherent frequency of the structure from the acceleration response under the structural environmental excitation,based on which a narrow band filter band is selected,a principal component analysis is carried out to obtain the vibration components of the structure of that order,and it is combined with the frequency to form a certain dimension of the structural state vector.The second type is to use the power spectrum in a band containing a large number of structural frequencies as the structural state vector,first calculate all the power spectrum curves during the monitoring period at a certain time interval,select the power spectrum vector in a particular band in the lossless state as the reference,and calculate the distance from the reference state vector at subsequent monitoring times.The distance values can be Euclidean distances or model assurance criterion(MAC).The Gaussian mixture model is then used to classify the distance or MAC values to achieve structural state classification and damage identification.In addition,the power spectrum curves at successive moments are also used to generate a time domain 3D diagram as a reference to assist in determining the current structural state by the stability of the ridges.To validate the proposed method,simulation data of ASCE-Benchmark structures,test data of laboratory experimental steel girders and monitoring data of a suspension bridge for 233 consecutive days were analyzed,and it is clear from the results that both types of structural state eigenvector indicators can well monitor the changes of the structure,and through Gaussian mixture model clustering analysis,the classification of structural state and the definition of threshold values for each state are achieved.The results show that both types of structural condition vector indicators can be used to monitor changes in the structure and to classify structural conditions and define thresholds for each condition through Gaussian mixture model clustering analysis. |
