A Statistical Approach to Ultrasonic Inspection for Complex Geometry

Abstract

Non-destructuve evaluation (NDE) is concerned with the inspection of a component for flaws, which can lead to a degradation in performance. Ultrasonic arrays have become a widely used tool in NDE as they may produce a visualisation of the scattering behaviour in the volume of a component through techniques such as full matrix capture and the total focusing method (TFM). By exploiting mode conversions as the wave skips from boundaries in the geometry, the TFM may be extended to provide additional views that can aid in detection of flaws. This has the benefit of increasing the coverage to evaluate areas from which the array does not have line of sight, or alternatively providing additional angles of incidence on a flaw which may aid in characterisation. Flaw detection is performed by distinguishing between anomalous signals which may indicate the presence of a defect, and benign signals such as boundary reflections or backscattering from the material. While the latter do not indicate that a defect is present, they may trigger a false positive indication, or otherwise act to obscure or mask anomalous signals. The quantity of geometrical artefacts will increase with the geometrical complexity and the number of skips used to construct a view, and backscattering from a material microstructure often results in an ultrasonic response of similar magnitude to that of a defect, therefore both have a significant impact for multiview TFM imaging.

First, this thesis examines the conventional approach to designing an inspection for a region which does not have line of sight with the array. The expected response from a candidate defect is simulated to identify the quality of the response from different views produced using the TFM, and identify the impact of coherent noise on detection ability. Arising from the difficulty experienced by this approach in predicting the location of imaging artefacts, a detection technique based on the stochastic properties of experimentally measured data is developed and applied to the non–line of sight geometry, enabling position-wise detection of anomalous signals in the image. This technique is then developed further to examine the detection ability in other sources of coherent noise. Materials with a high level of backscattering are examined, including carbon fibre–reinforced polymers and a rubber-sand mixture, in which defects have some finite spatial extent. Finally, an outlook on classification of signals using this statistical approach is provided.

Date of Award	18 Mar 2025
Original language	English
Awarding Institution	University of Bristol
Supervisor	Anthony J Croxford (Supervisor) & Paul D Wilcox (Supervisor)