We found a match
Your institution may have access to this item. Find your institution then sign in to continue.
- Title
Toward the Decarbonization of the Steel Sector: Development of an Artificial Intelligence Model Based on Hyperspectral Imaging at Fully Automated Scrap Characterization for Material Upgrading Operations.
- Authors
de la Peña, Borja; Iriondo, Ander; Galletebeitia, Aitzol; Gutierrez, Aitor; Rodriguez, Josué; Lluvia, Iker; Vicente, Asier
- Abstract
Due to decarbonization commitment made by steelmaking companies, the steel industry is tackling a technological transition from blast furnace (BF)–basic oxygen furnace (BOF) route to direct reduction iron (DRI)–electric arc furnace (EAF) route. Under this scenario, ferrous scrap becomes a critical factor for reaching CO2 reduction challenge. However, ferrous scrap can be considered one of the most complex industrial raw materials. In addition, scrap presents a huge heterogeneity in both physical and chemical characteristics. However, for producing high‐quality steel products, certainty on scrap specifics is required. Herein, an artificial intelligent model based on spectral information for the segmentation of different materials contained in the ferrous scrap is proposed. Developed solution offers a processing pipeline through a 2D–3D convolutional neural network algorithm based on a dataset with more than 428 million of pixels through hyperspectral cameras in the 400–1700 nm range. By this model, the detection of ferric fraction, stainless steel, aluminum, zinc, copper, sterile, and rubber and plastic materials are assessed. This work aims at increasing the reliability of the steelmaking process by lowering the number of steel quality noncompliance rejection due to lack of knowledge and uncertainties of these raw material compositions.
- Subjects
CONVOLUTIONAL neural networks; ARTIFICIAL intelligence; SCRAP materials; BLAST furnaces; CARBON dioxide mitigation; BASIC oxygen furnaces
- Publication
Steel Research International, 2023, Vol 94, Issue 11, p1
- ISSN
1611-3683
- Publication type
Article
- DOI
10.1002/srin.202200943