Research on bearing fault diagnosis based on multi-scale convolu-tional neural networks

A multi-scale CNN-SENet bearing fault diagnosis model based on Gramian Angular Field is pro-posed to address the problems of traditional fault features based on manual experience and incomplete feature extraction of ordinary convolutional neural networks. The bearing vibration signal data is transformed into 2D image data using Gramian Angular Field and then fed into the multi-scale CNN-SENet bearing fault diagnosis model, which is used to classify and extract features. The stockpile dataset was utilized by researchers at Case Western Reserve University to assess the efficacy of the model. Their experiments revealed that the transformation technique GADF out-performed others, resulting in the model achieving superior accuracy in diagnosing faults.