| Περίληψη: | The primary objective of this study was to investigate the use of conventional micro-CT and neural network (NN) algorithms for image processing to accurately visualize and detect the fibers in 3D printed composites and to use the information of the fiber detection in analytical micromechanical homogenization model in order to predict their homogenized elastic properties. To this end, a custom version of YOLO (You Only Look Once) v7 Tiny model was used. The model was trained on manually labeled images of micro-CT results for a 3D printed composite reinforced with continuous Kevlar fibers and another one reinforced with continuous carbon fibers. The micro-CT scans were performed in the Vrije Universiteit Brussel (VUB), in Brussels Belgium and were utilized in this study. The results demonstrate that the YOLOv7 algorithm was able to capture the boundaries of the fibers accurately leading to clearly defined fibers in the matrix material. Furthermore, the ability to provide results for thousands of images in a few minutes indicates its time efficiency and practicability. Due to the aforementioned factors, the custom version of YOLO v7 is a valuable addition to the scientific literature and can be used by both industry and academic professionals to obtain rapid and accurate results of fibers from micro-CT images. Finally, the results of the segmented images allowed to measure the volumes of the fibers and the matrix materials and estimate their volume fractions. In turn, the volume fractions were included in seven analytical micromechanical models to estimate the homogenized elastic properties of the single printed filaments of nylon reinforced with Kevlar fibers. The homogenized elastic properties were compared to tensile tests and showed great agreement. Out of the seven models, the best agreement to the most accurate one, the concentric cylinder model of Polyzos et. al., was achieved by the upper and lower bounds of the Hashin-Rosen model.
|
|---|
