Scholarlink

<p>Root cause analysis (RCA) systematically identifies the fundamental causes of faults and process deviations in industrial settings. In large chemical plants with interconnected control loops, RCA can be time-consuming due to the propagation of oscillatory and non-oscillatory faults, complicating manual detection from extensive operational data. Automated RCA algorithms are essential; however, existing techniques relying on plant topology or data-driven models often suffer from high computational costs, data unavailability, or limited accuracy. While some methods, like the spectral envelope approach, bypass causal maps, they only address specific fault types, lacking a comprehensive solution. This article introduces a method for ranking likely root cause variables in multivariate systems with interconnected control loops, without relying on topology or causal models. The approach utilizes cross-correlation with weighted lags among variables to identify the key variables responsible for anomalies in the system. The proposed algorithm is validated using synthetic data generated in MATLAB Simulink for various processes with interconnected control loops. While the proposed method does not provide formal causal guarantees, it is intended as a heuristic diagnostic ranking tool. The study’s results showed an overall accuracy of 94.41% for non-oscillatory faults and 88.65% for oscillatory faults for the case studies considered. Furthermore, the algorithm is also validated on synthetic data generated from two industrial case studies to exemplify the practical relevance of the approach.</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 cross-correlation-based framework for root cause detection in multivariable control networks

  • 期刊论文

  • 发表日期：

    2026年5月6日

查看全文
  • Abhishek Bansal,
  • Resmi Suresh,
  • Raghunathan Rengaswamy,
  • Prabirkumar Saha

摘要

Root cause analysis (RCA) systematically identifies the fundamental causes of faults and process deviations in industrial settings. In large chemical plants with interconnected control loops, RCA can be time-consuming due to the propagation of oscillatory and non-oscillatory faults, complicating manual detection from extensive operational data. Automated RCA algorithms are essential; however, existing techniques relying on plant topology or data-driven models often suffer from high computational costs, data unavailability, or limited accuracy. While some methods, like the spectral envelope approach, bypass causal maps, they only address specific fault types, lacking a comprehensive solution. This article introduces a method for ranking likely root cause variables in multivariate systems with interconnected control loops, without relying on topology or causal models. The approach utilizes cross-correlation with weighted lags among variables to identify the key variables responsible for anomalies in the system. The proposed algorithm is validated using synthetic data generated in MATLAB Simulink for various processes with interconnected control loops. While the proposed method does not provide formal causal guarantees, it is intended as a heuristic diagnostic ranking tool. The study’s results showed an overall accuracy of 94.41% for non-oscillatory faults and 88.65% for oscillatory faults for the case studies considered. Furthermore, the algorithm is also validated on synthetic data generated from two industrial case studies to exemplify the practical relevance of the approach.