Mask Gradient Response-Based Threshold Segmentation for Surface Defect Detection of Milled Aluminum Ingot

Area of Science:

• Materials Science
• Computer Vision
• Industrial Automation

Background:

• Surface quality of aluminum ingots is critical for downstream manufacturing processes.
• Automated defect detection is essential for efficient and reliable aluminum production lines.
• Existing methods struggle with the complex surface textures of milled aluminum ingots.

Purpose of the Study:

• To develop a novel, robust, and efficient two-stage defect detection system for milled aluminum ingot surfaces.
• To improve the accuracy and real-time applicability of defect detection in industrial settings.
• To address challenges like class imbalance in defect classification.

Main Methods:

• A novel mask gradient response-based threshold segmentation (MGRTS) for defect extraction.
• Combining MGRTS with Difference of Gaussian (DoG) for improved region of interest (ROI) detection.
• Utilizing an Inception-v3 network with data augmentation and focal loss for defect classification.

Main Results:

• The proposed MGRTS method effectively extracts various defects from complex surfaces.
• The combined MGRTS and DoG approach significantly enhances the defect detection rate.
• The Inception-v3 model achieved high classification accuracy, overcoming class imbalance issues.
• The system demonstrated efficiency and robustness in detecting diverse defects on milled aluminum surfaces.

Conclusions:

• The developed two-stage detection approach is highly effective and robust for identifying defects on milled aluminum ingot surfaces.
• The system successfully meets the accuracy and real-time requirements for application on actual production lines.
• This method offers a significant advancement for quality control in aluminum manufacturing.