The use of full-skip ultrasonic data and Bayesian inference for improved characterisation of crack-like defects

Abstract

Ultrasonic arrays are often used to detect and characterise crack-like defects in the field of non-destructive testing. The ultrasonic scattering matrix contains the far-field scattering coefficients of a defect for all measurable incident/scattering angles. This paper investigates the use of the scattering matrix for characterisation of small cracks in scenarios when the crack is steeply inclined, making direct imaging and analysis challenging. As well as the directly scattered signals, it is shown through experiments and simulations that additional characterisation information can be extracted from the full-skip ray path and used for improving the characterisation performance. Compared to the nearest neighbour approach, the mean errors of crack size and angle can be reduced by 12.1% and 17.1%, respectively, for 1.5 mm, 45° rough cracks by using a statistical modelling approach based on Bayesian inference.