Automated defect detection of CT projection image data using Monte Carlo simulation

During the last decade industrial computed tomography has become one of the most important metrological procedures for internal inspection, where it sees widespread application in additive manufacturing. Evaluating the CT volume data for defects is currently a lengthy process involving data acquisition, reconstruction, surface reconstruction, and nominal/actual comparison. The goal of the presented project is the development of a new pipeline for automated defect detection operating solely with projection data. Using this pipeline, the amount of necessary projections NP and therefore the measurement time of each object will be heavily reduced. Reference projection data of non-defect objects were generated using a multi-GPU Monte Carlo X-ray simulation. The innovative implementation of the Monte Carlo simulation on GPUs makes the photon number of 5x1011required for a proper simulation of an X-ray projection feasible for the first time. This generated reference data was then compared to real data and the differences evaluated. With this new processing pipeline, it is now possible to achieve a defect analysis with less than six projection images, which decreases the minimum measurement time tmby nearly two magnitudes.