Monitoramento Estrutural por Impedância Eletromecânica via Sistemas Neuro-Fuzzy

• Ano de defesa: 2024
• Autor(a) principal: Prudente, Fellipe André Diniz
• Tipo de documento: Dissertação
• Tipo de acesso: Acesso aberto
• Idioma: por
• Instituição de defesa: Universidade Federal de Uberlândia; Brasil; Programa de Pós-graduação em Matemática
• Palavras-chave em Português: ANFIS; HyFIS; Conjuntos Fuzzy; SBRF; SHM; Impedância; CNPQ::CIENCIAS EXATAS E DA TERRA::MATEMATICA::MATEMATICA APLICADA; ODS::ODS 4. Educação de qualidade - Assegurar a educação inclusiva, e equitativa e de qualidade, e promover oportunidades de aprendizagem ao longo da vida para todos.
• Link de acesso: https://repositorio.ufu.br/handle/123456789/41317; http://doi.org/10.14393/ufu.di.2024.24

The aim of Structural Health Monitoring (SHM) is to examine a structure for the identification, localization, and detection of potential damages. The focus of this study is on SHM in oil exploration and production companies. The data used comes from experiments funded by Petrobras. The objective is to account for temperature variations in data gathered from electromechanical impedance by employing two distinct neuro-fuzzy systems: the Hybrid Neural Fuzzy Inference System (HyFIS) and the Adaptive Neuro-fuzzy Inference System (ANFIS). Both systems generate Fuzzy Rule-Based Systems (FRBS), employing Mamdani and Takagi-Sugeno inference methods, respectively. The experiments took place in three different contexts. The initial experiment involved a piezoelectric PZT (PbLead Zirconate Titanate) sensor affixed to an aluminum plate under controlled temperature within a climatic chamber at the Mechanical Structures Laboratory (LMEst) of the Faculty of Mechanical Engineering. Subsequently, in the second scenario, data were collected using PZT patches mounted on a steel plate, subjecting it to five different damages in ambient conditions at the Glória Campus. For the third experiment, nine PZTs were linked to the circular section of a fuel tank prototype roof in an outdoor setting at the Santa Mônica Campus. In all experiments conducted at the Federal University of Uberlândia (UFU) sites, impedances linked to specific temperatures and frequencies were collected. A segment of this data was used to train the neuro-fuzzy networks constructing the FRBS, with temperature and frequency as input variables and impedance as the output variable. A comparative analysis was carried out by computing the Coefficient of Correlation Deviation (CCD) between the results obtained from the HyFIS and ANFIS-generated FRBS and the experimental data. The comparison results were satisfactory, showing HyFIS superiority with over 90\% accuracy in all experiments. In the second experiment, it was feasible to measure impedance differences between reference values and the five damages using ten distinct metrics derived from the FRBS obtained by HyFIS. This decision stemmed from the precision of validation values generated by HyFIS, surpassing those produced by ANFIS. The implementation of neuro-fuzzy networks in SHM holds potential for optimizing production processes across various engineering domains, particularly in the petroleum industry, which is the primary focus of this study.