Fault Diagnosis Method of Rolling Bearing Based on Improved One-Dimensional Convolutional Neural Network
[Purposes]To address the issues of overfitting and weak generalization ability when using tra-ditional one-dimensional convolutional neural network models for bearing fault diagnosis,this paper pro-poses an improved one-dimensional convolutional neural network(1DCNN)method for rolling bearing fault diagnosis.[Methods]Firstly,the proposed method utilizes a global average pooling layer to replace the fully connected layer of traditional one-dimensional convolutional neural networks,reducing the number of parameters in the model,decreasing model complexity,and enhancing the generalization abil-ity of the convolutional neural network.Secondly,by combining the Dropout regularization method,the issue of overfitting in the model is addressed.Finally,the classification is performed by the Softmax clas-sification function.[Findings]Using the Case Western Reserve University Bearing Fault Dataset for validation,the results show that the improved 1DCNN can achieve a high accuracy rate and good fitting effect with relatively fewer training iterations during fault diagnosis,with a fault accuracy rate of 99.42%.[Conclusions]This method significantly outperforms the fault diagnosis results presented by traditional one-dimensional convolutional neural networks,and holds important theoretical significance and appli-cation value for solving practical bearing fault problems.
rolling bearings1D convolutional neural networksfault diagnosis
NETL
NSTL
万方数据
