A Novel Method to Inspect 3D Ball Joint Socket Products Using 2D Convolutional Neural Network with Spatial and...

Area of Science:

• Industrial Manufacturing
• Computer Vision
• Artificial Intelligence

Background:

• Product defect inspection is crucial in manufacturing.
• High-precision 3D cameras are costly and slow for real-time inspection.
• Existing methods struggle with the complexity and diversity of industrial products.

Purpose of the Study:

• To develop a cost-effective and efficient real-time product inspection system.
• To improve defect detection accuracy for 3D objects.
• To address the limitations of traditional 3D inspection methods.

Main Methods:

• Utilized five 2D cameras for comprehensive product data capture.
• Developed a 2D convolutional neural network (CNN) incorporating spatial and channel attention (SCA) mechanisms.
• Employed channel attention (CA) to identify relevant feature maps and spatial attention (SA) to pinpoint critical regions.

Main Results:

• The proposed CNN with SCA achieved 98% classification accuracy in experiments.
• Demonstrated high efficiency for real-time product inspection.
• Showcased potential for detecting small product defects.

Conclusions:

• The proposed 2D CNN with SCA offers a viable solution for real-time industrial product inspection.
• This approach provides high accuracy and cost-effectiveness compared to traditional methods.
• The SCA mechanism enhances the model's ability to identify subtle defects.