Processamento de sinais de descargas parciais utilizando dicionários sobrecompletos e análise de componentes morfológicas

• Ano de defesa: 2019
• Autor(a) principal: André de Souza Oliveira Avelar
• Tipo de documento: Dissertação
• Tipo de acesso: Acesso aberto
• Idioma: por
• Instituição de defesa: Universidade Federal de Minas Gerais; UFMG
• Palavras-chave em Português: Atenuação de ruídos; Análise de componentes morfológicas; Representações esparsas; Dicionários sobrecompletos; Descargas parciais; Engenharia elétrica; Descargas elétricas; Processamento de sinais
• Link de acesso: http://hdl.handle.net/1843/RAOA-BCEHJ9

Electrical equipment insulation integrity is of great importance and a non-invasive test to verify this integrity is the Partial Discharge diagnosis (PD). This kind of diagnosis can be accomplished even with the equipment under operation. Undesired failures in the dielectric manufacturing process may result in small cavities in the material. When these cavities are exposed to an electric field, an ionization process occurs causing small localized electrical discharges, known as PD. On-site measured PD usually has high noise level from diverse sources, such as radio waves, transients of power electronic devices and the PD measurement system electronics itself. Noise is prejudicial to the PD processing system, since it may lead to an incorrect PD amplitude determination, a false detection or a failure to correctly detect a PD. This research work presents a new method of PD signal denoising based on Morphological Component Analysis (MCA) algorithm, which uses overcomplete dictionaries, sparse representations and signals prior information. The MCA is based on the assumption that the PD pulses and noise are morphologically different superimposed components and, as such, can be separated by analyzing their particular shapes. The method was tested on synthetic PD signals associated to amplitude modulated (AM), Gaussian and impulsive noise. The method was also tested on measured PD signal. The MCA method achieved high quality results on synthetic PD signals containing high impulsive and AM noise levels, even with amplitudes greater than the PD pulse amplitude. As for Gaussian noise, the MCA method achieved adequate results with low noise level, however did not achieve efficient performance with high noise level. Considering measured PD signals, once again the method presented high quality results for signals containing high impulsive and AM noise levels. The MCA method recovered the PD pulse waveform with a high cross-correlation, presented low pulse amplitude deviation and provided considerable improvement in signal-to-noise ratio for all the mentioned cases.