Fault detection has been deployed in many cases. It will help improve the stability of the system. Data-driven method can provide credible evidence for fault detection. For time series which may include a lot of noise, the performance of typical method may be affected. This article raises an enhanced gate recurrent unit (GRU) method in order to analyze unmanned aerial vehicle (UAV) flight data which are affected by the vibration of motor or wind. Firstly, the raw data are denoised and normalized to improve the effect of analysis. Secondly, a gate recurrent unit (GRU) model is built to estimate one of the sensor data based on others. Finally, to detect fault data, the method based on residuals and threshold is applied. To evaluate the effectiveness of the method, the simulation data of UAV are applied to the method, and it can be clearly found that the proposed method is effective in fault detection.