MTAGCN: Multi-Task Graph-Guided Convolutional Network with Attention Mechanism for Intelligent Fault Diagnosis of Rotating Machinery
Deep learning (DL)-based methods have shown great success in multi-category fault diagnosis due to their hierarchical networks and automatic feature extraction. However, their superior performance is mostly based on single-task learning, which makes them unsuitable for increasingly sophisticated eng...
Saved in:
| Main Authors: | , |
|---|---|
| Format: | Article |
| Language: | English |
| Published: |
MDPI AG
2025-04-01
|
| Series: | Machines |
| Subjects: | |
| Online Access: | https://www.mdpi.com/2075-1702/13/5/347 |
| Tags: |
Add Tag
No Tags, Be the first to tag this record!
|
| Summary: | Deep learning (DL)-based methods have shown great success in multi-category fault diagnosis due to their hierarchical networks and automatic feature extraction. However, their superior performance is mostly based on single-task learning, which makes them unsuitable for increasingly sophisticated engineering environments. In this paper, a novel multi-task graph-guided convolutional network with an attention mechanism for intelligent fault diagnosis, named MTAGCN, is proposed. Most existing fault diagnosis models are commonly bounded by a single diagnosis objective, especially when handling multiple tasks jointly. To address this limitation, a new multi-task fault diagnosis framework is designed, incorporating an attention mechanism between the task-specific module and task-shared modules. This framework enables multiple related tasks to be learned jointly while improving diagnostic and identification performance. Moreover, it is observed that most existing DL-based methods share incomplete fault representations, leading to unsatisfactory fault diagnosis. To overcome this issue, a graph convolutional network (GCN)-based fault diagnosis framework is introduced, which not only captures structural characteristics but also enhances diagnostic effectiveness. Comprehensive experiments based on three case studies demonstrate that the proposed MTAGCN outperforms state-of-the-art (SOTA) methods, striking a good balance between accuracy and multi-task learning. |
|---|---|
| ISSN: | 2075-1702 |