Fault Root Cause Analysis Based on Liang-Kleeman Information Flow and Graphical Lasso

Area of Science:

• Industrial Systems Engineering
• Data Science
• Fault Diagnosis

Background:

• Root cause analysis is crucial for system fault diagnosis, identifying fault locations and causes.
• Traditional causal analysis methods like transfer entropy can yield biased results, leading to inaccurate fault identification.
• Existing methods struggle with the high dimensionality and redundancy of industrial data.

Purpose of the Study:

• To develop a more accurate and reliable root cause analysis method for industrial systems.
• To address the limitations of transfer entropy in causal inference for fault diagnosis.
• To leverage information flow for superior fault detection and propagation analysis.

Main Methods:

• A novel root cause analysis method combining graphical lasso (Glasso) for dimensionality reduction and Liang-Kleeman information flow (LKIF) for causal inference.
• Application of Glasso to handle large-scale, high-dimensional industrial data by reducing redundancy.
• Utilizing LKIF to calculate information flow intensity and infer causal relationships between variables for fault tracing.

Main Results:

• The LKIF-Glasso method effectively identifies fault root causes and visualizes fault propagation on the Tennessee Eastman simulation platform.
• Comparative experiments demonstrate that information flow (LKIF) outperforms transfer entropy in root cause analysis accuracy.
• Detailed analysis of stripper step failure explains the superiority of information flow over transfer entropy.

Conclusions:

• The proposed LKIF-Glasso method offers a significant advancement in industrial fault diagnosis accuracy and reliability.
• Information flow is a more robust approach for causal inference in complex industrial systems compared to transfer entropy.
• This method provides a powerful tool for understanding and mitigating faults in industrial processes.