Combined Signal and Model-Based Sensor Fault Diagnosis for a Doubly Fed Induction Generator

The problem of multiplicative and/or additive fault detection and isolation (FDI) in the current sensors of a doubly fed induction generator (DFIG) is considered in the presence of model uncertainty. A residual generator based on the DFIG model is proposed using the structure of the classical generalized observer scheme. However, each observer in this scheme is replaced by a robust H_/H ∞ fault detection filter followed by a Kalman-like observer. The latter further attenuates the effect of the modeling uncertainties on the residuals. It exploits the specific pattern induced by the balanced three-phase nature of all the electric signals. It turns out that the FDI problem then amounts to detecting an abrupt change in the mean of the residual vector in the additive fault case, or the appearance of sine waves superimposed on a white noise vector in the multiplicative fault case. A decision algorithm made of a combination of generalized likelihood ratio algorithms allows us to detect and isolate the additive and multiplicative sensor faults. The complete FDI system is tested through simulations on a controlled DFIG.