Segmentation tool for the extraction of individual short fibers from micro-tomographs of additively manufactured reinforced polymer composites

Additive manufacturing of carbon fiber reinforced polymers (CFRP) offers great potential for producing high strength-to-weight parts of a geometric complexity that is impractical to obtain by classical manufacturing methods. The heterogeneity of the microstructure produced has been identified as a crucial challenge in terms of modeling. Now that the use of X-ray micro-tomography allows the volumetric imaging of such solids at the constituent scale, the absence of reliable segmentation tools by which the phases are discriminated and the individual fibers identified are a major hurdle to the use of such data in homogenization models. Short-fiber additively manufactured CFRPs have morphological features such as porosity, high fiber misalignment and high frequency of contact between fibers, which further complicate the segmentation task. To solve this problem, a segmentation tool has been developed that uses a machine learning package developed for unidirectional continuous fibers, and augments its capabilities by adding heuristics and processing that uses a priori knowledge about the morphology of short fibers to correctly label them in tomographic scans. With this tool, correct identification of individual fibers becomes possible even when the output of the machine-learning labelling is partially accurate. This considerably reduces the workload of producing manual expert labeling to train deep learning tools, and makes those tools more robust, as it compensates for their short-comings and narrow domain of validity. The reliance on distributed computing and optimized image processing libraries like OpenCV makes this an efficient solution for full-sized datasets, containing tens of thousands of individual fibers. By successfully extracting these fibers, insights can be gained into the relationship between processing parameters, microstructure and mechanical performance, opening up avenues of investigation and optimization of advanced composites.