Image-Segmentation-Guided Fragmentized Steel Scrap Tramp Material Characterization

Area of Science:

• Metallurgical Engineering
• Materials Science
• Computer Vision

Background:

• The UK steel industry is shifting to electric arc furnaces, increasing the need for high-quality steel scrap.
• Copper and other tramp elements in scrap significantly impact steel quality.
• Current chemical analysis methods are too costly and impractical for widespread scrap screening.

Purpose of the Study:

• To develop a fast, cost-effective method for estimating tramp element composition in steel scrap.
• To investigate the use of image-based classification for copper content analysis in fragmentized steel scrap.

Main Methods:

• A dataset of commercial-grade steel scrap images was created, featuring varied steel and copper mixtures.
• Images were annotated with weight measurements and segmentation data.
• A neural network was trained using image segmentation, with segmentation maps feeding a classifier.

Main Results:

• The image-based classification system achieved 86.67% accuracy in identifying copper composition classes.
• This demonstrates the feasibility of using computer vision for scrap analysis.

Conclusions:

• Computer vision offers a viable solution for rapid and economical steel scrap quality assessment.
• This technology can help manage tramp element levels in the transitioning steel industry.