The purpose of non-descructive enluation (NDE) is to acquire knowledge of the investigated sample. Digital x-ray imaging techniques such as radiography or computerised tomography (CI) produce images of the interior of a sample. The obtained image quality determines the possibility of detecting sample, ·elated features, e.g. details and flaws. this thesis presents a method of optinllsing the performance of industrial X-ray equipment for the imaging task at issue in order to obtain images with high quality.
CT produces maps of the X-ray linear attenuation of the sample's interior. CT can produce two-dimensional cross-section images or three-dimensional images with volumetric information on the investigated sample. The image contrast and noise depend on both the investig-Ated sample and the equipment and settings used (X-ray tube potential, X-ray filtration, exposure time, etc.). Hence, it is vital to find the optimal equipment settings in order to obtain images of high quality.
To be able to mathematically optimise the image guality, it is necessary to have a model of the X-ray imaging system together with an appropriate measure of image quality. The optimisation is performed with a developed model for an X-ray image-intensifier-based radiography system. The model predicts the mean value and variance of the measured signal level in the collected radiographic images. The traditionally used measure of physical image guality is the signal-to-noise ratio (SNR). To calculate the signal-to-noise ratio, a well-defined detail (flaw) is required. It was found that maximising the SNR leads to ambiguities, the optimised settings found by maximising the SNR were dependent on the material in the detail. When CT is performed on irregular shaped samples containing density and compositional variations, it is difficult to define which SNR to use for optimisation. This difficulty is solved by the measures of physical image quality proposed here, the ratios geometry-sensitivity/ noise, density-sensitivity/noise, and mass attenuation-sensitivity/noise. With these measures, a meiliod is presented that finds the optimal eguipment settings, where no improvement can be made without worsening at least one other sensitivity/noise ratio.
This thesis includes modelling and verification of the sharpness of the CT system in terms of the modulation transfer function, MTF. Together with the limiting perception factor and the maximised SNR, the detectability limits for any specific contrasting detail in the centre of a cylindrical sample can be determined. It is also demonstrated that the model can be used to suppress beam hardening when collecting CT-data. When homogeneous samples are imaged, the model can in addition be used to make post-processing corrections for suppressing the beam hardening artefacts.
Wavelet-based local tomography has been found to produce images with good accuracy from projection data only from a small region in a sample. Tlus technique is demonstrated on thermal barrier coatings, which contain internal cracks. With optimised eguipment settings and geometrical magnification of a region in the sample, wavelet-based local tomography produced high-resolution images of excellent quality. The increased resolution reveals features in the microstructure that cannot be resolved wiili traditional CT. This technigue will be a useful tool for characterisation of the microstructure in advanced materials.
Linköping: Linköpings universitet , 2000. , 34 p.