Application of Augmented Observer for Fault Diagnosis in Rotor Systems

Abstract

Observers are widely used to generate residuals for fault diagnosis processes. However the generated residuals are not only influenced by faults but also by disturbances and model uncertainties. In order to avoid false alarm and achieve a high fault detection rate disturbances and model uncertainties have to be considered in the fault detection and isolation (FDI) processes. In rotor systems the predominant disturbances and model uncertainties are forces excited by unbalances and gyroscopic effect which result in a rotary frequency dependent system behavior. The effect of unbalance forces and gyroscopic effect appear in a sinusoidal form with rotor rotary frequency. The disturbances and model uncertainties can be generally represented using unknown inputs and the signals of unknown inputs are sinusoidal. To take advantage of this characteristic of rotor systems augmented observers that account for sinusoidal unknown inputs are designed in this work. The augmented observer for sinusoidal signals can be used to estimate the distribution matrix of unknown inputs or to generate the residual for fault detection. In case that the faults (e.g. rotor disc wear) acting on the rotor are also sinusoidal, the augmented observer can be applied for fault isolation and identification. Different configurations of the observers are introduced in this paper according to their applications.