Abstract: We propose a fault diagnosis method based on the observer-based multi-innovation stochastic gradient algorithm in consideration of the pitch system faults of wind turbines. The proposed algorithm can improve the parameter estimation accuracy by extending the innovation length. With regards the observer canonical state space systems model, the multi-innovation stochastic gradient algorithm combined with the state observer is able to achieve the interactive estimation between the system states and the parameters. Here, the pitch system model is further transformed into the identification model by converting it into a canonical state space model. On the basis of the pitch system faults leading to the change of system parameters, the algorithm is adopted to estimate the systemstate and parameters. Then, the pitch system fault diagnosis problem is transformed into the parameter estimation issue. The simulation results demonstrate that the proposed method is capable of effectively diagnosing the pitch system faults.