Design of a class of reduced order unknown inputs nonlinear observer for fault diagnosis

The problem of fault detection in nonlinear systems is considered. Our goal is to extend the unknown inputs observer approach to nonlinear systems. The approach consists of two stages. At the first stage, a change of coordinates is found that decoupled the nonlinear system into two subsystems: one is independent of the unknown inputs and the other with the states is expressed as linear combinations of the outputs and the states of the first subsystem. At the second stage, a nonlinear observer is designed for the former. The use of such observer for robust fault detection purposes is then discussed, and finally illustrated by a numerical example.