Automated Detection and Quantification of Microstructural Defects in Images of ۳D-Printed Ti۶Al۴V Using Deep Learning

Detecting microstructural defects in additively manufactured Ti۶Al۴V components is crucial due to their impact on mechanical properties. This research utilizes an advanced deep learning method, YOLOv۱۰x, to identify three key defect types—Pore, Lack of Fusion (LOF), and Un-melted Powder (UP)—in scanning electron microscopy (SEM) images, achieving over ۹۰% confidence in detection. A Microstructural Defect Dataset was used for model training. To further characterize defects, a developed Python script calculates geometric features by inscribing an oval in the bounding box of detected defects and computing center coordinates, diameters, and area. These analyses enhance understanding of defect size, shape, and effects on microstructural integrity, aiding quality control in additive manufacturing. The results show the model’s robust capability for accurate automatic defect detection and characterization across various defect types. The approach could be promising for advancing automated inspection in additive manufacturing.

کلیدواژه ها:

Microstructural Defect Detection ، Deep Learning ، You Only Look Once (YOLO) v۱۰ ، Additive Manufacturing ، Ti۶Al۴V ، Scanning Electron Microscopy (SEM)