Desenvolvimento de sistema especialista Smart System para automatizar a construção de planos de manutenção: aplicação no setor hidrelétrico

• Autor(a) principal: Toni, Vinicius Martins Dias
• Data de Publicação: 2024
• Tipo de documento: Trabalho de conclusão de curso
• Idioma: por
• Título da fonte: Repositório Institucional da UTFPR (da Universidade Tecnológica Federal do Paraná (RIUT))
• Texto Completo: http://repositorio.utfpr.edu.br/jspui/handle/1/35314
• topic: Inteligencia Artificial; Algoritmo; Engenharia mecânica; Artificial intelligence; Algorithm; Mechanical engineering; CNPQ::ENGENHARIAS

This study investigates the creation of an algorithm for the design of an expert system to support maintenance in power generation units. Incorporating the premises of Industry 4.0, the study proposes an integrated approach that combines advanced techniques of Natural Language Processing (NLP) and Artificial Intelligence (AI) to analyze Service Order (SO) texts. The goal is to develop a proactive and intelligent maintenance plan, optimizing processes and contributing to the unit's predictive maintenance. Using data obtained from the Failure Mode, Effects, and Criticality Analysis (FMECA) tool, the developed algorithm seeks to establish correlations between the SOs and identify patterns of failures and critical components mentioned. The research employs a hybrid methodology, merging quantitative analysis through metrics and numerical data with qualitative analysis, aiming to extract valuable information from the textual narratives found in 112 maintenance records. For effective text processing, the study applies tokenization techniques, isolating each word in the SO records for a more granular analysis. This step is followed by filtering out stop words and punctuation, elements that do not contribute significantly to the analysis. Subsequently, the residual words are lemmatized, consolidating verbs into their infinitive form and nouns and adjectives into the masculine singular. The processing of textual data precedes the application of AI methods, specifically the TF-IDF technique and cosine similarity analysis, to determine the semantic proximity between terms. The algorithm implementation, carried out in Python with data storage via SQL, culminates in an efficient and cost-effective maintenance plan. The results indicate the need for maintenance actions focused primarily on failure modes identified as "Burnout", "Breakage", and "Poor contact", and on critical components such as the 'Heat Exchanger', ‘Oil Water Detector’ and 'Water Piping'.