Research on improving the bearing fault diagnosis method of MSCNN-ECA
Convolutional neural networks(CNNs),as powerful feature extraction tools,can effectively extract fault-bearing data from complex environments,thus improving recognition accuracy.This paper proposes a convolutional neural network-based bearing fault diagnosis method.First,rolling bearing data is sampled,and two-dimensional image data is generated through continuous wavelet transformation.Next,a Multiscale Convolutional Neural Network(MSCNN)is employed to extract features from the input data.Efficient convolution modules with residual structures maximize the retention of valid feature information,followed by feature selection using Channel Attention Modules(Efficient Channel Attention,ECA).Finally,after feature mapping via fully connected layers,the model predicts fault categories.Experimental validation is conducted by employing the dataset from Case Western Reserve University and the results generated from CNN-LSTM,ResNet,LeNet,and other models are compared.The proposed method consumes less time and achieves the highest diagnostic accuracy.Under single-load conditions,it achieves 100%diagnostic accuracy,while under multi-load conditions,it reaches an accuracy as high as 99.46%,surpassing other advanced algorithms.Additionally,the bearing data from Jiangnan University is employed for generalization validation,showing impressive transfer effects.
国家科技期刊平台
NSTL
万方数据
