On the Optimal Selection of Measured Data in Time Domain for Structural Identification

The results of structural identifications rely generally on the data measured in time domain. Reasonable results of structural identification can only be obtained on the basis of the measured data with sufficient information. This paper aimed to deal with the optimal selection of measured data for structural identification. The Fisher information matrix (FIM) of measured data in time domain was derived by the likelihood function that was obtained with the probability density function of these data. And the trace of FIM might be effective in selecting the most informative among the available measured data as the optimal one for structural identification. In order to implement the calculation of FIM, the formulae of calculating the partial derivatives of calculated data by using the corresponding mathematical model with respect to identification parameters, were developed. With the studied case of one multi-degree-of-freedom spring-mass system, the identification results, based on different simulated measured data while on the other identical conditions for calculations, were compared. The results of comparison showed that, among them, the one corresponding to the most informative data selected by using the proposed method was the most reasonable, which demonstrated the validity of the method.