Research On Modal Parameter Identification And Online Monitoring Of Structure Based On Subspace Algorithm

Modal parameter identification for engineering structures has been widely applied in many fields. Classic modal analysis methods use the relationship between the excitation and response to identify the result. Information of excitation can not be gotten from large-scale structures during the experiment, so the methods, which identify parameter using output data only, has been paid more attentions. Based on the stochastic subspace identification method, the paper focuses on the modal parameter identification and online diagnosis from ambient vibration data. The content is list below:1) The stochastic subspace method has been introduced deeply. The influence using different weighted matrix has been researched. After that the precision and stability of the method is analyzed carefully.2) Based on the stochastic subspace method, the projection from future data row space to past data row space is modified to an update mode for subspace tracking. The left singular vector of the projection has been tracked by the projection approximation subspace tracking algorithm. Then the modal parameter is obtained by least square method. At last an experiment is made using a cantilever beam to prove the efficiency and stability of the method by changing the frequency domain of the exciting signal. The results show that good precision and stability will be obtained at same time by choosing an appropriate forgetting factor B.3) The relationship between blind source separation and signal of vibration response has been introduced. Every mode of vibration response is separated by the technique of blind source separation. In order to increase the precision of separation, using the IMF (Intrinsic Mode Functions) given by EMD (Empirical Mode Decomposition) as the reference signal, calculating the covariance matrix between the IMF and pre-whitening signal. The decorrelation matrix has been gotten by joint diagonalization of several matrixes with different time delay. The all process of separation has been optimized and the precision of separation has been improved. The ending effect of EMD happen to be corrected during separation and the characteristic of EMD, analyzing nonstationary signal, has been preserved, and same time adaptability of the separation algorithm has been improved4) Expend the SSI from time domain to frequency domain. A subspace method in frequency domain has been found by using the relationship of power spectra of time t and time t+1. Because the engineering restriction of number of reference point, based on the original arithmetic, frequency domain subspace method using one reference point and part referent points has been developed. Adaptability of the method has been improved greatly. Because frequency domain subspace method do not relay on the interval of frequency, efficiency of the method has been increased by using part data round of peak of power spectra. The simulation shows that result get by using part data also have good precision.5) Using online subspace method, the online diagnosis of engineering structure has been realized from ambient vibration data though the way of mode shape of curvature. Genetic algorithms are used to optimize the sensor placement though dualistic coding, when the number of sensors are restricted.