Research On Techniques Of CT ART's Parellel Acceleration And Optimization Of Image Quality

Industrial CT (ICT) is a testing technology that obtains the inner information including geometric structure and material ingredient of objects by reconstructing gray images of the intersection slices in the non-destructive condition. Given the merit of insubjection to the restrictions of material and structure, in many developed countries, ICT is widely applied into non-destructive testing (NDT), evaluation and reverse engineering in many fields. Considered as the best NDT approach in the modern world, ICT has broad application perspective and favorable developing future.Algebraic Reconstruction Technology (ART) is the most efficient method to replace the filter back-projection reconstruction method and resolve the problem of incomplete projection data's reconstruction. However, conventional ART bears some shortcomings such as time-consuming computation, huge data storage, low image quality and ill reconstruction efficiency, which become the bottle-neck of its utilization. On this research background, based on the mathematic principles of ART, comprehending computing parallel technology, this dissertation will research on ART's parallel acceleration and quality optimization of reconstruction images, and bring forward several approaches of NDT of huge parts in aerospace crafts whose sizes are larger than that of CT detector.The primary research content and fruits as following:1. Rapid and High Efficiency Projection Arithmetic Based on Phantom.Aiming at the requirements of ART, combining the specialties of CT simulation condition, this dissertation researches on the 2D and 3D forward projection arithmetics based on phantom model. A rapid voxel-traversing method is brought forward to promote the projection speed;complex Simpson integral is used to improve the projection accuracy of parts with heterogeneously distributing materials. The results of simulation experiments illustrate that, in the condition of satisfying the accuracy requirement of functionally gradient materials' projection, compared to the common UG arithmetic, this method remarkably speeds up the computation with accelerating ratio of more than 100.2. Rapid Parallel ART. Based on the projection method above, researching on the mathematic principles of ART, integrating the SIMD technologies such as MMX, SSE(2) of Intel CPU, a rapid parallel ART is advanced. This dissertation will discuss the accelerating technique in iterative process and optimized memory accessing mode in detail, and demonstrate the final procedure of the arithmetic. Compared to conventional ART, this method obtains accelerating ration of up to 20, which strengthens the applicationalability of ART.3. Reconstruction Methods of Huge Parts. Several resolvents is created to reconstruct parts with huge sizes and broad and flat shapes. Simulation experiments has proved the feasibility of these methods, and received results meeting the requirements of engineering application.4. Quality Optimization of Reconstruction Image. According to projection accessing principle, an improved method is advanced to optimize the accessing sequence of projection angles. Ground on this, this dissertation has finished a projection accessing approach based on the feature of the part, which enhances the efficiency of computing the sequence of projection angles and is provided with self-adaptive ability to parts with different shapes. This one dramatically improves the reconstruction accuracy and results just 1/3 error of previous projection accessing methods.