AI-Powered Analysis of Industrial CT Data

Abstract

Extensive research and development has been conducted in the field of AI-powered analysis of medical CT data during the past years – with significant progress. Although voxel data from industrial parts differ from medical data in the contrast level and the resolution, the medical DL approach is promising. Therefore, the aim of this work is exploring and developing neural network models for detecting defects in industrial CT data. Network architectures, successfully applied to medical CT data, were investigated and derivates were developed. Different neural network models were trained utilising a mixture of synthetic data and real data. The evaluation showed very good results for a modified U-Net neural network.