Research On Key Technologies Of Local Tomography For Planar Object In Industrial CT

As one of the best non-destructive testing(NDT) ways, interior structure of imaging object can be non-destructively reconstructed with X-ray projections by industrial Computed Tomography(CT). The field of view(FOV) of CT is limited and determined by X-ray cone angle, detector area, geometry magnification, etc. If the object cannot be completely surrounded by the FOV of CT, its projection will be truncated, which leads to local tomography problem. In local CT, the image reconstructed by traditional image reconstruction algorithm will be contaminated by truncation artifact. Truncation artifact in CT image is seen in the form of a dc bias and bright ring-like artifact on the edge of FOV, which is a challenge to NDT. Different from the theory of traditional reconstruction methods, local reconstruction algorithms based on a new mathematic theory intend to reduce truncation artifact, so it plays a key role in local CT. Since local reconstruction not only needs exact geometry to reduce geometry artifact but also fast speed to satisfy time requirement in practical applications, geometry calibration algorithm, local reconstruction algorithm and fast computation technique of local reconstruction are key in local tomography of industrial CT.There are a lot of planar parts such as printed circuit board and integrated circuit in modern electronic equipment. In the field of CT, planar part is called planar object and industrial CT based NDT is very important for normal running of it. However, testing efficiency of planar object may be decreased by truncation artifact in local CT. Because planar object has large area to thickness ratio, its local reconstruction is especial and different from other objects. Unfortunately, there is a lack of report on the local reconstruction of planar object. Since investigation achievements of local tomography for planar object can broaden applications of theory of local tomography and improve NDT efficiency of planar object in industrial CT, it has great academic meaning and practical value.In this work, key technologies of local tomography for planar object in circular CT, half-covered CT and helical CT are discussed, which are geometry calibration method, local reconstruction algorithm and fast computation technique of local reconstruction. The main researches and innovations of the work are described as follows.1. An interval subdividing based geometry self-calibration method for circular local CT is proposed. Self-calibration method can save hardware cost, but measurement of geometry artifact in reconstruction image may be influenced by the bright-like truncation artifact in local CT. In the method, two geometry parameters, which are the abscissa of projection of rotation center and rotation angle of detector along its normal, are calibrated, respectively. In the calibration process, initial interval of geometry parameter is set and subdivided equally firstly. Secondly, the subinterval that the similarity measurement of reconstruction images on its two end points is the greatest is selected and subdivided equally again. Repeat the two above steps until the stop criterions of subdividing are reached. To reduce bright ring-like truncation artifact, the Derivation Hilbert Back-projection(DHB) method is applied. To meet the time requirement, only a two dimensional slice image is reconstructed in each image reconstruction. Simulation and real experimental results show that geometry artifact in the local reconstruction image can be fast and significantly reduced by the proposed method.2. A local reconstruction method for local reconstruction of planar object in circular CT(LRPOC) is presented. Let d be the thickness of planar object, r the radius of FOV and δ the spatial resolution of CT, local reconstruction for planar object by LRPOC method can be divided into two cases. In the first case that d<2*sqrt[r2-(r-δ)2], vertical PI lines are selected and reconstructed by Back-projection Filtration(BPF) algorithm. In the second case that d≥2*sqrt[r2-(r-δ)2], two reconstructions are included. Firstly, vertical and non-truncated PI lines are selected and reconstructed by BPF method. Secondly, project onto convex set(POCS) algorithm is applied to reconstruct horizontal and truncated PI lines with image on vertical PI lines as aprior knowledge. Simulation and real experimental results show that planar object can be exactly reconstructed by LRPOC method.3. A method based on parameters of two ellipses is proposed to calibrate geometry of the half-covered CT. The half-covered CT can improve efficiency of NDT by extending horizontal FOV of CT. However, projection of marker in calibration object is a half of ellipse, which leads to that geometry parameters cannot be determined by existing offline calibration methods. In the proposed method, projections of a calibration object with two steel balls as markers are acquired firstly. Secondly, projection of each marker is fitted into an ellipse by the least square method, and rotation angle of detector is calculated. Thirdly, the projection of each marker is rotated with the calculated rotation angle and fitted again. At last, the remaining geometry parameters are determined. Simulation and real experimental results show that the geometry of half-covered CT can be well calibrated and geometry artifact in corrected image can be significantly reduced by the proposed method.4. A fast half-covered BPF(fHC-BPF) method for local reconstruction of planar object in half-covered CT is presented. In half-covered CT, image can be exactly reconstructed by the PI-line based HC-BPF algorithm. However, integral intervals of different PI lines are different, which decreases computation efficiency of HC-BPF method in computer implementation. Let d be the thickness of planar object, R the scanning radius of CT and Np the total number of projections in circular scan, that one of integral limitations of each PI line can be fixed if d<2R*sin(2π/Np) is demonstrated firstly. Then the other integral limitation of each PI line can be slightly enlarged and fixed because there is no projection data in this enlarged interval. By this way, integral intervals of all PI lines are equal. Simulation and real experimental results show that fHC-BPF method can improve computation efficiency of HC-BPF algorithm but without loss of image quality.5. A segmentation and interpolation based method is proposed to determine geometry parameters of helical CT. The Helical CT can improve efficiency of NDT by extending FOV along z-axis. However, geometry calibration is very troublesome because geometry parameters of each acquired projection need to be estimated. The proposed method has two steps. In the first step, the ascending axis of helical CT is segmented into several pieces and geometry parameters of circular CT at each section point are calculated. In the second step, geometry parameters of each projection in a piece are interpolated with those of the two end points of the piece. Simulation and real experimental results show that geometry artifact in reconstruction image of helical CT can be fast and significantly reduced by the method.6. A helical DHB(H-DHB) method for local reconstruction of planar object in helical CT is proposed. Because computational complexity of exact reconstruction algorithm based on PI-line theory is very high, approximate reconstruction method with low complexity is more attractive in practice. DHB algorithm is derived by replacing the ramp filter of FDK algorithm with first derivative operator and Hilbert transform operator. So the bright ring-like truncation artifact can be well reduced by DHB method. Similar to the helical FDK(H-FDK) algorithm, the H-DHB method is derived from DHB algorithm. Simulation and real experimental results show that local reconstruction image quality of H-DHB method is better than that of H-FDK method. Compared to the Laplace operator based reconstruction algorithm(LORA), the computation speed of the method is much faster.7. To improve the computational efficiency of local reconstruction algorithm for planar object, a fast computer implementation is proposed. Because the thickness of planar object is very thin, reconstruction of full volume determined by detector size will waste a lot of time and resource. To minimize reconstruction image data quantity of planar object, a fast method which determines the reconstruction area with projection of planar object is proposed. Since the back-projection has very high computational complexity, it is the most time-consuming step of analytic local reconstruction algorithms. To improve the computational speed of back-projection, CUDA based acceleration strategies of back-projection on graphic processing unit(GPU) are divided into constant and inconstant categories. And the optimal configuration parameters of inconstant strategies are estimated by minimizing run time of back-projection using the genetic algorithm. Experimental results show that reconstruction image data quantity of planar object can be significantly reduced and the optimal configuration parameters of back-projection on GPU can be searched with a few iterations by the proposed method. The method is tested on the RabbitCT platform and run time of back-projection is less than one second.