Recognition of Converter Steelmaking State Based on Convolutional Recurrent Neural Networks

The converter steelmaking process is an important part of metallurgical production, and the flame characteristics at the furnace mouth indirectly reflect the smelting conditions inside the furnace. Effectively recognizing and predicting the smelting conditions of converter steelmaking is a challengi...

Full description

Saved in:
Bibliographic Details
Published in:Metallurgical and materials transactions. B, Process metallurgy and materials processing science Process metallurgy and materials processing science, 2024-06, Vol.55 (3), p.1856-1868
Main Authors: Huang, Chengyong, Dai, Zhangjie, Sun, Ye, Wang, Zijiao, Liu, Wei, Yang, Shufeng, Li, Jingshe
Format: Article
Language:English
Subjects:
Accuracy
Characterization and Evaluation of Materials
Chemistry and Materials Science
Feature recognition
Industrial applications
Materials Science
Metallic Materials
Metallurgy
Nanotechnology
Neural networks
Original Research Article
Production methods
Recurrent neural networks
Smelting
Steel converters
Steel making
Structural Materials
Surfaces and Interfaces
Thin Films
Citations: Items that this one cites
Online Access:Get full text
Tags: Add Tag
No Tags, Be the first to tag this record!
Description
Summary:The converter steelmaking process is an important part of metallurgical production, and the flame characteristics at the furnace mouth indirectly reflect the smelting conditions inside the furnace. Effectively recognizing and predicting the smelting conditions of converter steelmaking is a challenging and critical issue in industrial production. However, traditional image-based methods using a single static flame image as input have low recognition accuracy and cannot accurately reflect changes in smelting conditions. To address this problem, a new recognition model is proposed in this study, which first preprocesses the flame video sequences at the furnace opening, and then applies a convolutional recurrent neural network (CRNN) to further learn the spatio-temporal features and obtain recognition results. In addition, in order to further improve the accuracy of the model, we introduced the channel attention mechanism and verified the effectiveness of the model through the feature layer visualization technique. In addition we quantitatively evaluate the model performance by accuracy, precision, recall, and F 1-score, and plot the confusion matrix with AUC–ROC curves. The experimental results show that the method is not only effective but also robust and has a large potential for industrial applications.
ISSN:1073-5615
1543-1916
DOI:10.1007/s11663-024-03071-9