Continuous casting is an important link in steel production, and its process parameters directly affect the quality of the casting and the economic benefits of enterprises. However, the continuous casting production involves complex nonlinear heat transfer and mass transfer processes, and traditional numerical simulation and laboratory testing optimization methods are inefficient and difficult to meet the demands of modern steel enterprises for efficient production.
Currently, optimization methods for continuous casting production mainly rely on traditional physical models, which cannot handle the multivariable nonlinear coupling relationships. In addition, due to insufficient data engineering capabilities, enterprises find it difficult to effectively utilize production data, leading to resource waste.
Recently, Professor Zhou Lejun and his team from Central South University conducted a systematic analysis of the application of machine learning in the continuous casting process, summarizing the latest progress of machine learning in anomaly prediction, quality detection, and process optimization. They proposed an intelligent production framework based on machine learning, significantly improving casting quality and production efficiency. The research also focused on how to combine physical models with machine learning methods to create a hybrid prediction framework to enhance the model’s reliability and generalization ability.
The relevant research results were published in the “China Metallurgy” 2024, Vol. 34, No. 11, titled “Research Progress of Machine Learning in Continuous Casting Process,” with authors Zhou Lejun, Wang Wanlin, Ji Yi*, and Chen Jiaxi.
Research results and conclusions:
-
Machine learning methods can significantly improve the prediction accuracy of continuous casting anomaly events, including key issues such as sticking, steel leakage, nozzle blockage, and liquid level fluctuations.
-
A strategy combining visual inspection with process parameter prediction addresses the problems of low efficiency and high subjectivity of manual inspection.
-
A hybrid framework combining machine learning with physical models was proposed, allowing the model to demonstrate higher stability and flexibility in continuous casting anomaly forecasting, casting defect detection, and process optimization.
-
Data-driven optimization strategies significantly enhance the efficiency of adjusting continuous casting process parameters and reduce uncertainty in the production process.
There are a total of 4 images in this paper, some of which are shown below:
Figure 1. Feature extraction based on the geometric characteristics and propagation characteristics of the sticking area
Figure 2. Visualization of the blockage index of steel grades
Figure 3. Prediction method based on the time series of liquid level fluctuations in the crystallizer
Figure 4. Techniques used in the image acquisition and preprocessing stages
Figure 5. Rapid detection process for casting corner cracks based on the YOLOv5-SFA algorithm
-
If you need the original paper, please leave a message “2501” at the bottom of this page. If you have any specific literature materials you would like to see, please also leave a message to the editor.
-
You can click on “#Continuous Casting” below and then “Subscribe” to this collection, and the system will more accurately recommend related information to you.
-
Welcome to submit, and feel free to “like, view“!