Defect detection in the presence of geometrical artefacts

Abstract

Defect detection in non-destructive testing relies on the ability to detect signals arising from defects and distinguish them from those arising from noise. Ultrasonic arrays are capable of producing multiple views of a component through combinations of mode conversions and reflections of waves from boundaries, which potentially contain additional information about defects. Noise in these views can come from random sources such as electrical noise and thermal fluctuation, and coherent sources including stochastic microstructural noise, and boundary reflections giving rise to geometrical artefacts. The number of geometrical artefacts increases with both the geometrical complexity of the part under test, and with the number of reflections used to construct a view. This paper tackles the challenge of defect detection in the presence of geometrical artefacts by characterising the intensity distribution at each point in every view produced within a pristine sample, and using a hypothesis test as a means of defect detection. This information is combined with state-of-the-art data fusion routines, improving detection performance. Experimental work shows that this method improves performance compared to other methods of defect detection over the same scan area, which assume that the defect does not occupy the same location as an artefact in at least one view. Provided the pristine sample is characterised sufficiently well, hypothesis testing provides a robust method of defect detection in the presence of geometrical artefacts.