Study On Fault Diagnosis Method Of Wind Turbine Generator Bearing

As an important rotating component of a wind turbine,rolling bearings have an important influence on the normal operation of the transmission chain and even the unit.Therefore,it is of great practical significance to carry out fault detection and diagnosis of rolling bearings.The fault data has different characteristics at different stages of the wind farm operation.The fault data in the early stage of wind farm construction is incomplete,and it is impossible to model bearing fault types.When the wind farm is continuously operated for a long period of time,various fault types have accumulated data,which provides sufficient data support for model establishment,and more accurate fault location can be performed based on this data.This paper proposes corresponding bearing fault diagnosis methods based on deep learning based on whether the wind field fault data is complete or not.In the case of incomplete fault data at the scene,in order to solve the problem that the traditional Encoder-Decoder model cannot effectively use the signal characteristics,a MALEDbased anomaly detection model is proposed.The model has the ability to automatically extract local frequency features and time features from time-frequency signals,and at the same time can avoid the problem of signal dilution,which enhances the model's ability to extract features.The model was tested with actual data.Experiments show that the recognition rate of the method in this chapter for bearing abnormality detection is over 97%,which verifies the effectiveness of the method in this chapter.Under the condition that the fault data is complete,a single wavelet transform cannot fully express the signal characteristics and the standardization of the standard convolutional neural network has poor generalization ability.A fault diagnosis method based on Mw-1DConvLSTM is proposed.This method can enhance the representation of vibration data and improve the learning performance of the model.By studying various performance indexes of the model under different speeds and different types of bearings,it proves that this method is superior to other methods combining time-frequency transform and convolutional neural network.