A Knowledge Graph And Graph Neural Network-Based Framework For Autonomous Fault Detection And Isolation In Large-Scale Networks

Abstract

Modern networks face unprecedented complexity and scale challenges that traditional fault detection and isolation methods struggle to address effectively. This article presents an innovative method that combines knowledge graphs with graph neural networks to create an autonomous fault detection and isolation framework for large-scale networks. By integrating the structural and semantic representation capabilities of knowledge graphs with the adaptive learning power of graph neural networks, the system enables context-aware anomaly detection, automated root-cause localization, and continuous learning in dynamic network environments. The framework ingests diverse network data to construct comprehensive knowledge graphs, applies sophisticated feature engineering techniques, and leverages message-passing neural architectures to identify fault patterns and propagation paths. The above is, to a large degree, proven by extensive testing in enterprise and telecommunications testbeds, showing large increases in detection accuracy, isolation performance, and system flexibility as compared to legacy methods. The approach is particularly well-suited to telecommunications, cloud computing, IoT, and enterprise IT practices, and has wider implications toward environmental sustainability, economic resiliency, and social service reliability. This transformational shift to infrastructure self-management deals with the increasingly daunting task of network reliability at scale.