A Bearing Fault Diagnosis Method Based on 2D Images and Transformer

Bearings are an important part of rotating machinery., so the accurate diagnosis of bearing faults is essential. Therefore., this paper proposes a bearing fault diagnosis method that combines two-dimensional (2D) images and the Swin Transformer model. The Swin Transformer model takes the attention mechanism as its core., which can efficiently capture global information of images. First., we use three different methods to convert bearing vibration signals into 2D images., namely gramian angular field (GAF)., short-time Fourier transform (STFT)., and wavelet transform (WT). Secondly., the 2D images are input into the Swin transformer model for feature extraction and fault classification through transfer learning. Finally., the bearing dataset of Case Western Reserve University (CWRU) is used for experiments. The results show that the combination of three different types of 2D images with the Swin Transformer model achieved excellent fault diagnosis results., which is superior to many traditional deep learning methods.