Reconhecimento de padrões de defeitos de soldagem utilizando classificadores treinados com sinais ultrassônicos simulados numericamente

• Autor(a) principal: Murta, Raphaella Hermont Fonseca
• Data de Publicação: 2018
• Tipo de documento: Dissertação
• Idioma: por
• Título da fonte: Repositório Institucional da Universidade Federal do Ceará (UFC)
• Texto Completo: http://www.repositorio.ufc.br/handle/riufc/30042
• topic: Ciência dos materiais; Ultrassom; Métodos de simulação; Soldagem; Redes neurais; Artificial neural network; Welding defects; Simulation; Ultrasound

Non-destructive evaluation based on ultrasound propagation is widely used to detect and to size up discontinuities. Time of Flight Diffraction (TOFD), an ultrasonic technique, has been increasingly used in welding joints inspections due to quick inspection and reliability. However, the classification of the kind of discontinuity from the ultrasound signals acquired during an inspection requires a high skilled professional. This task can be done by using pattern recognition algorithms, which are able to quickly process a great amount of data. In this present work, three types of discontinuities usually found in welding joint (lack of penetration, porosity and crack) were embedded in a bidimensional modelated media. Following, the finite volume method (FVM) was used to simulating wave propagation in the modelated media, mimicking the ultrasonic testing. Simulated ultrasonic signals were pre processed and submitted to pattern recognition algorithms (K-Nearest Neighbors, Artificial neural networks e K-means). This work aims to evaluate the use of simulated signals during the training stage of pattern recognition tools which will be used to classify signals acquired during welded joints inspections.