Aplicação de redes neurais artificiais para classificar padrões de ruído eletromagnético conduzido nas linhas de alimentação de circuitos integrados

• Ano de defesa: 2022
• Autor(a) principal: Borba, Douglas
• Orientador(a): Vargas, Fabian Luis
• Tipo de documento: Dissertação
• Tipo de acesso: Acesso aberto
• Idioma: por
• Instituição de defesa: Pontifícia Universidade Católica do Rio Grande do Sul
• Programa de Pós-Graduação: Programa de Pós-Graduação em Engenharia Elétrica
• Departamento: Escola Politécnica
• País: Brasil
• Palavras-chave em Português: Ruído Conduzido; Interferência Eletromagnética; Rede Neural Artificial; Sistema Embarcado Robusto; Anaconda Navigator; Tensor Flow
• Palavras-chave em Inglês: Conducted Noise; Electromagnetic Interference; Machine Learning; Deep Neural Network; Robust Embedded System; Anaconda Navigator; Tensor Flow
• Área do conhecimento CNPq: ENGENHARIAS
• Link de acesso: https://tede2.pucrs.br/tede2/handle/tede/10166

With the increasing use of embedded systems in our daily lives and the increasing level of electromagnetic noise in the environment in which these systems are exposed, the need for reliable operation is paramount. When the vital components of these systems are exposed to a large amount of noise, which can be both conducted and radiated, these noises can lead to serious and unavoidable failures, so it is essential to analyze and reliably test these components. In this scenario, this work presents a study on the use of machine learning techniques (artificial neural networks) to carry out the identification and classification in the field of different types of electromagnetic noise conducted in the power lines of integrated circuits (ICs), according to with a specific set of IEC standards. This work details the development of a methodology in which the waveforms of the phenomena contained in the standards are simulated through software due to the time available for a master's work plus the pandemic bias and the difficulty of accessing the laboratories due to this scenario, and also by the risk involving the time and complexity to obtain these waveforms in hardware using a microprocessor, for example. Experimental results obtained in this methodology suggest that the proposed approach is very effective to achieve the objective of identifying the types of electromagnetic noise conducted. In this way, the methodology developed allows the insertion of artificial intelligence in the context of tests, allowing the developers of systems and integrated circuits a new approach to assess the susceptibility to conducted EMI, enabling the development of new techniques to increase the reliability and robustness of the projects.