A new procedure for parameter estimation by exciting limit cycles of specified periods

This paper describes a new procedure for identification of plant parameters using limit cycles of specified period when output measurements are heavily corrupted by noise. A simple and elegant scheme is presented in which the plant output is not applied directly to the input of the relay to excite the limit cycles. Instead, the noise-corrupted error signal of the feedback control system is sampled, filtered through a special cyclic filter and a DC eliminator and then applied to the input terminal of the relay. The net result is a stable oscillation of definite period which is synchronous with respect to the sampling pulse train. Thus the effect of noise in the output measurements is easily eliminated. The cyclic filtering scheme enables simple on-line identification procedures such as the minimum norm and LSQ methods to be applied to yield highly accurate results. The dynamics of the cyclic filter introduced does not affect the parameters to be estimated in any manner. To accelerate the convergence of the parameters, the input to the system is also cyclically averaged. © 1997 by John Wiley & Sons, Ltd.