An Improved Fault Detection Method based on Canonical Variate Analysis for Tricky Faults of Wind Turbine

Wind power is prospecting as a renewable and clean energy. However, wind energy is usually unstable due to the influence of region, season and weather. Considering the safety, stability and economy of the generation process, it is essential to develop an efficient condition monitoring and fault detection system for wind turbines. There are some tricky faults in wind turbines which are not so far from the normal conditions that are hard to detect using traditional methods. To solve this problem, this paper proposes an improved fault detection method which combines partial least-squares regression and canonical variate analysis, mapping the canonical variables to another feature space using partial least-squares regression. The divergence of canonical variables under tricky failure is expanded in the new space, thus the properties of fault detection are effectively improved. Compared with the other three traditional methods, the simulation on the wind turbine benchmark and data sets on blade icing in real wind turbines verifies that the accuracy and universality of the proposed method have been significantly improved and are better than canonical variate analysis.