Fault Diagnosis Method for Bearings Based on ECA-ResNet
Aimed at the problem that traditional fault diagnosis methods for bearings cannot extract the key features sufficiently under strong noise and variable load,a fault diagnosis method for bearings is proposed based on effective channel attention-residual network(ECA-ResNet).Firstly,the bearing vibration signals are converted into three-channel color images using short-time Fourier transform and pseudo-color processing method,which is used as input sample set.Secondly,the shallow edge features of data are extracted and compressed based on convolution and pooling operation,and the residual network block is introduced into efficient channel attention network(ECA-Net)to solve the problem of network degradation,establish the connection between channels,and extract the deep key features adaptively.Finally,Dropout is introduced to suppress the model overfitting,and the Softmax layer is used as classifier for fault diagnosis.The verification is conducted using bearing data sets from Case Western Reserve University and Jiangnan University.The results show that the average accuracy of the proposed method can reach 97.5%and 93.69%under strong noise and variable load respectively,having better noise resistance,generalization and versatility.
rolling bearingfault diagnosisFourier transformdeep learningresidual
NETL
NSTL
万方数据
