Research Of Structural Parameter Identification Using Incomplete And Noise-Polluted Testing Information

Structural parameter identification problem is a kind of optimization inverse problem, which is summed up to an optimization problem based on matching of modal parameter. Many methods attempt to minimize a nonlinear error function between the analytical and measured modal data. But by the virtue of noise existing in observed experimental data, parameter identification problem is always ill-posed. Thus the existence, uniqueness, stability of the solution can not be guaranteed. Dozens of parameter identification methods have been presented before, but methods using incomplete and noise-polluted testing information are insufficient. According to the present conditions, this paper introduces methods of parameter identification using incomplete and noise-polluted tested information, including the following aspects:Firstly, this paper introduces the theory of structural dynamic sensitivity analysis, which can then be applied for the analysis of modal parameter sensitivity, parameter selection, and sensitivity-based parameter updating problems.Secondly, this paper introduces the effect of error function on parameter identification, and then presents the detectability index of the elements. By means of Monte-Carlo simulations, a conclusion is obtained that sensitivity analysis before the iteration is necessary, because it can exclude the elements that can not be identified easily. Also it is effective to suppress the noise influence on parameter identification by selecting a proper error function.Regularization method can effectively suppress the noise influence on parameter identification. According to the specified objective function, the gradient regularization method was introduced in this paper. By means of Monte-Carlo numerical simulations, a conclusion is got that the theory presented in this paper not only guarantees the iterations going on, suppresses the noise influence on parameter identification, but also improve the stability of the solution. The generalized cross-validation leads to a good selection of regularization parameters also. The regularization method depends somewhat upon the initial values for the iterative update, focusing on this, the Homotopy idea is brought into the parameter identification problem. By selecting the optimal Homotopy parameter, the method greatly improves the convergence territory of the regularization method.Lastly, the research is summarized and the future extensions of the relevant study are presented.