Industrial Plant Causal Decision Support for Root Cause Analysis

The patent creates a simplified copy of the complex causal relationships in the industrial plant by building a causal graph model that replicates the essential cause-effect relationships. This model serves as a lightweight alternative to expensive high-fidelity process models, providing training and decision support functionality without requiring costly OTS systems. The causal graph captures the essential plant behavior in a computationally efficient manner.

The patent employs inexpensive computational models (causal graphs and probabilistic models) that can be rapidly deployed and updated without significant investment. These lightweight models provide sufficient functionality for operator support and training purposes, replacing the need for expensive, long-lived OTS infrastructure while achieving the core objective of improving operator capability.

The patent enables the plant operation system to provide its own decision support and diagnostic capabilities through automated causal inference. The system independently analyzes process data, identifies root causes of abnormalities, and suggests corrective actions without requiring experienced operator intervention. This self-service capability embeds expert knowledge into the control system itself, maintaining reliability while reducing dependency on expensive experienced personnel.

The patent introduces a causal inference system as an intermediary between the complex plant processes and the operators. This intermediary automatically processes process data, applies causal models to identify root causes, and presents actionable insights to operators. It serves as a knowledge bridge that compensates for operator inexperience, maintaining plant reliability without requiring experienced operators.

The patent extracts only the essential causal relationships from the complex plant system by building a causal graph that identifies key cause-effect connections. Rather than processing all available process data equally, the system extracts and focuses on the most relevant variables and relationships that drive plant behavior. This extraction approach maintains diagnostic accuracy by focusing on critical factors while reducing information processing complexity through selective attention to essential causal pathways.

The patent applies different processing strategies to different parts of the system based on their causal importance. The causal graph model identifies which process variables and relationships are most critical for diagnosis, allowing the system to apply more sophisticated analysis to key areas while using simpler monitoring for less critical parameters. This local quality approach optimizes diagnostic accuracy where it matters most while minimizing overall processing complexity.