Design of excitations for structure discrimination of nonlinear systems, using the best linear approximation

Abstract

Block-oriented nonlinear models are very popular, since they are composed of well-understood elements, namely linear dynamic parts and static nonlinearities. However, the identification problem of most types of block-oriented nonlinear models can be very difficult (especially the generation of initial estimates) and time-consuming. Therefore it is of crucial importance to select the most appropriate model structure directly from the measurement data, before proceeding with the actual nonlinear identification. In this work, we explore methods to detect the internal structure of the system (in particular, the presence of nonlinear feedback), based on the best linear approximation paradigm. Two different strategies (varying DC (mean value) and standard deviation (STD) of the excitation signal) are compared. The methods are applied on two real systems with a static nonlinear block in the feedback path.