Detecção de outliers usando data stream com contextualização de falhas orientada por ontologia na indústria 4.0

• Ano de defesa: 2022
• Autor(a) principal: Miodutzki, Dionei
• Tipo de documento: Dissertação
• Tipo de acesso: Acesso aberto
• Idioma: eng
• Instituição de defesa: Universidade Tecnológica Federal do Paraná; Curitiba; Brasil; Programa de Pós-Graduação em Computação Aplicada; UTFPR
• Palavras-chave em Português: Localização de falhas (Engenharia); Big data; Processamento eletrônico de dados - Processamento distribuído; Ontologia; Inferência (Lógica); Fault location (Engineering); Electronic data processing - Distributed processing; Ontology; Inference; CNPQ::CIENCIAS EXATAS E DA TERRA::CIENCIA DA COMPUTACAO; Ciência da Computação
• Link de acesso: http://repositorio.utfpr.edu.br/jspui/handle/1/29691

Outlier detection is important in several sectors of the economy, the academy and the government. In the industrial sector, these techniques make it possible to quickly and accurately identify equipment failures, product defects and safety risks. The evolution of Industry 4.0, however, is bringing challenges previously uncommon in the area. The large number of data constantly generated by a multitude of sensors represents a processing challenge and can ultimately lead to the identification of a large number of outliers simultaneously. The scale and complexity of this scenario slow the troubleshooting process, delaying the identification of the source of the fault and increasing costs and downtime. This work presents a solution that tackles the problem in two fronts: (i) distributed processing of data streams for outlier detectiong; and (ii) ontology-based contextualization of the detected outliers. Our proposal supports decision-making in a widespread failure scenario, where there are multiple outliers detected in a set of equipment with known dependencies between them. Dependencies are represented using ontologies, as a way to provide a clear and user-facilitated interpretation. An inference engine implemented as a graph database is responsible for identifying the most probable causes of the failure. Performance tests demonstrate the scalability of our implementation.