Blarr, JulianeJulianeBlarrKunze, PhilippPhilippKunzeKresin, NoahNoahKresinLiebig, WilfriedWilfriedLiebigInal, KaanKaanInalWeidenmann, Kay AndréKay AndréWeidenmann2024-03-262024-03-262024https://publica.fraunhofer.de/handle/publica/46451210.1016/j.ndteint.2024.1030672-s2.0-85185563337In order to determine fiber volume contents (FVC) of low contrast CT images of carbon fiber reinforced polyamide 6, a novel thresholding method and a convolutional neural network are implemented with absolute deviations from experimental values of 2.7% and, respectively, 1.46% on average. The first method is a sample thickness based adjustment of the Otsu threshold, the so-called “average or above (AOA) thresholding”, and the second is a mixed convolutional neural network (CNN) that directly takes 3D scans and the experimentally determined FVC values as input. However, the methods are limited to the specific material combination, process-dependent microstructure and scan quality but could be further developed for different material types.enCarbon fiber reinforced polymersDeep learningLow contrastThermoplasticsX-ray tomographyjournal article