Graph Neural Network-Based Cross-Domain Fault Prediction for Smart Manufacturing Systems

Abstract

The Internet of Things in smart manufacturing and Industry 4.0 is rapidly developing, changing the way traditional factories operate. These changes have brought new issues related to proactive equipment upgrades and real-time fault monitoring. Traditional machine learning methods cannot effectively address issues related to domain differences or variations across different regions. It still cannot manage distributed domains. This paper proposes a new method based on Graph Neural Networks (GNN) for cross-domain fault prediction in smart manufacturing facilities. Based on the aforementioned findings, this study has developed a new architecture called RGN-FCN. The study proposes an innovative method that combines Graph Neural Network (GNN) technology with Fast RPN for target recognition. The empirical analysis of this study shows that the GNN-based method outperforms traditional methods such as CNN and LSTM; it is also applicable under non-traditional conditions beyond simple input distributions. Adding explicit structural connections and domain-independent representation learning improves cross-domain generalization ability; accurately identifying less common or complex faults. Introduce how this framework helps deploy the next generation of AI-based, scalable, and resilient smart factory operation and maintenance systems. The future research goal is to further reduce the impact of high domain differences on performance and deepen causal interpretability, so that it can be applied in commercial environments.