Research On Bearing Fault Diagnosis Of Wind Turbine Based On Convolutional Neural Network

Wind power industry is in rapid development,and the condition monitoring and fault diagnosis of wind turbines is of great significance for reducing maintenance costs,reducing losses,and improving system reliability.Bearings are one of the most frequently faulted components in wind turbines.Therefore,accurate and effective bearing fault diagnosis methods help ensure the safe and stable operation of wind turbines.Traditional diagnosis methods often rely on artificial experience and expert knowledge for fault feature extraction,which increases the complexity and difficulty of fault diagnosis and is not conducive to the realization of automatic fault detection and intelligent operation and maintenance of wind turbines.Therefore,in order to realize fault feature learning and fault classification simultaneously,this paper studies a bearing fault diagnosis method based on convolutional neural network.This paper proposes a flowchart of intelligent fault diagnosis for bearings based on vibration signal detection,signal transformation,and convolutional neural network recognition.A deep convolutional neural network with three pairs of convolution-pooling layer pairs and two fully connected layers is designed.Firstly,the validity of the proposed method is verified by simulation data.Then,the case of comparison of three signal analysis methods,including vibration gray-scale image,short-time Fourier transform and continuous wavelet transform,is studied under the same model structure using the public bearing dataset from Case Western Reserve University.At the same time,the diagnostic results of the method based on the deep convolutional neural network and the method based on the time-domain statistical features and shallow machine learning models on the same data set are compared,which proves the superiority of the proposed method in this paper.In addition,weak fault diagnosis under noise interference and compound fault diagnosis are also studied.A novel strategy based on improved cost function and training sample interference is proposed to reduce the accuracy decline under low signal-to-noise ratio conditions.For the problem of compound faults diagnosis,single fault signals are utilized to construct pseudo-compound fault samples.Combined with a multi-label classification strategy,the model has a better ability to diagnose compound faults with only single fault samples for training.