Publikation

GPU-accelerated Computed Laminography with Application to Nondestructive Testing

Michael Maisl, Lukas Marsalek, Christian Schorr, Jan Horacek, Philipp Slusallek

In: Proceedings of the 11th European Conference on Non-Destructive Testing. European Conference on Non-Destructive Testing (ECNDT-2014) October 6-10 Prague Czech Republic ISBN 978-80-214-5018-9 10/2014.

Abstrakt

Computed tomography (CT) is a very powerful tool in medicine and non-destructive testing but it is unsuitable for planar objects like printed circuit boards or fiber reinforced plastics sheets, due to their strongly varying penetration lengths and spatial restrictions during the scan. A solution can be found in the use of computed laminography (CL), a technique where the object is irradiated by an oblique angle, thereby circumventing the problems arising in CT. The innovative scanner system CLARA (Computed laminography and radioscopy device) realizes this geometry in a new and efficient way, compared to existing industrial systems. Instead of four translational axes, CLARA only needs one rotational axis, greatly reducing both the costs and the calibration errors. Due to the limited amount of angular coverage and the specific geometric setup, filtered back projection methods used in CT cannot be employed for the reconstruction of laminographic projections. More flexible iterative algorithms like SART (simultaneous algebraic reconstruction technique) provide an answer to this challenge and also allow to incorporate a priori knowledge about the object to increase the reconstruction quality. The drawback of these algorithms lies in their high computational demands, resulting in typical reconstruction times in the order of hours. For certain practical applications this may be too time-consuming and therefore an acceleration of the algorithms is necessary. Compared to desktop central processing units (CPUs), modern graphics processing units (GPUs) contain an order of magnitude more cores designed specifically for massive data parallel processing. By designing new algorithms to fully exploit the GPU architecture, reconstruction times can be reduced from hours on a CPU to mere minutes on the GPU, thus allowing the practical use of computed laminography in industrial settings. Areas of application include the inspection of printed circuit boards, fiber reinforced plastics parts of wind energy plants and cars, as well as thegeneral testing of sheet-like objects. We explain the advantages of the CLARA scanner and demonstrate the feasibility of computed laminography by means of CLARA results of industrially relevant inspection challenges using GPU-enhanced algorithms.

Deutsches Forschungszentrum für Künstliche Intelligenz
German Research Center for Artificial Intelligence