YOLOV10-S For 3D Printing Anomaly Detection and Classification

The rapid advancement of additive manufacturing, particularly Fused Deposition Modeling (FDM) 3D printing, has transformed prototyping and small-scale production. However, anomalies such as extrusion problems, spaghetti, stringing, and warping continue to compromise print quality, leading to material waste and inefficiencies. These challenges highlight the need for automated, real-time defect detection systems aligned with the United Nations Sustainable Development Goals 9 and 12. This study developed and evaluated a real-time anomaly detection and classification system using the YOLOv10-S model. A dataset of common 3D printing defects, compiled from public repositories, was preprocessed and used to train and validate the model under various image conditions, including RGB and grayscale. Paired T-test revealed significant differences between training and validation losses, indicating localization challenges. One-way MANOVA showed that anomaly type influenced detection performance, with spaghetti achieving the highest accuracy and extrusion problems the lowest. RGB images outperformed grayscale, highlighting the advantage of color features in detection. The final model demonstrated strong and consistent performance across validation and testing, confirming its effectiveness for realtime defect detection in FDM 3D printing.