35 / 2025-03-28 10:30:09

Spacecraft Fault Inference based on Multi-source and Multi-level Knowledge Graphs

spacecraft,fault diagnosis,knowledge grap,deep learning

This paper classifies spacecraft fault knowledge graphs into two categories based on fault knowledge sources and system structure levels: data-driven and knowledge-driven fault knowledge graphs. Distinct fault inference methods are proposed for each type. For data-driven fault knowledge graphs, a multilayer perceptron model is employed to capture semantic information from measuring points and on-orbit data, enabling spacecraft health monitoring and fault location identification. For knowledge-driven fault knowledge graphs, the Graph Attention Deep Deterministic Policy Gradient (GADDPG) is utilized to achieve precise fault path inference. These methods are integrated within a unified fault knowledge graph of the overall control system, constructing a comprehensive fault diagnosis system that combines knowledge and data. Experimental validations using on-orbit data and actual fault cases demonstrate the system's capability to achieve accurate fault localization and hierarchical fault diagnosis.