Answers are grounded in this paper and cite their sources.
Evaluation of transformer models improves defect detection in injection molding, suggesting a move away from expert reliance.
The startup process in injection molding, traditionally reliant on expert personnel, has become critical due to small-batch production and a shortage of skilled workers. This shift requires efficient quality monitoring, especially for visually demanding plastic products. This study evaluates transformer-based computer vision models like RT-DETR for detecting defects characterized by domain-specific patterns and high intra- and inter-class variability. The research compares general models with specialized models trained for specific component-defect combinations, focusing on key computer vision metrics and computational demands. Injection molding features such as representative dimensions and aspect ratio are analyzed to assess knowledge transferability from pre-trained models to new scenarios. The findings could enhance defect detection accuracy, reduce reliance on experts, and boost production efficiency in small-batch, visually critical manufacturing.
No takes yet. Share an insight, caveat, or question.
Andrea Pieressa (2025) studied this question.
Synapse has enriched 10 related papers. Consider them for comparative context: