The influence of surface determination in x-ray computed tomography

Abstract

X-ray computed tomography (XCT) is a non-destructive technique that enables the dimensional inspection of the internal and external features of modern components. All dimensional attributes, including form, size and surface texture, are derived from the surface of the component, therefore, surface determination (SD) is a critical step in X-ray computed tomography (XCT) for dimensional metrology. Threshold-based and gradient-based algorithms are widely used. However, these algorithms are often sensitive to systematic errors and can be largely influenced by the input parameters selected by operators. Current research gaps include the limited knowledge related to the quantifiable impact of SD algorithms on the traceability of the XCT geometric measurements and to the SD algorithm robustness to threshold setting. It was postulated here that Marker-controlled watershed (MCW), successfully applied in the medical field, can overcome the potential threshold errors encountered with current SD algorithms and provide robust results even in the presence of typical XCT errors. As a result, this thesis presents the development of a robust SD algorithm for XCT based on the MCW method, which combined weighted gradient voxel refinement and automated marker generation. The validation and comparison of MCW were accomplished by employing calibrated workpiece for experimental XCT scanning, and simulation. The work has further evaluated the performance of MCW algorithm and two commonly applied SD algorithms, Canny and ‘advanced mode in VGStudio’ (VG) algorithms, under beam hardening and noise effect. SD thresholding parameters were optimised by Taguchi methods when evaluating different types of surface conditions, including smooth and rough surfaces. Canny presented the worst performance during the measurement of rough surfaces but performed well in smooth surface measurement cases. VG performed worst in bi-directional measurement under the influence of beam hardening but successfully computed the rough surface when beam hardening was corrected. The proposed MCW demonstrated that it is an all-rounded algorithm. Compared to other surface determination algorithms, the MCW algorithm presented consistent results with little influence from the beam hardening, noise and threshold settings. This demonstrates a great potential that the proposed SD algorithm can be fully automated in XCT.