| REN Yi,HAN Jiaxin,LUAN Fangjun,YUAN Shuai.A BiGRU-CCT Based Hybrid Imaging Method for Bearing Fault Diagnosis[J].航空发动机,2026,52(1):38-47 |
| A BiGRU-CCT Based Hybrid Imaging Method for Bearing Fault Diagnosis |
| DOI：10.12482/ISSN.1672-3147.20240914001 |
| Key Words:bearing fault diagnosis bidirectional gated recurrent unit compact convolutional transformer hybrid imaging methods |
| Author Name | Affiliation | | REN Yi | School of Computer Science and Engineering,Shenyang Jianzhu University,Shenyang 110168,China | | HAN Jiaxin | School of Computer Science and Engineering,Shenyang Jianzhu University,Shenyang 110168,China | | LUAN Fangjun | School of Computer Science and Engineering,Shenyang Jianzhu University,Shenyang 110168,China | | YUAN Shuai | School of Computer Science and Engineering,Shenyang Jianzhu University,Shenyang 110168,China |
|
| Hits: 1593 |
| Download times: 659 |
| Abstract:To solve the problem of insufficient utilization of fault information in bearing fault diagnosis,A hybrid imaging method
based on bidirectional gated recurrent unit(BiGRU)and compact convolutional transformer(CCT)for bearing fault diagnosis was proposed.
The bidirectional structure of BiGRU was used to capture the implicit time correlation in bearing vibration signals.A three-channel image
was synthesized by the Gramian angular field(GAF),Markov transition field(MTF),and recurrence plot(RP)to obtain the multi-dimensional
characteristics of the signal.The convolution blocks of the CCT model were employed to extract local features and reduce model parameters.
The features fused by BiGRU and the convolution blocks were fed into the Transformer and sequence pooling module for fault diagnosis,
enabling more comprehensive monitoring of the bearing operating conditions.To better explain the proposed bearing fault diagnosis method,
the confusion matrix and t-distributed Stochastic Neighbor Embedding(t-SNE)were used to visualize the results.The results demonstrate
that,compared to the 2D CNN and Vision Transformer(ViT)utilizing hybrid imaging methods,the fault diagnosis method incorporating
BiGRU and CCT is found to be capable of fusing various spatial features,thereby facilitating the extraction of temporal information from
fault signals,resulting in an improvement in both F 1 score and efficiency. |
| View Full Text View/Add Comment Download reader |
|
|
|
