Quality assurance of CT scanning for industrial applications

Jais Andreas Breusch Angel

Research output: Book/ReportPh.D. thesis

1169 Downloads (Pure)


X-ray computed tomography (CT) gives major possibilities by looking through the industrial parts with complex geometries, but one of the largest challenges is the quality assurance of measurements. This Ph.D. project at DTU Mechanical Engineering deals with the development of procedures for quality assurance of CT for industrial measurements both in the manufacturing and in the meat processing industries. Various methods and reference objects have been developed in this project to establish metrological traceability of measurements. Moreover investigations as well as international comparisons in the field of application on the two different areas have been carried out.
Different reference objects have been developed and introduced for the manufacturing industry: step gauge, step cylinder and a cylindrical multi-material assembly. These objects can be used for correction of measurement errors in the CT model. Moreover, two reference objects are calibrated objects from the manufacturing industry: a threaded tube from the medical industry and a LEGO brick from the toy industry. Establishment of traceability for all objects is performed using coordinate measuring machines (CMMs) with known uncertainty. The stability has been documented for all reference objects except for the step cylinder and the cylindrical multi-material assembly.
A design of experiment (DOE) was performed on measuring errors arising in a CT, in terms of material density and orientation of scanned step gauges. An analysis of variance (ANOVA) shows that all main factors and their interactions are significant. The maximum deviation from the reference value can be reduced by compensating for systematic errors, but it is more complicated to correct for vertical orientations in high density materials.
In an interlaboratory comparison involving 27 laboratories from 8 countries, measurements were carried out using CT on two common objects from the manufacturing industry, a threaded tube and a LEGO brick. The comparison has shown that CT measurements on the industrial parts used lie in the range 6-53 μm, with maximum values up to 158 μm, compared to average uncertainties below 5.5 μm using CMMs.
A test was performed to check if X-ray contrast modalities can be applied for metrological purposes. Traditionally, segmentation between multi-materials in CT scanning is done by using different edge detection techniques and threshold algorithms, but these are only available for multi-materials where the densities are not close to each other. X-ray contrast modalities overcome this problem by constructing dark field, phase contrast and transmission images. Measurement results show that further development related to stability issues on the used CT is needed to create a metrological tool using X-ray contrast modalities.
Two synthetic reference phantoms have been developed by Danish Meat Research Institute (DMRI) and introduced for the meat processing industry. The phantoms represent real pig carcasses and are made of several polymer components, representing tissue types such as lean meat, fat, and bone. Establishment of traceable volume measurements for the phantoms is performed using the gravimetric method (also called water displacement). The stability has been documented for the two phantoms.
For the meat processing industry concerned, a similar interlaboratory comparison using two reference phantoms from the meat processing industry was carried out using CT, and involved four laboratories from 4 countries. The comparison has shown that CT measurements on the phantoms used lie in the range 1-1090 mL, with maximum values up to 1348 mL, compared to average uncertainties below 10 mL using the gravimetric method.
DMRI and DTU Compute have previously developed advanced image analysis software (PigClassWeb) which performs virtual dissections in pig carcasses. A DOE was carried out to document the performance of PigClassWeb through volume comparisons to real dissections of pig carcasses. For the real dissections, volumes of tissue types such as bone, lean meat and fat, are estimated using commercial VolumeGraphics software. It is detected that the ANOVA and the residuals from the virtual dissection fail the normality test. The reason can be that the simulation data has special problems and challenges which are difficult to overcome by using current regression software.
Original languageEnglish
PublisherDTU Mechanical Engineering
Number of pages181
Publication statusPublished - 2014

Fingerprint Dive into the research topics of 'Quality assurance of CT scanning for industrial applications'. Together they form a unique fingerprint.

Cite this