Shopfloor Terminology for Retrieval-Augmented Generation (RAG): Aligning Operator Language with Engineering Knowledge

Abstract

As industrial work becomes increasingly digitalized, integrating human expertise into intelligent systems is essential for reliability and adaptability. This study investigates how curated terminology can improve Large Language Model-based Retrieval-Augmented Generation (RAG) systems for industrial knowledge management. It addresses a key linguistic issue that operators often use colloquial or locally coined terms that differ from standardized terminology found in technical documentation. This can lead to retrieval failures and inconsistent responses.A domain-specific dataset comprising 35 operator questions derived from a wire harness manufacturing manual is used to compare two types of RAG queries: natural-language operator queries and terminology-enhanced queries expanded with curated synonyms. Human evaluators assessed the correctness of generated answers. Terminology-enhanced queries achieved on average 67% correct answers compared to only 11% for nonterminology-enhanced ones.These results demonstrate the importance of terminology alignment for the reliable and effective use of LLMs in industrial contexts. Curated terminology bridges the gap between operator language and formal documentation, supporting tacit knowledge externalizationand improving retrieval reliability. This preliminary study highlights the feasibility and practical relevance of integrating terminology into RAG pipelines and outlines future directions towards adaptive, human-centered knowledge systems in manufacturing.

Keywords: RAG, Terminology, Query Expansion, Human-AI interaction, Tacit Knowledge, Operator language

DOI: 10.54941/ahfe1006998