报告详情
  1. 活动首页
  2. 报告清单
  3. 报告详情
An Uncertainty-Aware Mamba Framework for Open-Set Fault Diagnosis of Rotating Machinery
编号：99 访问权限：仅限参会人 更新：2026-09-27 08:59:02 浏览：4次 张贴报告

报告开始：暂无开始时间（Asia/Shanghai）

报告时间：暂无持续时间

所在会场：[暂无会议] [暂无会议段]

暂无文件

摘要
Fault diagnosis of rotating machinery is a critical task in equipment health monitoring. Although deep learning has achieved remarkable success in intelligent fault diagnosis, most existing methods rely on Softmax-based deterministic predictions and are unable to quantify the epistemic uncertainty of model outputs. Consequently, they tend to produce overconfident predictions when encountering noisy samples, class-boundary samples, or data outside the training distribution, thereby reducing diagnostic reliability. To address this issue, this paper proposes EviMamba, a rotating machinery fault diagnosis framework that integrates the Mamba state space model with evidential deep learning. The proposed method combines efficient temporal feature modeling with evidential learning to jointly perform fault classification and uncertainty estimation. Furthermore, an Uncertainty Margin Constraint (UMC) is introduced to encourage low uncertainty for reliable samples while suppressing overconfident predictions on boundary and hard samples, thereby learning a more discriminative uncertainty distribution. Experimental results demonstrate that the proposed method can effectively identify unknown faults without using any out-of-distribution (OOD) samples for threshold calibration. The adaptive uncertainty estimation strategy establishes reliable open-set decision boundaries by distinguishing in-distribution (ID) and OOD samples, improving the detection capability of unseen faults while maintaining stable recognition performance for known fault categories. 
 
关键词
Evidential deep learning; Mamba; Fault diagnosis; State space model; Epistemic uncertainty.
报告人
Qiuying Zhao
Student Shenyang University of Technology

稿件作者
Yue Sun Shenyang University of Technology
Qiuying Zhao Shenyang University of Technology
发表评论
验证码 看不清楚，更换一张
全部评论
重要日期

  • 会议日期

    11月06日

    2026

    至

    11月08日

    2026

  • 10月15日 2026

    初稿截稿日期

主办单位
IEEE Instrumentation and Measurement Society
承办单位
Sichuan University
联系方式
登录查看完整联系方式
历届会议
移动端
在手机上打开
小程序
打开微信小程序
客服
扫码或点此咨询