Automatic detection and severity assessment of stringing defect in FDM 3D printing using deep learning
Author
Herman, Damian Dariusz
Term
4. semester
Publication year
2024
Submitted on
2024-05-31
Pages
73
Abstract
The thesis investigates defects occurring during the Additive Manufacturing process and proposes a CNN-based solution to detect and assess the severity of the stringing defect. A YOLOv8 and MobileNetV3 solution to first detect and then assess stringing defects’ severity have been proposed to solve the problem. A labeled object detection dataset with severity labels have been created. Tests on the dataset and as well on a real-life capture from a webcam overseeing the printing process have been conducted, to see whether the proposed solution is viable to solve the problem of both manufacturing supervision and defect severity assessment. To find out if the solution is viable, several metrics have been calculated to measure, how well the solution performs, as well resource usage have been measured, to find out if the solution won’t interrupt the manufacturing process. On the test split of created dataset, the detection model performs well, however it fails to generalize on the real-life capture and the severity assessment model is not accurate enough.
Keywords
Documents