An LLM-Based Decision-Support System for Mitigating Document-Induced Human Error Risks in Flight Test Occupational Safety
Multi-version iteration and cross-aircraft configuration differences in flight test restriction documents pose significant challenges to crew situational awareness. This paper presents an intelligent parsing and multi-dimensional comparison system based on large language models. An anchor-driven hierarchical partitioning strategy converts unstructured Word documents into a structured knowledge base covering all aircraft and all versions. A fault-tolerant comparison engine following a rule pre-screening, LLM fine-grained judgment and rule fallback pipeline extracts semantic differences across adjacent versions, cross-aircraft configurations and full historical traceability. A dual-layer importance assessment mechanism that integrates domain rules with LLM reasoning produces task-oriented and explainable priority rankings. Experiments on 29 authentic documents from five test aircraft show that the parsing accuracy reaches 97.8%, the F1 scores of three comparison tasks all exceed 89%, the module coverage of generated PDF reports reaches 98.3% and structural compliance reaches 100%. The system reduces crew document review time from hours to minutes. From an occupational health and safety perspective, fragmented and ambiguous safety-critical documentation represents an information hazard that may increase the probability of human error during flight-test preparation and execution. By transforming multi-version restriction documents into an auditable, importance-ranked, and task-oriented report, the proposed system acts as a digital risk-control measure that reduces crews' exposure to information discontinuity, highlights safety-relevant changes, and supports more consistent knowledge transfer during personnel rotation. These findings suggest that LLM-assisted document intelligence can contribute not only to document-processing efficiency, but also to the prevention of document-induced human-error risks in high-risk aviation workplaces.