Cross-Factory Polarizer Sheet Surface Defect Inspection System Based on Multiteacher Knowledge Amalgamation

To ensure the quality control of the polarizer sheet, the surface defect inspection plays a pivotal role as the final step in the production process. Surface defect inspection of polarizer using auto-optical inspection (AOI) poses several challenges, including the difficulty of capturing clear and f...

Full description

Saved in:
Bibliographic Details
Published in:IEEE transactions on instrumentation and measurement 2024, Vol.73, p.1-16
Main Authors: You, Mingyu, Ren, Baiyu, Han, Xuan, Zhou, Hongjun
Format: Article
Language:English
Subjects:
Adaptation models
Amalgamation
Assembly lines
Computational modeling
Data models
Defect detection
Domain different
End user training
End users
Error reduction
Factories
feature fusion
Harnesses
Industrial plants
Inspection
knowledge amalgamation
Optical data processing
polarizer sheet defect auto-optical inspection (AOI)
Polarizers
Production facilities
Production lines
Quality control
selection strategy
Surface defects
task aligned
Task analysis
Citations: Items that this one cites
Online Access:Get full text
Tags: Add Tag
No Tags, Be the first to tag this record!
Description
Summary:To ensure the quality control of the polarizer sheet, the surface defect inspection plays a pivotal role as the final step in the production process. Surface defect inspection of polarizer using auto-optical inspection (AOI) poses several challenges, including the difficulty of capturing clear and flat images of the polarizer, the presence of residue and dirt on the production line, which increases classification complexity and variations in data across different production lines and factories, limiting model generalizability. To address these issues, this article presents a polarizer sheet AOI equipment. With this equipment, we can capture images that are ideal for inspection purposes. In addition, we introduce a novel approach that incorporates task-aligned and domain-adapted multiteacher knowledge amalgamation for the first time based on the captured images. Specifically, we introduce two selection strategies: one leveraging the selection strategy based on judgment of student model (Stu-MKA) and the other leveraging the selection strategy based on attention mechanism (Atn-MKA). These strategies guide the amalgamation of knowledge from diverse models, yielding a student model that harnesses the strengths of each teacher model, trained on disparate data domains. This approach effectively addresses the challenge of limited annotated samples in cross-factory training and the pronounced disparities in cross-domain data, culminating in commendable overall performance. The AOI equipment has been successfully applied to multiple production lines in multiple factories. It not only improves production capacity but also reduces human detection and operation errors, and the defect inspection results have received unanimous praise from end users. Our implementation is available at: https://github.com/Herrera21-a/MTKA .
ISSN:0018-9456
1557-9662
DOI:10.1109/TIM.2024.3417596