Related Experiment Video
Updated: Aug 5, 2026
05:47
Simulation of a Scaled Assembly Process with Collaboration of a Robotic Arm and Monitoring through a Vision System for Quality Control
Published on: August 29, 2025
Deep learning based load station inspection for smart manufacturing with limited data
Yasir Ijaz1, Sonya Coleman1, Dermot Kerr1
1School of Computing, Engineering and Intelligent Systems, Ulster University, Northland Road, Londonderry, BT48 7JL UK.
The International Journal, Advanced Manufacturing Technology
|August 3, 2026
Summary
YOLOv8 demonstrates superior data efficiency for industrial Load Station inspection, requiring fewer training samples than YOLOv5. Smaller YOLOv8 variants excel in data-scarce environments, offering reliable inspection with limited datasets.
Area of Science:
- Deep Learning
- Computer Vision
- Manufacturing Automation
Background:
- Industrial Load Station (LS) inspection is critical for semiconductor wafer handling.
- Data scarcity poses a significant challenge for deploying deep learning in manufacturing.
- Ensuring proper LS alignment and occlusion-free conditions is vital for pin pack assembly.
Purpose of the Study:
- To investigate the impact of training data size on deep learning inspection reliability.
- To determine minimum data requirements for effective LS inspection in smart manufacturing.
- To compare the data efficiency of YOLOv5 and YOLOv8 variants in a real-world industrial setting.
Main Methods:
- Utilized YOLOv5 and YOLOv8 variants across five training set sizes.
- Employed 5-fold cross-validation with multiple random seeds for evaluation.
- Conducted an ablation study on combining real and synthetic obstruction samples.
Main Results:
- YOLOv8 exhibited superior data efficiency, achieving high accuracy (0.981 ± 0.014) and mAP@0.5 (0.842 ± 0.050) with only 40 samples per class (T-40).
- Smaller YOLOv8 variants (YOLOv8n, YOLOv8s) outperformed larger models in data-scarce conditions.
- Combining real and synthetic obstruction samples was essential for reliable inspection.
Conclusions:
- YOLOv8 offers a more data-efficient solution for industrial LS inspection compared to YOLOv5.
- Smaller YOLOv8 models are recommended for deployment in data-limited smart manufacturing environments.
- The study provides statistically grounded recommendations for deep learning deployment with limited data.