2nd International Symposium on NDT in Aerospace 2010 - We.1.A.4 Advanced Reconstruction Techniques Applied to an On-Site CT System Jonathan HESS, Markus EBERHORN, Markus HOFMANN, Maik LUXA Fraunhofer Development Center for X-Ray Technology EZRT, Fuerth, Germany Abstract. The 3D inspection of large or already mounted modules is not possible by conventional means up to now. This poses a challenge to non-destructive testing. It is shown that by using a system consisting of two robots and by using suitable calibration procedures a mobile CT system in short Robo-CT can be realized. For these new calibration procedures a method for image based pose estimation was devolved which allows a precise determination of the system alignment. By using new calibration methods and advanced reconstruction techniques, 3D volumes can be generated. 1. Introduction The 3D inspection of large or already mounted modules is not possible by conventional means up to now. This poses a challenge to non-destructive testing. Today s 3D x-ray systems depend on a circular movement of the specimen, while x- ray source and detector are fixed. By mounting the source and/or the detector on an independent manipulation system, the increased degrees of freedom can be used to gather a series of 2D projections while moving around the specimen. The Fraunhofer IIS developed new methods for image based pose estimation, allowing a precise determination of detector and source positions. In the majority of cases the traverse path is limited by the dimensions of the specimen. Therefore the trajectories are adapted to fit in the available work envelope. A demonstration system located at the Development Center for X-Ray Technologies (EZRT) in Fuerth, consisting of two robots holding an x-ray source and detector, shows the capabilities of these methods. 2. System Set-up The Fraunhofer Development Center X-Ray Technology (EZRT) has developed a prototype system (Figure 1) consisting of two robots, a mini focus x-ray tube and a digital flat panel detector. Using this prototype a new calibration procedure and advanced reconstruction techniques for on-site CT systems are evaluated. Licence: http://creativecommons.org/licenses/by-nd/3.0 1
Figure 1: System set-up mobile Computed Tomography 2.1 Robots Robot arms of the Stäubli RX90 series have a maximum payload of 7 kg and a maximum range of 900 mm when using the standard six axis set-up. Taking into account these conditions a suitable x-ray source and detector were chosen which do not surpass the maximum load and have a compact size. Thus, a minimal limitation of operating range is assured. It should be mentioned that positioning precision and repeatability decrease with increasing payload. Both arms are mounted on mobile sockets that allow manual positioning. Four pedestals ensure a stable base of the robots after positioning. The tool mount is adequate for the simultaneous mounting of the UltraCal laser system[1] and the x- ray components. 2.2 X-ray Components The x-ray detector Hamamatsu C9311DK comes with an active pixel size (Pitch) of 100 x 100 µm and a CsI Scintillator. The x-ray source MXR-160HP/11 manufactured by Comet AG comes with a focal spot size of 0.4 mm and 1.0 mm according to EN12543. The voltage range is about 20 160 kv with a maximum dose output of 800 W on the small focal spot and 1800 W respectively on the big spot. A fist setup was realized with an detector which was available at the Fraunhofer IIS. The payload of the robots was limited, so an optimal matching of focal spot size and detector pitch was not possible. 2
2.3 Laser-calibration system The UltraCal laser system is consisting of a class 2 laser with integrated laser sensor and a retro reflector that reflects incident laser light parallel. The retro reflection is done by a triple mirror consisting of three perpendicular mirrors. The included sensor offers a laser shift in two dimensions with a precision of ±25 µm. 2.4 Manipulator control The control system of the Robo-CT was transparently mapped onto the interfaces of a conventional CT system. Thus, it is possible to use existing software libraries and acquisition methods. The manipulation of the object carrier common in industrial CT systems is mimicked by the simultaneous movement of both robots. Together with the six degrees of freedom for the manipulation of the x-ray source and the detector no limitations arise for the system's trajectories. 3. Initial system calibration The two robots are fully independent and therefore a recalibration is necessary, after they have been positioned at the testing site. The UltraCal laser system is used for the initial system calibration which determines the relative position of the robots towards each other. The values calculated by the UltraCal Laser system have a deviation of around 300 µm, but for the reconstruction an even higher precision is necessary. Additionally the relative position of the robots to the object has to be determined. These values can be calculated by analysing the x-ray-projections of a calibration-object. This technique is called pose estimation. 4. Pose estimation As already mentioned, the relation of x-ray tube and detector to each other, as well as the relation of the specimen to the robots is unknown. A calibration object is placed next to the specimen and visible on each projection. The geometry of the calibration object was previously determined by a conventional coordinate measuring machine. A special calibration object was designed for the use with the Robo- CT. It is basically a small rod with attached spheres in helical pattern which is shown in figure 2. The distribution of the spheres in space is an advantage in means of misalignments. The helical pattern is robust against misalignments and it allows a positioning of the object, where no overlap of the spheres during a measurement occurs. If the measurement geometry and the calibration object are well known, it is possible to calculate the translation and rotation of the x-ray- source and the detector. For this calculation the POSIT algorithm is used [2]. Designed for use with cameras and LEDs, the algorithm can be applied to the Robo-CT. 3
Figure 2. A three dimensional model of the calibration rod With this information the pose of the calibration object in every single projection can be calculated. Using the calibration object of the first projection as a reference and relating the poses of all other projections to this first one, a set of position data for the ART (Algebraic Reconstruction Technique) reconstruction [3] can be generated. Experiments have shown that the quality of the reconstruction can be improved significantly by using object calibration. The figures 3 and 4 show impact damages on a fibre reinforced composite plate. Evidently, artefacts and noise are reduced. The plate was inspected using a 90 degrees limited angle CT Figure 3. Uncorrected reconstruction 4
Figure 4. Corrected reconstruction 6. Summary and Outlook It is shown, that by combining high accuracy position determination and advanced reconstruction techniques, the generated 3D volumes can be improved. The used algorithms can cope with imprecise measuring setups and make it possible to use arbitrary geometries for acquiring projections. Acknowledgements This Project is co-financed by the European Union and the Free State of Bavaria. References [1] UltraCal Robot Calibration System, Robo-Technology GmbH, http://www.robo-technology.de/ultracal_v5.pdf [2] Daniel F. DeMenthon und Larry S. Davis: Model-Based Object Pose in 25 Lines of Code. Computer Vision Laboratory, University of Maryland, Maryland, 1995 [3] Avinash C. Kak und Malcom Slanley: Principles of Computerized Tomographic Imaging. IEEE Press, New York, 1. Auflage, 1999 5