Wind turbine fault diagnosis based on unscented Kalman Filter

A wind turbine is a nonlinear system with multi-variables that can be measured subject to considerable noise. To realize intelligent fault detection for wind turbines as part of condition monitoring, an unscented Kalman Filter (UKF) approach, widely used for signal tracking of nonlinear systems, is adopted in this paper. Its capability for reliable fault detection is assessed. A model has been developed to represent temperature variation of generator stator winding of a healthy wind turbine under operational conditions. This model is developed based on heat generation principle within the stator winding and the thermal dynamics of heat loss. This model is further incorporated into an Unscented Kalman Filter (UKF) for generator temperature prediction and thus fault detection. Two failure modes are introduced into the thermal model, which results are compared with predicted results of the UKF model. Effectiveness of the UKF for wind turbine fault detection is demonstrated.