real-time data-driven microstructural defect detection in scanning electron microscopy images of additively manufactured ti6al4v using advanced deep learning method

The automated detection of microstructural defects in additively manufactured ti6al4v materials presents significant challenges due to the lack of comprehensive datasets and the variability of defect types. this study introduces a novel methodology for addressing these challenges by developing a microstructural defect dataset (mdd) specifically tailored for scanning electron microscopy (sem) images. we trained and evaluated multiple yolov8 models—yolov8n, yolov8s, yolov8m, yolov8l, and yolov8x—using this dataset to assess their effectiveness in detecting various defects. the principal results demonstrate that yolov8m achieves a balanced trade-off between precision and recall, making it suitable for reliable defect identification across diverse defect types. yolov8s, on the other hand, excels in efficiency and speed, particularly for detecting ’pore’ defects. the study also highlights the limitations of yolov8n in detecting specific defect types and the computational challenges associated with yolov8l and yolov8x. our methodology and findings contribute to the scientific understanding of automated defect detection in additive manufacturing. the development of the mdd and the comparative evaluation of yolov8 models advance the state of knowledge by providing a robust framework for detecting microstructural defects. future research should focus on expanding the dataset and exploring advanced ai techniques to enhance detection accuracy and model generalization.