Aplicação de Sensor Virtual baseado em Aprendizado de Máquina para Aprimoramento da Eficiência Global em Manufatura Digital

• Ano de defesa: 2023
• Autor(a) principal: Souza, Bruno Vilela de [UNIFESP]
• Orientador(a): Santos, Sérgio Ronaldo Barros dos [UNIFESP]
• Tipo de documento: Dissertação
• Tipo de acesso: Acesso aberto
• dARK ID: ark:/48912/001300001xt78
• Idioma: por
• Instituição de defesa: Universidade Federal de São Paulo
• Palavras-chave em Português: Aprendizado de Máquina; Sensor Virtual; Random Forest; Extreme Gradient Boosting; Métodos Ensembles; OEE; Manufatura Discreta
• Link de acesso: https://repositorio.unifesp.br/handle/11600/69541

This work addresses the use of machine learning algorithms to propose a learning architecture approach, that allows the creation of a virtual sensor model capable of detecting production failures using operational data from machines in a manufacturing process. The specific algorithms used are Ensembles models such as the Random Forest, Gradient Boosting, and Extreme Gradient Boosting. The modeling of a virtual sensor is relevant as after its implementation it can prevent equipment from being stopped due to issues with physical sensors, directly impacting the Overall Equipment Effectiveness (OEE), particularly through equipment availability for manufacturing. The primary contribution of this work is the proposal of an architecture grounded in machine learning approaches for a package machine in a discrete manufacturing line that enables the enhancement of production quality by predicting when a product will be defective and improving equipment availability with the aim of maximizing production efficiency. By accurately detecting manufacturing faults, especially those related to the positioning of packaging for producing target products, the proposed algorithms make a significant contribution to improving OEE (Overall Equipment Effectiveness). This approach can represent a significant advancement for the industry, providing valuable tools for monitoring and controlling manufacturing processes with the goal of enhancing product quality and operational efficiency.