Comparative benchmarking of the small and nano versions YOLOv8 and YOLOv11 for FDM 3D printing fault detection
Abstract
Fused Deposition Modelling (FDM) 3D printing is widely used in industrial and consumer applications, but printing failures are often detected only after substantial material and production time have been lost. This study presents a controlled benchmark of four lightweight YOLO object detectors, YOLOv8n, YOLOv8s, YOLOv11n, and YOLOv11s, for in-process FDM fault detection. All models were fine-tuned under identical conditions on 5793 annotated images covering six fault classes. Performance was evaluated on clean test images and under three types of synthetic camera degradation: Gaussian blur, directional motion blur, and additive Gaussian noise. Deployment-related metrics, including parameter count, GFLOPs, ONNX model size, GPU latency, and CPU inference latency, were also measured. On clean images, YOLOv8s achieved the highest [email protected] of 0.7905, while YOLOv11n provided comparable nano-scale accuracy with 14% fewer parameters. Stringing remained the most difficult fault class, with [email protected] values of approximately 0.45–0.48. The robustness experiments showed that blur caused a gradual reduction in detection performance, while directional motion blur was more severe than Gaussian blur. Additive Gaussian noise produced the largest performance degradation. Across the extended degradation grid, YOLOv11n achieved the lowest Degradation Sensitivity Index (DSI) of 28.80%, and also provided the fastest CPU inference at 48.11 ms/image and the smallest model footprint at 10.61 MB. These results provide a practical benchmark for selecting lightweight YOLO models and designing robust FDM monitoring systems.
Related
- Toward deployment-oriented benchmarking of intrusion detection systems: integrating threshold dynamics, distribution shift, and interpretability into comparative model evaluation
- A Comparative Review and Benchmarking Evaluation of Step Detection Methods for Longitudinal Gait Assessment
- CDA-YOLO: an improved YOLOv11 model for PCB defect detection
- YOLOv11-AHFE: adaptive high-frequency signal enhancement mechanism for surface detection
- Illumination-guided and reliability-aware YOLOv11 for low-light object detection
BibTeX
@article{hany2026comparative,
title = {{Comparative benchmarking of the small and nano versions YOLOv8 and YOLOv11 for FDM 3D printing fault detection}},
author = {Hesham Hany and Abdelgwad Elashry},
journal = {Scientific Reports},
year = {2026},
month = oct,
doi = {10.1038/s41598-026-73375-0},
url = {https://www.nature.com/articles/s41598-026-73375-0},
note = {CC BY 4.0},
}
RIS
TY - JOUR
TI - Comparative benchmarking of the small and nano versions YOLOv8 and YOLOv11 for FDM 3D printing fault detection
AU - Hesham Hany
AU - Abdelgwad Elashry
PY - 2026
DA - 2026/10/29
JO - Scientific Reports
DO - 10.1038/s41598-026-73375-0
UR - https://www.nature.com/articles/s41598-026-73375-0
AB - Fused Deposition Modelling (FDM) 3D printing is widely used in industrial and consumer applications, but printing failures are often detected only after substantial material and production time have been lost. This study presents a controlled benchmark of four lightweight YOLO object detectors, YOLOv8n, YOLOv8s, YOLOv11n, and YOLOv11s, for in-process FDM fault detection. All models were fine-tuned under identical conditions on 5793 annotated images covering six fault classes. Performance was evaluated on clean test images and under three types of synthetic camera degradation: Gaussian blur, directional motion blur, and additive Gaussian noise. Deployment-related metrics, including parameter count, GFLOPs, ONNX model size, GPU latency, and CPU inference latency, were also measured. On clean images, YOLOv8s achieved the highest [email protected] of 0.7905, while YOLOv11n provided comparable nano-scale accuracy with 14% fewer parameters. Stringing remained the most difficult fault class, with [email protected] values of approximately 0.45–0.48. The robustness experiments showed that blur caused a gradual reduction in detection performance, while directional motion blur was more severe than Gaussian blur. Additive Gaussian noise produced the largest performance degradation. Across the extended degradation grid, YOLOv11n achieved the lowest Degradation Sensitivity Index (DSI) of 28.80%, and also provided the fastest CPU inference at 48.11 ms/image and the smallest model footprint at 10.61 MB. These results provide a practical benchmark for selecting lightweight YOLO models and designing robust FDM monitoring systems.
SN - 2045-2322
ER -