Partial Cross Mapping Based on Sparse Variable Selection for Direct Fault Root Cause Diagnosis for Industrial...

Area of Science:

• Process engineering and industrial automation
• Data-driven fault diagnosis and root cause analysis
• Complex systems modeling and causality inference

Background:

• Effective root cause diagnosis is crucial for safe and efficient process industry operations.
• Conventional methods like contribution plots suffer from smearing effects, hindering accurate diagnosis.
• Existing methods like Granger causality (GC) and transfer entropy struggle with indirect causality in complex industrial settings.

Purpose of the Study:

• To propose a novel regularization and partial cross mapping (PCM)-based framework for root cause diagnosis.
• To enable efficient direct causality inference and fault propagation path tracing in industrial processes.
• To overcome the limitations of traditional methods in handling complex industrial process faults.

Main Methods:

• Generalized Lasso-based variable selection for identifying candidate root cause variables.
• Formulation of Hotelling T² statistic and Lasso-based fault reconstruction.
• Application of partial cross mapping (PCM) for root cause diagnosis and fault propagation path tracing.

Main Results:

• The proposed framework successfully infers direct causality and traces fault propagation paths.
• Validation across four diverse cases: numerical example, Tennessee Eastman benchmark, wastewater treatment, and steel decarburization processes.
• Demonstrated rationality and effectiveness in identifying root causes and propagation pathways.

Conclusions:

• The regularization and PCM-based framework offers a robust solution for root cause diagnosis in process industries.
• This approach enhances the ability to ensure safe production and improve operational efficiency.
• The method effectively addresses challenges posed by smearing effects and indirect causality.