A Deep-Learning-based 3D Defect Quantitative Inspection System in CC Products Surface.

Area of Science:

• Materials Science
• Computer Vision
• Artificial Intelligence

Background:

• Surface quality inspection is critical in continuous casting (CC) production.
• Existing methods struggle with real-time, accurate defect identification in high-temperature environments.

Purpose of the Study:

• To develop an intelligent 3D quantitative inspection strategy for CC production lines.
• To improve the accuracy and efficiency of surface defect detection and delineation.

Main Methods:

• An improved 3D laser image scanning system (3D-LDS) integrating binocular imaging and deep learning.
• Optimization of CCD laser image scanning for high-temperature industrial applications.
• A novel region proposal method for 3D region of interest (ROI) initial depth location to reduce false positives.
• A two-step defect inspection strategy using a fused deep convolutional neural network (CNN) model for classification and delineation.

Main Results:

• The 3D-LDS effectively suppresses redundant candidate bounding boxes caused by pseudo-defects.
• The fused deep CNN model accurately classifies and delineates surface defects.
• The system demonstrates applicability for surface quality inspection of slab, strip, and billet products.

Conclusions:

• The developed 3D-LDS provides an intelligent and effective solution for CC surface inspection.
• The novel defect inspection strategy addresses key challenges in real-time defect analysis.
• This approach contributes to enhanced quality control in steel production.