Intelligent Inspection Method and System of Plastic Gear Surface Defects Based on Adaptive Sample Weighting Deep...

Area of Science:

• Materials Science
• Computer Vision
• Manufacturing Engineering

Background:

• Plastic gears commonly have surface defects after injection molding, challenging traditional inspection methods.
• Existing visual inspection systems lack comprehensive defect detection and accuracy.
• There is a critical need for advanced, accurate plastic gear defect inspection solutions.

Purpose of the Study:

• To develop an intelligent algorithm and system for comprehensive plastic gear defect inspection.
• To improve accuracy and capability beyond current single-category inspection methods.
• To address challenges posed by complex defect types and imbalanced datasets.

Main Methods:

• Developed PGD-net (plastic gear defect network) for detecting subtle surface defects.
• Integrated an adaptive sample weighting method with an improved Focal-IoU loss function.
• Incorporated CoordConv layers to enhance model generalization.
• Constructed a dataset of 16 plastic gear defect types for training and testing.

Main Results:

• PGD-net achieved a mean average precision (mAP) of 95.6% across 16 defect types.
• The algorithm effectively captures subtle defect features at arbitrary locations.
• An online inspection system was developed and integrated into production lines.
• The system successfully inspects over 60,000 gears daily.

Conclusions:

• PGD-net offers a robust and accurate solution for plastic gear surface defect inspection.
• The developed online system enables automated quality control and sorting in manufacturing.
• This technology significantly enhances the efficiency and reliability of plastic gear production.