A Comparative Study of Causality Detection Methods in Root Cause Diagnosis: From Industrial Processes to Brain...

Area of Science:

• Engineering
• Neuroscience
• Data Science

Background:

• Causal knowledge extraction from process measurements is crucial for fault root cause analysis (RCA).
• Existing causality detection methods often have idiosyncratic implementations, limiting accessibility.
• A unified comparison framework is needed for diverse research communities.

Purpose of the Study:

• To provide a comprehensive comparison of data-based causality detection methods for root cause diagnosis.
• To evaluate 11 distinct methods across two complex domains: industrial processes and human brain networks.
• To offer insights and a taxonomy for selecting appropriate causality detection techniques.

Main Methods:

• A unified evaluation framework was designed to compare 11 causality detection methods.
• Methods were implemented in a standard way to infer causal interactions from multi-variable signals.
• Experiments were conducted on plant-wide oscillations in an industrial process and epileptogenic focus localization in brain networks.

Main Results:

• The study presents a cross-domain investigation comparing the performance of 11 causality detection methods.
• Findings offer practical insights into the application and effectiveness of different techniques.
• An interpretative pitfall common to causality detection methods is identified and discussed.

Conclusions:

• Data-based causality detection methods can be effectively applied to complex systems like industrial plants and brain networks.
• The comprehensive comparison provides valuable guidance for researchers and engineers in RCA.
• Awareness of potential interpretative pitfalls is essential for accurate causal inference.