We present a new theoretical model and reconstruction results for a new class of fast x-ray CT machine -- the Real Time Tomography (RTT) system, which uses switched sources and an offset detector array. We begin by reviewing elementary properties of the Radon and X-ray transforms, and limited angle tomography. Through the introduction of a new continuum model, that of sources covering the surface of a cylinder in R³, we show that the problem of three-dimensional reconstruction from RTT data reduces to inversion of the three-dimensional Radon transform with limited angle data. Using the Paley-Wiener theorem, we then prove the existence of a unique solution and give comments on stability and singularity detection. We show, first in the two-dimensional case, that the conjugate gradient least squares algorithm is suitable for CT reconstruction. By exploiting symmetries in the system, we then derive a method of applying CGLS to the three-dimensional inversion problem using stored matrix coefficients. The new concept of source firing order is introduced and formalised, and some novel visualisations are used to show how this affects aspects of the geometry of the system. We then perform a detailed numerical analysis using the condition number and SVD of the forward projection matrix $A$, to show that the choice of firing order affects the conditioning of the problem. Finally, we give reconstruction results from both simulated phantoms and real experimental data that support the numerical analysis.

The aim of this report is to study how machine learning can be used to reconstruct 2 dimensional computed tomography images from limited angle data. This could be used in a variety of applications where either the space or timeavailable for the CT scan limits the acquired data.In this study, three different types of models are considered. The first model uses filtered back projection (FBP) with a single learned filter, while the second uses a combination of multiple FBP:s with learned filters. The last model instead uses an FNO (Fourieer Neural Operator) layer to both inpaint and filter the limited angle data followed by a backprojection layer. The quality of the reconstructions are assessed both visually and statistically, using PSNR and SSIM measures.The results of this study show that while an FBP-based model using one or more trainable filter(s) can achieve better reconstructions than ones using an analytical Ram-Lak filter, their reconstructions still fail for small angle spans. Better results in the limited angle case can be achieved using the FNO-basedmodel.

La tomographie par rayons X ou CT pour "Computed Tomography" est un outil puissant pour caractériser et localiser les défauts internes et pour vérifier la conformité géométrique d’un objet. Contrairement au cas des applications médicales, l’objet inspecté en Contrôle Non Destructif (CND) peut être très grand et composé de matériaux de haute atténuation, auquel cas l’utilisation d’une trajectoire circulaire pour l’inspection est impossible à cause de contraintes dans l’espace. Pour cette raison, l’utilisation de bras robotisés est l’une des nouvelles tendances reconnues dans la CT, car elle autorise plus de flexibilité dans la trajectoire d’acquisition et permet donc la reconstruction 3D de régions difficilement accessibles dont la reconstruction ne pourrait pas être assurée par des systèmes de tomographie industriels classiques. Une cellule de tomographie X robotisée a été installée au CEA. La plateforme se compose de deux bras robotiques pour positionner et déplacer la source et le détecteur en vis-à-vis. Parmi les nouveaux défis posés par la tomographie robotisée, nous nous concentrons ici plus particulièrement sur la limitation de l’ouverture angulaire imposée par la configuration en raison des contraintes importantes sur le mouvement mécanique de la plateforme. Le deuxième défi majeur est la troncation des projections qui se produit lorsque l’objet est trop grand par rapport au détecteur. L’objectif principal de ce travail consiste à adapter et à optimiser des méthodes de reconstruction CT pour des trajectoires non standard. Nous étudions à la fois des algorithmes de reconstruction analytiques et itératifs. Avant d’effectuer des inspections robotiques réelles, nous comptons sur des simulations numériques pour évaluer les performances des algorithmes de reconstruction sur des configurations d’acquisition de données. Pour ce faire, nous utilisons CIVA, qui est un outil de simulation pour le CND développé au CEA et qui est capable de simuler des données de projections réalistes correspondant à des configurations d’acquisition définies par l’utilisateur. / X-ray computed tomography (CT) is a powerful tool to characterize or localize inner flaws and to verify the geometric conformity of an object. In contrast to medical applications, the scanned object in non-destructive testing (NDT) might be very large and composed of high-attenuation materials and consequently the use of a standard circular trajectory for data acquisition would be impossible due to constraints in space. For this reason, the use of robotic arms is one of the acknowledged new trends in NDT since it allows more flexibility in acquisition trajectories and therefore could be used for 3D reconstruction of hardly accessible regions that might be a major limitation of classical CT systems. A robotic X-ray inspection platform has been installed at CEA LIST. The considered system integrates two robots that move the X-ray generator and detector. Among the new challenges brought by robotic CT, we focus in this thesis more particularly on the limited access viewpoint imposed by the setup where important constraints control the mechanical motion of the platform. The second major challenge is the truncation of projections that occur when only a field-of-view (FOV) of the object is viewed by the detector. Before performing real robotic inspections, we highly rely on CT simulations to evaluate the capability of the reconstruction algorithm corresponding to a defined scanning trajectory and data acquisition configuration. For this purpose, we use CIVA which is an advanced NDT simulation platform developed at CEA and that can provide a realistic model for radiographic acquisitions and is capable of simulating the projection data corresponding to a specific CT scene defined by the user. Thus, the main objective of this thesis is to develop analytical and iterative reconstruction algorithms adapted to nonstandard trajectories and to integrate these algorithms in CIVA software as plugins of reconstruction.

Cette thèse traite du couplage d'un tomographe optique par luminescence (LCT) et d'un tomographe par rayons X (XCT), en présence d'une contrainte sur la géométrie d'acquisition du XCT. La couverture angulaire du XCT est limitée à 90 degrés pour satisfaire des contraintes spatiales imposées par le LCT existant dans lequel le XCT doit être intégré. L'objectif est de dériver une information anatomique, à partir de l'image morphologique issue du XCT. Notre approche a consisté i) en l'implémentation d'un algorithme itératif régularisé pour la reconstruction tomographique à angle limité, ii) en la construction d'un atlas anatomique statistique de la souris et iii) en l'implémentation d'une chaîne automatique réalisant la segmentation des images XCT, l'attribution d'une signification anatomique aux éléments segmentés, le recalage de l'atlas statistique sur ces éléments et ainsi l'estimation des contours de certains tissus à faible contraste non identifiables en pratique dans une image XCT standard. / This thesis addresses the combination of an Optical Luminescence Tomograph (OLT) and X-ray Computerized Tomograph (XCT), dealing with geometrical constraints defined by the existing OLT system in which the XCT must be integrated. The result is an acquisition geometry of XCT with a 90 degrees angular range only. The aim is to derive an anatomical information from the morphological image obtained with the XCT. Our approach consisted i) in the implementation of a regularized iterative algorithm for the tomographic reconstruction with limited angle data, ii) in the construction of a statistical anatomical atlas of the mouse and iii) in the implementation of an automatic segmentation workflow performing the segmentation of XCT images, the labelling of the segmented elements, the registration of the statistical atlas on these elements and consequently the estimation of the outlines of low contrast tissues that can not be identified in practice in a standard XCT image.

This thesis describes the analysis, design and implementation of a high performance DSP-based servo drive for a limited-angle torque motor used in thermal imaging applications. A limited-angle torque motor is an electromagnetic actuator based on the Laws' relay principle, and in the present application the rotation required was from - 10° to + 10° in 16 ms, with a flyback period of 4 ms. To ensure good quality picture reproduction, an exceptionally high linearity of ±0.02 ° was necessary throughout the forward sweep. In addition, the drive voltage to the exciting winding of the motor should be less than the +35 V ceiling of the drive amplifier. A research survey shows that little literature was available, probably due to the commercial sensitivity of many of the applications for torque motors. A detailed mathematical model of the motor drive, including high-order linear dynamics and the significant nonlinear characteristics, was developed to provide an insight into the overall system behaviour. The proposed control scheme uses a multicompensator, multi-loop linear controller, to reshape substantially the motor response characteristic, with a non-linear adaptive gain-scheduled controller to compensate effectively for the nonlinear variations of the motor parameters. The scheme demonstrates that a demanding nonlinear control system may be conveniently analysed and synthesised using frequency-domain methods, and that the design techniques may be reliably applied to similar electro-mechanical systems required to track a repetitive waveform. A prototype drive system was designed, constructed and tested during the course of the research. The drive system comprises a DSP-based digital controller, a linear power amplifier and the feedback signal conditioning circuit necessary for the closed-loop control. A switch-mode amplifier was also built, evaluated and compared with the linear amplifier. It was shown that the overall performance of the linear amplifier was superior to that of the switch-mode amplifier for the present application. The control software was developed using the structured programming method, with the continuous controller converted to digital form using the bilinear transform. The 6- operator was used rather than the z-operator, since it is more advantageous for high speed sampling systems. The gain-scheduled control was implemented by developing a schedule table, which is controlled by the DSP program to update continuously the controller parameters in synchronism with the periodic scanning of the motor. The experimental results show excellent agreement with the simulated results, with linearity of ±0.05 ° achieved throughout the forward sweep. Although this did not quite meet the very demanding specifications due to the limitations of the experimental drive system, it clearly demonstrates the effectiveness of the proposed control scheme. The discrepancies between simulated and experimental results are analyzed and discussed, the control design method is reviewed, and detailed suggestions are presented for further work which may improve the drive performance.