Análise de agrupamento de sinais obtidos por simulação do ensaio ultrassônico pela técnica TOFD para reconhecimento de defeitos de soldagem modelados

Abstract

The non-destructive testing by ultrasound is widely used to check for discontinuities inside the material. The Time Of Flight Diffraction is an ultrasonic technique widely used and combined with the phased array detects far smaller discontinuities in the material. However, the classification of the type of discontinuity from the ultrasonic signals obtained during an inspection requires a highly skilled professional, and a long time if considering the volume of data to be analyzed. This task may be performed by classification and clustering algorithms able to quickly process a large amount of data. In this work was simulated discontinuities in software COMSOL three types of welding defects (lack of penetration, pore and crack); then it was applied two pattern recognition methods: Principal component analysis and the k-means, to check how the data is separated. This work is a first step for the numerical simulation of use of ultrasonic testing to produce simulated signals and use them in pattern recognition tools training, to be later tested with experimental ultrasonic signals.