Leveraging LLMs for Information Extraction in Manufacturing

In this paper, we present an approach using open-source Large Language Models (LLMs) for structuring free-text fields in the context of manufacturing. Our process involves three main steps: (1) creating prompts that extract information from free text fields, (2) converting the extracted information into a structured format accessible to computational analysis, and (3) evaluate the accuracy of the extraction by comparing the structured output with a predefined ground truth. We present the approach using a case from an actual manufacturer: We apply the process to texts from free text fields containing problems and solutions from quality control. Using LLMs, we extract quality problems and their solutions from the text, cluster the quality problems and identify common quality issues. Our findings demonstrate the potential of LLMs to automate knowledge extraction and the time-consuming manual pre-processing of text necessary for subsequent analytics and machine learning.