Generative Synthesis of Defects in Industrial Computed Tomography Data

Abstract

The need for data increases as more and more companies try to take their first steps with AI to improve their efficiency and processes. Addressing this problem, we propose a solution for synthetic generation of industrial CT data for further AI applications using a deep learning approach packaged in a process pipeline. Based on a few individual CT scans of components with internal defects, the pipeline is able to generate STLs of any component with a large variation of artificially generated defects inside. Using this data with CT simulation, for example, provides access to creating large databases to overcome data lag and enrich further AI applications.