Utilização de aprendizado de máquina na categorização de dados de manutenção em equipamentos eletrônicos

• Ano de defesa: 2024
• Autor(a) principal: Alves, Clayton da Silva
• Tipo de documento: Dissertação
• Tipo de acesso: Acesso aberto
• Idioma: por
• Instituição de defesa: Universidade Federal de Santa Maria; Brasil; Administração Pública; UFSM; Programa de Pós-Graduação em Gestão de Organizações Públicas; Centro de Ciências Sociais e Humanas
• Palavras-chave em Português: Aprendizado de máquina; Processamento de linguagem natural; Framework; Manutenção de equipamentos; Instituições federais de ensino superior; Machine learning; Natural language processing; Equipment maintenance; Federal higher education institutions; CNPQ::CIENCIAS SOCIAIS APLICADAS::ADMINISTRACAO::ADMINISTRACAO PUBLICA
• Link de acesso: http://repositorio.ufsm.br/handle/1/33496

The research aimed to develop a framework utilizing machine learning (ML) and natural language processing (NLP) techniques to categorize maintenance data for electronic equipment in federal higher education institutions. This initiative seeks to optimize maintenance management and enhance operational efficiency. The study adopted a qualitative, descriptive, and documentary research approach, analyzing a database from the LAMI-UFSM department, which included 21,177 maintenance records from 2003 to 2023. Among the key findings, ML techniques were employed to automate the categorization of textual observations related to maintenance services. This automation facilitated the identification of various defective components, led to a gradual reduction in labor demand, and provided insights into the predominance of specific equipment types, such as desktops and printers, which were identified as the primary items requiring maintenance. The study also uncovered seasonal patterns in maintenance needs. Additionally, a prescriptive analysis of the categorized data, particularly concerning computers, was conducted. This categorization enabled a detailed examination of the maintenance performed, identification of critical components using the ABC method, and recommendations for maintaining safety stock to optimize resources and time. Ultimately, the study proposes a maintenance data categorization framework based on ML and NLP techniques, specifically tailored for federal higher education institutions to enhance maintenance management and improve operational efficiency.