Scholarlink

Purpose <p>Confronted with the challenges that insufficient fusion of multi-channel signals leads to degraded fault diagnosis for rotating machinery, an innovative information fusion fault diagnosis is developed in this paper.</p> Methods <p>First, the symmetrized dot pattern (SDP), which transforms signals into snowflake-shaped patterns, is improved for multi-channel signal fusion and end-to-end fault diagnosis, named pixelated SDP (PSDP). The proposed PSDP can generate pixel-level representations of signals, and data-level fusion is also accomplished. Second, to extract features from PSDP images and raw vibration signals, a multi-level fusion network (MLFNet), which consists of a multi-scale attention fusion module (MSAFM) and a supplemental feature extraction module (SFEM), is designed. The implication of ‘multi-level’ can be summarized into two points: one is that multi-scale features of images are fused by MSAFM, and the other is that multi-modal features are jointly utilized for fault recognition.</p> Results <p>The proposed method achieves outstanding fault diagnosis based on the HUST dataset (Acc: 99.63% &amp; 94.40%) and the MCC5-THU dataset (Acc: 91.67% &amp; 95.23%), and the model also has satisfactory noise robustness. It also maintains a high testing speed of 24.22 ms/sample. Moreover, the effectiveness of multi-channel signal fusion and multi-modal fusion is validated through ablation study.</p> Conclusion <p>This paper contributes an advanced information fusion fault diagnosis framework for rotating machinery, and the proposed methodology has potential for real-world applications.</p>

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

A Novel Information Fusion Fault Diagnosis Framework for Rotating Machinery Based on Pixelated SDP and Multi-level Fusion Network

  • 期刊论文

  • 发表日期：

    2026年1月7日

查看全文
  • Ziyao Geng,
  • Shihua Zhou,
  • Zhaohui Ren,
  • Tianzhuang Yu,
  • Fangrui Dong,
  • Xingyu Wang

摘要

Purpose

Confronted with the challenges that insufficient fusion of multi-channel signals leads to degraded fault diagnosis for rotating machinery, an innovative information fusion fault diagnosis is developed in this paper.

Methods

First, the symmetrized dot pattern (SDP), which transforms signals into snowflake-shaped patterns, is improved for multi-channel signal fusion and end-to-end fault diagnosis, named pixelated SDP (PSDP). The proposed PSDP can generate pixel-level representations of signals, and data-level fusion is also accomplished. Second, to extract features from PSDP images and raw vibration signals, a multi-level fusion network (MLFNet), which consists of a multi-scale attention fusion module (MSAFM) and a supplemental feature extraction module (SFEM), is designed. The implication of ‘multi-level’ can be summarized into two points: one is that multi-scale features of images are fused by MSAFM, and the other is that multi-modal features are jointly utilized for fault recognition.

Results

The proposed method achieves outstanding fault diagnosis based on the HUST dataset (Acc: 99.63% & 94.40%) and the MCC5-THU dataset (Acc: 91.67% & 95.23%), and the model also has satisfactory noise robustness. It also maintains a high testing speed of 24.22 ms/sample. Moreover, the effectiveness of multi-channel signal fusion and multi-modal fusion is validated through ablation study.

Conclusion

This paper contributes an advanced information fusion fault diagnosis framework for rotating machinery, and the proposed methodology has potential for real-world applications.