Parameter Identification of Vibration Systems Using Sensitivity Analysis : Consideration of Improvement of Convergence

Abstract

Previous papers discussed a method for identifying model parameters of a vibration system by means of a non-linear least square minimization procedure of an evaluation function which consists of the difference of natural frequencies, mode shape coefficients of natural modes between the measured data and the data calculated from the estimated model parameters. The method using the previous evaluation function tends to necessitate a large number of iterations. This paper proposes a method using a new evaluation function for the purpose of decreasing the number of iterations and steadying convergences. It contains scalar values which evaluate the similarity of mode shape between the measured natural mode and the calculated one of each order, instead of each mode shape coefficient of natural modes. The effects of using the above-defined scalar values are discussed. Numerical models are examined to verify the efficiency of the method using a new evaluation function.