Machine-Learning Algorithms for Process Condition Data-Based Inclusion Prediction in Continuous-Casting Process: A...

Area of Science:

• Materials Science and Engineering
• Manufacturing Process Optimization
• Data Science and Machine Learning

Background:

• Continuous casting is crucial for steel slab production, demanding robust quality control.
• Intelligent manufacturing and data-driven techniques offer advanced solutions for process monitoring.
• Predicting defects like slag inclusions is vital for maintaining high-quality steel products.

Purpose of the Study:

• To develop and evaluate a machine learning algorithm for predicting slag inclusion defects in continuous casting.
• To leverage process condition sensor data for defect prediction.
• To identify the most effective machine learning model for this quality control task.

Main Methods:

• Analysis of a large dataset comprising sensor data from approximately 7300 casting samples.
• Application of Empirical Mode Decomposition (EMD) for processing multi-modal time series data.
• Comparative evaluation of various machine learning algorithms including K-Nearest Neighbors, Support Vector Classifiers, Decision Trees, Random Forests, AdaBoost, and Artificial Neural Networks.
• Implementation of over-sampling and under-sampling techniques to address imbalanced data distribution.

Main Results:

• The optimized Random Forest algorithm demonstrated superior performance compared to other evaluated machine learning models.
• The Random Forest model achieved high recall and ROC AUC scores, indicating effective prediction of slag inclusion defects.
• The study successfully identified a data-driven approach for improving quality control in continuous casting.

Conclusions:

• Machine learning, particularly optimized Random Forests, offers a powerful tool for predicting slag inclusion defects in continuous casting.
• The developed algorithm provides valuable insights for real-time quality control and process optimization in steel manufacturing.
• Data-driven defect prediction enhances the efficiency and reliability of the continuous casting process.