Large language model-driven intelligent architecture for aerospace fault management

To address the issue of requirements for efficient processing and precise decision-making of multi-source heterogeneous fault data during implementing modern aerospace missions， a large-language-model-based aerospace equipment fault scheme is proposed， which follows agent architecture to build a four-layer closed-loop system of "data-knowledge-decision-application" that integrates core functions such as fault process mining， historical case matching， process node response and plan report generation. In terms of technical realization， a two-stage mining algorithm based on process intermediate representation is used to extract the structured disposal process， historical cases are associated through combining BM25 with semantic vector hybrid matching algorithms， and knowledge graph embedding technology is relied to achieve semantic alignment of faults and process nodes， and the automation driven fault disposal plans is ultimately generated. This architecture is verified in a leakage current scenario within the control system of a launch vehicle. The intelligence and standardization level of fault handling is significantly improved and presented technical support can serve for rapid response to aerospace equipment fault plans.