Design and development of latex mark visual detection system
Artificial intelligence has experienced notable growth and plays a significant role nowadays. A latex former or mould can cause uneven pickup of the latex. As a result, the latex former can produce defective gloves in the production line, which causes a low passing rate in output and wastage. The ma...
Saved in:
| Main Authors: | , , , |
|---|---|
| Format: | Article |
| Language: | en |
| Published: |
UiTM Press
2025
|
| Subjects: | |
| Online Access: | https://ir.uitm.edu.my/id/eprint/122917/1/122917.pdf https://ir.uitm.edu.my/id/eprint/122917/ https://jmeche.uitm.edu.my/ |
| Tags: |
Add Tag
No Tags, Be the first to tag this record!
|
| Summary: | Artificial intelligence has experienced notable growth and plays a significant role nowadays. A latex former or mould can cause uneven pickup of the latex. As a result, the latex former can produce defective gloves in the production line, which causes a low passing rate in output and wastage. The manual latex ark former defect detection that utilises human resources is a temporary solution, as it is time-consuming and comes with a high human error. The proposed latex visual detection system uses artificial intelligence technologies to provide reliable detection of latex mark defects on the latex former. The vital parts in developing a highly efficient latex former defect detection model include designing a mechanical frame structure and developing and evaluating a deep learning model. One of the focuses is to apply the You Only Look Once fifth version (YOLO v5) model and investigate the performance of other versions of YOLO, average loss, and mean average precision (mAP) performance. For the validation, quality inspections are also conducted using the Acceptance Quantity Limit (AQL) standard with 315 sampling sizes in each trial run of the system, and the inspection is rejected with a maximum of 15 pieces of false rejection. In conclusion, the YOLO v5 model is used. With the 14 stages of algorithm development, including tagging and training, the YOLO v5 model achieved an average loss of 5.2% and mAP performance of 99.3% accuracy, achieving the AQL 2.5 standard with less than 15 pieces of false detection. |
|---|