Scholarlink

This paper proposes the Joint Low-Rank and Sparse Double Dictionary Learning (LSDDL) for industrial process monitoring. The LSDDL utilizes two dictionaries, a low-rank dictionary and a sparse dictionary, to simultaneously capture global structural features and local features in multivariate time-series data from complex industrial environments. The low-rank constraint limits the learnable dictionary atoms in a low-dimensional subspace, enhancing steady-state features. The sparse constraint with \({l}_{2,1}\) -norm focuses on key local features, improving transient fault detection. An online monitoring strategy is proposed, which utilizes the T2 and SPE statistics for fault detection and estimates the control limits through kernel density estimation. Experimental results on the Tennessee Eastman Process show that LSDDL outperforms traditional and single-dictionary methods in Fault Detection Rate (FDR), demonstrating its effectiveness in complex industrial process monitoring.

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

Joint Low-Rank and Sparse Double Dictionary Learning for Industrial Process Monitoring

  • 会议论文

  • 发表日期：

    2026年

查看全文
  • Yan Pan,
  • Jipeng Guo,
  • Haopeng Yang,
  • Yuchen Wang,
  • Wenxuan Chen,
  • Youqing Wang

摘要

This paper proposes the Joint Low-Rank and Sparse Double Dictionary Learning (LSDDL) for industrial process monitoring. The LSDDL utilizes two dictionaries, a low-rank dictionary and a sparse dictionary, to simultaneously capture global structural features and local features in multivariate time-series data from complex industrial environments. The low-rank constraint limits the learnable dictionary atoms in a low-dimensional subspace, enhancing steady-state features. The sparse constraint with \({l}_{2,1}\) -norm focuses on key local features, improving transient fault detection. An online monitoring strategy is proposed, which utilizes the T2 and SPE statistics for fault detection and estimates the control limits through kernel density estimation. Experimental results on the Tennessee Eastman Process show that LSDDL outperforms traditional and single-dictionary methods in Fault Detection Rate (FDR), demonstrating its effectiveness in complex industrial process monitoring.