대규모 언어 모델 기반 지능형 표준 물관리 시험검사 현장 데이터 자동화 시스템 개발

초록

This study proposes a data automation system for water management testing and inspection operations that integrates application records, field photographs, measurement data, checklists, and location information into a unified, accession-number-centered structure. To address missing data, duplicate entries, and repetitive manual processing, the system combines OCR- and LLM-based document understanding with deterministic rule-based post-processing, including data alignment, deduplication, date correction, identifier assignment, and format normalization. A local vLLM-based architecture was adopted for sensitive documents to enhance data control and reduce dependence on external APIs. A functional review confirmed support for accession-number-based data integration, OCR and LLM output inspection, automatic identifier assignment with manual override, CSV validation, graph-based analysis, and pre-transmission JSON verification. The results demonstrate the feasibility of integrating heterogeneous field data through combined generative AI and deterministic post-processing, offering a scalable and practical framework for automating field data management in water resource monitoring contexts.

키워드