Characterisation of defects from ultrasonic array data
A Velichko
University of Bristol, UK
The use of ultrasonic arrays in non-destructive evaluation brings many benefits. It has recently become possible to capture the full matrix of raw array data from all transmit-receive element combinations and then perform other operations in post-processing. The advantage of this method of array data acquisition is that transmit-receive datasets contain the maximum possible amount of information that can be measured by an array from a given position. In this case various imaging techniques can be applied, which produce high-quality images with increased defect detection sensitivity. However, in some practical applications, defect detection alone is not enough and an understanding of the nature of defects is required for the assessment of structural integrity of a system. In this presentation, the fundamental questions associated with the defect characterisation challenge are addressed. In particular, what information about scattering behaviour of a defect can be reliably extracted from an array data and how this information can be efficiently used to achieve quantitative defect characterisation.
A new way of looking at this problem, termed ‘parametric-manifold mapping’, involves constructing a surface in principal component space that represents all possible defects of a given type. The characterisation challenge then becomes one of finding the closest approach of measured data to this surface. The shape of this surface reveals fundamental insights into the nature of the defect characterisation information, indicating, for example, which defects are easy/hard to characterise. Crucially, the method also enables probability maps of defect characterisation to be plotted. This leads to the potential to understand what measurement (for example pulse-echo or array) is best suited to a particular characterisation problem. All findings are supported by numerical simulations and experimental measurements.
A new way of looking at this problem, termed ‘parametric-manifold mapping’, involves constructing a surface in principal component space that represents all possible defects of a given type. The characterisation challenge then becomes one of finding the closest approach of measured data to this surface. The shape of this surface reveals fundamental insights into the nature of the defect characterisation information, indicating, for example, which defects are easy/hard to characterise. Crucially, the method also enables probability maps of defect characterisation to be plotted. This leads to the potential to understand what measurement (for example pulse-echo or array) is best suited to a particular characterisation problem. All findings are supported by numerical simulations and experimental measurements.
