ABSTRACT
The striking duality between the time- and frequency-domain representations for a linear system means that there are a number of approaches to linear system identification based in the different domains. In fact, the duality extends naturally to nonlinear systems where the analogues of both the impulse response and transfer functions can be defined. A minor problem with discrete-time system identification for the structural dynamicist is that the model coefficients have no physical interpretation. However, although it is difficult to convert the parameters to masses, dampings and stiffnesses, it is relatively straightforward to obtain frequencies and damping ratios. In the system identification literature, it is usually said that an input signal must be persistently exciting if it is to be of use for system identification. It is usual in nonlinear system identification to fit a linear model first. This gives information about the degree of nonlinearity and also provides guidance on the appropriate values for the lags.
