X-ray Computed Tomography (CT) scanning is a well-established technique that combines X-ray images collected at different angles through a part using a reconstruction algorithm to produce a three-dimensional distribution of the X-ray attenuation. The resulting attenuation is a function of the photon energy and the mass density and atomic number of the materials being imaged. Classical reconstruction algorithms assume the X-ray source is monochromatic and so artifacts occur with most industrial CT systems that use the broad spectrum produced by an X-ray tube. This paper describes the use of multi-energy CT imaging to reduce or eliminate beam-hardening artifacts while at the same time providing enhanced information about the material(s) being imaged. The goal is to be able to separate and identify different materials and to be able to quantify mass density and atomic number.
Multi-energy CT scanning is performed by imaging a part using X-rays from two or more distinct energy bands. Data processing algorithms take advantage of the independent information from each of the bands to extract more information than is available with conventional single-energy CT scanning. In this project, two general methods of data processing are being investigated. First is a data-driven method where calibration data sets are used to correct beam-hardening artifacts and to identify multiple material types in the part being imaged. To identify N different materials, at least N different X-ray energy spectra must be used. The second method being investigated is a model-based approach. In this approach a model of the X-ray generation and detection for the CT scanner being used is combined with a model of material X-ray absorption to extract additional information about the part being scanned. At the photon energies used in this study, X-ray absorption is a function of the mass density and the effective atomic number of the material being imaged. The goal of the model-driven approach is to estimate the mass density and effective atomic number from the acquired multi-spectral CT images.
The data-driven approach has been completed and demonstrated by separating stainless steel and aluminum using a dual-energy CT scan. The separate steel and aluminum 3D images have significantly reduced beam-hardening artifact as expected. For the model-driven approach, the model of image formation for the CT scanner being used has been completed and verified. Current work is focused on solving the inverse problem to extract the density and effective atomic number parameters.