Inteligencia artificial em lean manufacturing: aplicação de processamento de linguagem natural no controle de Andon

Abstract

This study applies Artificial Intelligence techniques, through Natural Language Processing (NLP), to the control of the Andon system, in alignment with Lean Manufacturing principles. The research was conducted in a company located in the Industrial Hub of Manaus (PIM), where the control of Andon activations is performed manually using Exel spreadsheets, which limits the analysis and interpretation of data, especially records that do not result in critical stoppages. The objective is to transform historical Andon records into organized and accessible information through the application of NLP models, combined with a Dashboard developed in Power BI to support decision-making. The methodology is characterized as a case study with a quantitative approach, using 955 records, of which 645 were considered valid after data preprocessing. The NLP model, based on the GPT-4 architecture, was configured with historical records using the Retrieval-Augmented Generation (RAG) technique, allowing for the automatic interpretation of textual information such as symptoms, causes, and actions, overcoming limitations related to textual ambiguity and lack of standardization. In parallel, a Power BI Dashboard was developed for quantitative analysis, enabling the identification of failure patterns, critical workstations, and accumulated downtime. A prototype of a virtual assistant was also created in Streamlit, connected to the OpenAI API, capable of responding to natural language queries based on the processed data. The results demonstrate that the integration of data analysis and NLP significantly contributes to improving process control, enhancing information organization, reducing operational failures, and supporting decisionmaking in industrial management.