Extreme learning machine-based non-linear observer for fault detection and isolation of wind turbine
Ayoub El Bakri, Miloud Koumir, Ismail Boumhidi · Australian Journal of Electrical & Electronics Engineering · 2019
This paper presents a robust fault detection and isolation (FDI) scheme for a variable speed wind turbine. The proposed scheme (extreme learning machine–state-dependent differential Riccati equation (ELM-SDDRE)) is an observer model-based approach, especially, a non-linear observer using SDDRE based on an improved model of the wind turbine by using the ELM. The standard SDDRE can be used for small model uncertainties. However, when the uncertainties are large, the SDDRE cannot detect and isolate the faults. The main objective of the ELM is the prediction of unknown nominal model dynamics to construct a new improved nominal model used by the observer for FDI. This makes the effect of uncertainties weak and consequently allows better faults detection. The faults considered in this paper are sensor faults in the rotating speeds of the rotor and generator outputs. The effectiveness of the proposed approach is illustrated through simulation.