Abstract:  Aiming at the problems of severe wear of the pitch bearing of wind turbines and large economic losses, a modeling method based on data acquisition and monitoring and control (SCADA) data for wind turbine pitch bearing wear warning was proposed. New feature variables were constructed using sliding window statistics based on the operating parameters of the wind turbine such as pitch motor current, pitch angle, wind speed and power. Then the feature variables and label data were imported into the random forest algorithm for model training and verification. Finally, an early warning model for monitoring the wear of the pitch bearing was established. With the historical data of 20 wind turbines of a wind farm were taken as the input, and the unit fault record was taken as the output mark, a reasonable early warning rule was set and an early warning model was established and verified. The test results show that the pitch bearing wear warning model can issue early warning information 5 days to 30 days in advance, and the accuracy rate can reach 87.9% on average. Through trial operation, it is found that the model has the characteristics of low cost, high efficiency and strong interpretability, which is of great significance for improving the safe operation time of the unit and reducing the operation and maintenance cost of the unit.

Key words:  early warning model; pitch bearing; bearing wear; fault early warning; random forest

GUO Peng-fei, LIU Wei-jiang, ZHU Peng-cheng, et al. Early warning method of wind turbine pitch bearing wear［J］.Journal of Mechanical & Electrical Engineering, 2021,38(8):1045-1050.