Hybrid identification with time-series data and frequency response data for accurate estimation of linear characteristics

The purpose of this paper is to estimate the linear characteristics accurately by separating the nonlinear characteristics from the time-series data that measures the frequency response characteristics of the plant. In general, it has not been easy to separate linear and nonlinear characteristics because measurement data includes both of the characteristics. However, it is possible to separate by assuming a model of nonlinear characteristics, searching for parameters of the nonlinear model, and estimating transfer function from the Frequency Response Data (FRD) without the effect of the nonlinearity. We call this method hybrid identification of time-series data and frequency data since FRD is used to estimate the linear transfer function, and time-series data is used to estimate the parameters of nonlinear characteristics. Moreover, Bayesian optimization is used as an efficient search method of the parameters of the nonlinear model. The effectiveness of the proposed identification method is verified by ball screw and rolling friction.