A numerical database for ultrasonic defect characterisation using array data: Robustness and accuracy

Elastodynamic scattering matrices are known to contain geometrical information about a given scatterer, such as its size and shape. Here, the extent to which this scattered information can be retrieved using an ultrasonic array and used to characterise defects for Non-Destructive Evaluation is explored. Experimentally measured defect scattering matrices are compared to a database of possible scatterers and the nearest neighbour used to characterise the defect's geometry in terms of crack length and orientation. As an example, a database of scattering matrices for small (lengths 0.2–2.0 mm) cracks at a range of frequencies (2-20 MHz) is formed. The short range similarity (i.e. that between close neighbours) and the long range similarity (i.e. uniqueness) are used to understand the uncertainties inherent in this approach. In addition, the effect of spatially coherent noise, such as grain scattering in a polycrystalline metal, on the scattered information content is quantified. It is shown that as the noise level or frequency increases, so the information retrievable from a given crack is reduced, setting bounds on the accuracy of characterisation possible from a given ultrasonic dataset.