Fault diagnosis is the best discipline to control the operation and maintenance costs of the wind turbine system. However, the fault diagnosis of wind turbine finds difficulties with the variation of wind speed and electrical energy (generator torque).

In this work, the proposed fault diagnosis approach is based on the Feature set algorithm, manifold learning and the Support Vector Machine classifier. First, the construction of the feature set is very important step, with the high dimension after application the MAED (Manifold Adaptive Experimental Design) algorithm on the data set. Moreover, the NPE(Neighborhood Preserving Embedding)manifold learning algorithm is applied for dimensionally reduction of feature set by the eigenvectors; it is easy to use as the input for the last step. Finally, the low dimension of eigenvectors is exploited by the Support Vector Machine classifier for recognition fault and making the maintenance decision.

This approach is implanted on the faults of the benchmark wind turbine and gives the best performance.