风力发电机组发电机前轴承故障预警及辨识

Fault warning and identification of front bearing of wind turbine generator

为实现风电机组发电机前轴承故障预警及辨识,将监控和数据采集系统(SCADA)时间序列数据和状态监测系统振动数据相结合,提出了一种时频域建模方法。首先,利用SCADA数据建立基于门控循环单元神经网络的发电机前轴承温度模型,并计算其温度残差特征;其次,提取发电机前轴承振动信号时域特征和频域特征;最后,将温度残差特征和振动信号时频域特征相融合,建立基于极限梯度提升的前轴承故障辨识模型,从而辨识发电机前轴承正常、内圈损伤、外圈损伤、轴不平衡、滚动体损伤5类情况。实验研究表明,该方法比单独利用振动信号特征开展前轴承故障预警辨识的准确率高,其正常、内圈损伤、外圈损伤的平均辨识准确率从87%、58.5%、65%,分别提升到88.5%、67.5%和74%。

In order to realize the fault warning and identification of front bearing of wind turbine generator,in this paper a time-frequency domain modeling method is proposed,which integrates the time series data of supervisory control and data acquisition(SCADA)system with the vibration data of condition monitoring system(CMS).Firstly,the temperature model of generator front bearing based on gated recurrent unit(GRU)neural network is established using the SCADA data,and the temperature residual features are calculated.Secondly,the time domain features and frequency domain features of the vibration signal of generator front bearing are extracted.Finally,the temperature residual features and the time-frequency domain features of the vibration signal are fused,and the extreme gradient boosting(XGBoost)based fault identification model of the front bearing is established,which can identify five working conditions of the generator front bearing,including normal,inner ring damage,outer ring damage,shaft imbalance and rolling body damage.Extensive experiment results demonstrate that the proposed method can achieve higher identification accuracy compared with the front bearing fault warning identification method using the vibration signal characteristics alone.The average identification accuracy for normal,inner ring damage and outer ring damage conditions increase from 87%,58.5%and 65%to 88.5%,67.5%and 74%,respectively.