Knowledge-Assisted Multi-Graph Dependency Learning for Industrial Anomaly Detection

This method integrates process knowledge into Graph Neural Networks to improve multivariate time-series anomaly detection in multi-stage industrial environments. It addresses the limitations of standard GNNs that fail to capture critical, predefined relational structures between sensors and process stages.