Abstract: A time2varying autoregressive model with time-varying coefficients is introduced in this paper to extractmodal parameters from non-stationar y vibration signals. With this model, the relationship between time-varyingmodal par ameters and time-varying autoregr essive model coefficients is established. The time-varying autoregressivemodeling is employed in order to obtain the high resolution of t he time-frequency distribution of non-stationary responses. A new algor ithm is presented by combining the use of base functions with t he time2fr equency transform ofsignals, in which the base funct ions have shift the time-varying modeling to a linear time-invariant identificationproblem. A simulated linear time2var ying system of three degr ees of freedom with time-varying stiffness is presentedto validate the proposed time2varying autoregr essive method.

Key words: time-varying system, modal parameter identification, time-frequency transform, autoregressive model