This paper deals with fault detection strategy used to ensure wind turbine reliability. Based on unknown iput nonlinear observer, the proposed approach have to estimate not only the full system state but also some actuator faults that can be considered as unknown inputs. Compared to some usually used algorithms, this method is caracterized by calculation time earn as well as development effort and accuracy which makes it useful for online implementation even for fast process. Used for linear systems, such approaches demonstrated interesting performances and results. The problem becomes harder for nonlinear systems where models are characterized by complex and coupled behaviors. More over, faults have to be detected as earlier as possible to avoid catastrophic and irreversible damages. In this work, fault detection algorithm based on unknown input high gain observer is proposed for a class of nonlinear systems site of actuator devations. Applied to a simulated wind turbine plant to reconstruct faults altering the electromechanical torque subpart, the results confirmed the accuracy and time convergence performances of the proposed observer which make it an intersting candidate to an online implementation.