AD-STLN: A Neural-Symbolic Framework with Dictionary Learning for Transparent Bearing Fault Classification
编号:60 访问权限:仅限参会人 更新:2025-11-10 11:31:24 浏览:21次 口头报告

报告开始:2025年11月23日 08:50(Asia/Shanghai)

报告时间:20min

所在会场:[S2] Parallel Session 2 [S2-2] Parallel Session 2-23 AM

暂无文件

摘要
This paper tackles the interpretability challenge in rolling bearing fault diagnosis by introducing the Adaptive Dictionary-based Sparse Temporal Logic Network (AD-STLN). Unlike traditional approaches using fixed wavelet kernels, AD-STLN employs data-driven dictionary learning to adaptively extract fault-specific features. The framework comprises three key modules: an adaptive dictionary convolution layer for customized feature extraction, a sparse attention transformer encoder for highlighting salient signal regions, and a temporal logic reasoning layer that constructs weighted Signal Temporal Logic (wSTL) expressions to explain diagnostic decisions. Experiments on the CWRU dataset show that AD-STLN delivers competitive accuracy while offering clear, logic-based interpretability, supporting more transparent and trustworthy fault diagnosis.
 
关键词
Adaptive-dictionary-based Sparse Temporal Logic Network (AD-STLN),signal temporal logic (STL),dictionary learning,interpretable fault diagnosis
报告人
Peixi Yang
Student South China University of Technology

稿件作者
Peixi Yang South China University of Technology
Gang Chen South China 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21日

    2025

    11月23日

    2025

  • 10月20日 2025

    初稿截稿日期

  • 11月23日 2025

    注册截止日期

主办单位
IEEE Instrumentation and Measurement Society
South China University of Technology
承办单位
South China University of Technology
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