In order to keep away wind turbines from catastrophic conditions due to sudden breakdowns, it is important to detect faults as soon as possible. For diagnosis, a model-based approach is chosen. There are many works that use this fault detection design, but the majority of them consider this system as a linear time invariant (LTI) model. The objective of this paper is, first, to find an LPV model of the system using the subspace identification technique of linear parameter-varying (LPV). Second, we focus on fault diagnosis based on residual generation which is obtained as a comparison between the measured variable and the estimated one using this LPV model. In this work, a benchmark of a wind turbine case is proposed with six predefined faults (three sensor fault scenarios, two actuator fault scenarios and a system fault scenario).

Peer Reviewed