“…We can reconstruct the output (ẑ(t)) of the linear dynamic subsystem using the inverse of the nonlinear static function (f −1 (•), identiÿed by the previous relay test). Then, we can estimate a complex linear model from the reconstructed output (ẑ(t)) and the given process input (u(t)) using well-established linear system identiÿcation approaches such as methods using transform (Cheng and Hsu, 1982;Chou et al, 1999;Sagara and Zhao, 1989;Eitelberg, 1988;Johansson et al, 1999;Sung et al, 1998) subspace method , prediction error method and numerous techniques to determine the model structure for linear systems such as by considering physical insight, examining the spectral analysis estimate, testing ranks in covariance matrices, examining the information matrix, minimizing Akaike's information theoretic criterion (AIC) and checking whiteness of residuals (Ljung, 1987).…”