Diagnosis of Device Exception Based on Causality of Device Indicators

Intelligent Exception diagnosis has achieved significant success in extracting essential information from sensor data. However, these methods suffers from unstable feature extraction performance and difficulties in detecting anomalies in high-dimensional spaces. To address these challenges, in this work, we propose an anomaly detection and fault diagnosis framework that integrates causal reasoning with graph neural networks (CGNN). Specifically, our model constructs causal graphs of monitoring variables using causal reasoning algorithms and employs graph neural networks to extract information from these graphs. Experimental validation on the extended-TE process dataset demonstrates that ourCGNN framework outperforms baseline algorithms, achieving faster fault identification and higher prediction accuracy. Importantly, our approach can accurately diagnose faults even with limited prior causal knowledge among variables, making it highly adaptable to various industrial scenarios. This framework not only offers practical value for engineering applications but also provides significant insights for improving the interpretability and reliability of anomaly detection systems in complex industrial environments.