Abstract: For the sensor fault reconstruction of a wind turbine drive train system that includes unknown inputs and disturbances that cannot be decoupled, we propose a new unknown input observer (UIO) for fault reconstruction. Unlike the general UIO, this method solves the partial decoupling problem experienced by a wind turbine as a result of disturbances. First, we build the drive train sensor fault model of a wind turbine and then convert this system model to an augmented system. We divide the unknown input into two parts:the unknown inputs that can be decoupled are decoupled by the disturbance decoupling principle, and the disturbances that cannot be decoupled are attenuated by using LMI optimization. By using the Lyapunov stability theorem, we prove the stability of an unknown input observer and the ultimate boundedness of dynamic fault errors, and then achieve the estimation of system states and fault reconstruction, although some disturbances cannot be decoupled completely. Simulations are performed to demonstrate the effectiveness of the proposed algorithm.