重庆理工大学学报2025，Vol.39Issue(9)：179-184,6.DOI:10.3969/j.issn.1674-8425(z).2025.05.022
双路Transformer在轴承故障诊断中的应用
Application of dual-path Transformer in bearing fault diagnosis
摘要
Abstract
As it is unable to make full use of the signal characteristics,traditional algorithm has certain limitations in the bearing fault diagnosis task.To address the issue,a Dual-Path Transformer method is proposed to diagnose and classify bearing faults.The self-attention mechanism of the Transformer ensures the deep extraction of global temporal correlation information from long sequential frequency data.The Dual-Path Transformer employs convolution kernels of different sizes and attention mechanisms with different characteristics on two paths to extract the high and low-frequency features of the signal.Thus,it effectively identifies high and low-frequency features that represent bearing faults from the multiple spectra of the signal sequence,increasing the richness of feature information.Additionally,a multi-scale feature fusion module is designed to fuse the high and low-frequency features containing global correlation information extracted by the Dual-Path Transformer,obtaining deep fault features to achieve efficient diagnosis of different fault types.Experimental validation on the bearing dataset of the Society for Machinery Failure Prevention Technology demonstrates the Dual-Path Transformer achieves an accuracy of 97.44%under certain convergence rates.It surpasses traditional diagnostic algorithms in both accuracy and robustness.关键词
轴承故障诊断/双路 Transformer/多尺度特征融合/MFPT数据集/自注意力机制Key words
bearing fault diagnosis/Dual-Path Transformer/multi-scale feature fusion/MFPT dataset/self-attention mechanism分类
机械工程引用本文复制引用
邰志艳,侯婷悦,刘铭,于子奇,冯子懿..双路Transformer在轴承故障诊断中的应用[J].重庆理工大学学报,2025,39(9):179-184,6.基金项目
国家自然科学基金项目(61503150) （61503150）
吉林省发改委省级产业创新专项资金项目(2017C033-4) （2017C033-4）
