Effective early warning of main bearing failures in wind turbines is a key issue to improve their reliability and availability. In order to find out wind turbine main bearing faults in time and avoid greater loss

an XGBoostKDE-based wind turbine main shaft bearing temperature prediction and fault warning method is proposed. The relevant characteristic parameters in supervisory control and data acquisition(SCADA) system are selected as input variables to predict the main bearing temperature under normal operating conditions of the wind turbine

and the residuals between the predicted value and the operating data during normal operation are obtained. Afterward

the non-parametric kernel density estimation(KDE) method is applied to determine the residual warning threshold

and the sliding window analysis method is combined to achieve the main bearing fault warning of the wind turbine.Moreover

a 2 MW-level wind turbine is used as the research object

and the operating data in the SCADA system is read for verification. The experimental results show that

this method can predict the main bearing temperature under normal operating conditions with accuracy of 97.6%

respond to the fluctuation of the temperature curve when the main bearing fails

and give an effective warning signal to the main bearing failure of the wind turbine nearly one month in advance.