Abstract: In order to realize early warning of wind turbine gearbox faults, aiming at problems of insufficient feature extraction of existing models in time domain and frequency domain and poor adaptability of traditional threshold setting methods, a condition monitoring method of the wind turbine gearbox based on IACrossformer-ESPOT was proposed. Firstly, the SCADA data of wind turbine were preprocessed, and variables related to the oil temperature of the gearbox were selected as the model input through the maximum information coefficient (MIC). Then, with the feature extraction ability of interactive convolution (ICB) and adaptive spectrum (ASB), the complex pattern changes in the time-frequency domain of the gearbox oil temperature data were effectively captured based on the IACrossformer algorithm, and the normal oil temperature model of the gearbox was established. Finally, the ESPOT algorithm was used to adaptively set the threshold of the prediction residual to achieve efficient residual analysis and condition monitoring. Results show that the proposed method can accurately establish the normal oil temperature model of gearbox, and the threshold setting algorithm has good parameter adaptability, which can realize the early warning of gearbox fault.