[home] [Personal Program] [Help]

---

Session: Session 11: Digital engineering II
Session starts: Wednesday 28 June, 10:50
Presentation starts: 10:50
Room: Theatre room: plenary

---

Hyeon-Gyu Park (Korea Advanced Institute of Science and Technology)
Kyu-Jin Lee (Korea Advanced Institute of Science and Technology)
Jung-Ryul Lee (Korea Advanced Institute of Science and Technology)

Abstract:
ABSTRACT Since 3D printed composites are capable of arbitrary three-dimensional shape design and have good mechanical properties, many studies have been conducted as parts materials for the aerospace industry [1][2]. However, during the long-term process of the 3D printing composite, the offset of the nozzle changes finely or defects caused by twisting have a problem of degrading mechanical properties. In addition, there is also a problem that is subjective and takes a lot of time when evaluating internal defects. Therefore, to address these problems, we aim to develop sophisticated defect diagnosis AI that can detect the location of depth direction defects. In this study, pulse-echo laser ultrasound examination techniques and six-degree-of-freedom robotic arm are used to examine real structures of complex shapes. The pulse-echo laser ultrasound examination technique is a method in which the excitation laser excites the structure of interest, and then the laser Doppler Vibrometer (LDV) receives it and performs a non-destructive examination. Six-degree-of-freedom robotic arm corrects the curvature so that the pulse-echo laser is vertically incident on any three-dimensional shape. Continuous fiber 3D printer was used to make specimens for learning data, Onyx with nylon-based carbon short fiber and carbon continuous fiber filaments were used as materials, and lamination was performed through the Material Extrusion method. The C-Scan image visualized using the pulse-echo laser ultrasonic was binarized by matching the visible result by minimum intensity projection through X-Ray Microscope, and then labeled for AI learning data. Deep ensemble CNN was modelled to extract advanced features by considering Time series and spatial information simultaneously. The final network was optimized through hyperparameter tuning, and the reliability of diagnostic AI was verified through 5-Fold validation. Through this study, defects on real-structure specimens of continuous fiber 3D printed CFRP composites were visualized through defect diagnosis AI, and defect detection target accuracy for test specimens was achieved. This study is expected to give many advantages such as establishing a process strategy by presenting the location of the defect beyond simply determining the presence or absence of the defect. [1] N Shahrubudin et al. Procedia Manufacturing. 35:1286-1296, 2019 [2] Sanei, S.H.R.; Popescu, D. 3D-Printed Carbon Fiber Reinforced Polymer Composites: A Systematic Review. J. Compos. Sci. 2020, 4, 98. https://doi.org/10.3390/jcs4030098