Diagnosisformer: An efficient rolling bearing fault diagnosis method based on improved Transformer

Abstract

Aiming at the problems of low accuracy and robustness of traditional deep learning fault diagnosis methods, a novel attention-based multi-feature parallel fusion model Diagnosisformer is proposed for rolling bearing fault diagnosis utilizing Transformer as the basic network. Firstly, frequency domain features of the original data are extracted by Fast Fourier Transform (FFT), and then normalization operations and embeddings are performed on the model input. Secondly, the designed multi-feature parallel fusion encoder is exploited to extract the local and global features of the bearing data. The extracted features are fed to a cross-flipped decoder, followed by a classification head for fault classification. Finally, experimental verification is performed using data collected by the rotating machinery fault diagnosis experimental platform and the Case Western Reserve University (CWRU) bearing dataset. The average experimental results on the two fault diagnosis datasets are 99.84% and 99.85%, respectively. The results show that our diagnosis method significantly outperforms the state-of-the-art in accuracy, generalization, and robustness.