ScholarWorks@경상국립대학교

초록

This study proposes a prompt engineering–based framework for proactive quality management, focusing on unstructured defect data from aircraft fuselage manufacturing. Unlike conventional Non-Conformance Report (NCR) systems that mainly emphasize post-defect corrective actions, a Large Language Model (LLM) was applied to structure defect types, causes, and corrective actions, and to generate preventive scenarios for recurring defects. Using 453 NCR cases from Company A (2023–2025), four prompt engineering techniques—Zero-shot, Few-shot, Role-based, and Output Formatting Control—were evaluated against four criteria: defect summarization accuracy, consistency of cause interpretation, specificity of corrective actions, and feasibility of preventive scenarios. The analysis showed that Output Formatting Control provided the highest performance in structured automation, while Role-based prompting aligned best with practitioner perspectives. Academically, this study validates the effectiveness of LLM-driven approaches for structuring unstructured quality data. Practically, it offers a framework to enhance defect analysis efficiency and support preventive quality management.

키워드