Neural Network Applications in Steelmaking Processes

The integration of neural networks into steelmaking has revolutionised process control, quality assurance and energy efficiency across primary steelmaking routes. These data-driven models are deployed in basic oxygen furnaces and electric arc furnaces to predict critical endpoints—such as carbon, phosphorus and temperature—enabling dynamic adjustments in blowing patterns and power input. Spectral feature extraction from flame emissions, combined with rough set reduction and optimisation algorithms, allows high-precision inference of molten steel properties. Hybrid approaches that fuse neural architectures with first-principle thermodynamic models have delivered robust forecasts of oxygen consumption and slag chemistry. Advances in recurrent and convolutional networks now underpin adaptive control systems responsive to feedstock variability and operational disturbances. This convergence of machine learning and steelmaking has enhanced throughput, improved product uniformity and supported decarbonisation by optimising energy use and minimising material waste on a global scale.

Research from Nature Portfolio

Recent studies have employed spectral feature extraction and genetic-algorithm-optimised attribute reduction to identify key wavelengths for predicting converter temperature and carbon content. A back-propagation neural network trained on eight representative spectral indices achieved accuracies above 99% for temperature and carbon predictions, with average errors below 4 K and 0.02% respectively. Mutual information analysis confirmed the universality of the selected features across different operating conditions, demonstrating the model’s robustness for real-time process monitoring.

Research from all publishers

A comprehensive review of machine learning in basic oxygen furnace steelmaking categorised developments into static, dynamic and intelligent prediction stages, highlighting back-propagation networks and hybrid metaheuristic-enhanced models for end-point carbon and phosphorus estimation. Empirical comparisons between linear regression and neural networks for phosphorus endpoint prediction identified oxygen flow and sub-lance phosphorus concentration as the most influential variables, with neural models delivering superior accuracy.

In stainless electric arc furnaces, support vector regression with a radial basis function kernel has been applied to real-time tap temperature prediction and control. Deployment of this model yielded a 17% reduction in temperature deviation and a 282 kWh-per-heat saving in power consumption, with an internal rate of return of 35.8%.

A principal component analysis–genetic algorithm–back propagation neural network has been developed for simultaneous prediction of end-point phosphorus and oxygen contents in basic oxygen furnaces. PCA removed input collinearities, while GA optimised network weights, producing root-mean-square errors of 0.0015% for phosphorus and 0.0049% for oxygen—levels of precision aligned with modern process control requirements.

Neural Network Applications in Steelmaking Processes publication trend

The graph below shows the total number of articles in neural network applications in steelmaking processes across all publications each year (not limited to Nature Index journals).

Technical terms

Basic Oxygen Furnace (BOF): A primary steelmaking vessel in which high-purity oxygen is blown through molten iron to reduce carbon content and refine chemistry.

Electric Arc Furnace (EAF): A reactor that melts scrap or direct-reduced iron using electrical arcs, offering rapid cycle times and lower carbon emissions.

Back Propagation (BP) Neural Network: A multilayer feedforward network trained by adjusting connection weights via gradient descent to minimise prediction error.

Support Vector Regression (SVR): A kernel-based machine learning method that fits a regression hyperplane within a defined error margin around training data.

Principal Component Analysis (PCA): A statistical technique for reducing data dimensionality by transforming correlated variables into orthogonal components capturing maximum variance.

Long Short-Term Memory (LSTM): A recurrent neural network unit engineered to capture long-range dependencies and temporal patterns in sequential data.

References

1. Toward learning steelmaking—A review on machine learning for basic oxygen furnace process. Materials Genome Engineering Advances (2023).
2. Comparison Between Empirical Strategies for Predicting Endpoint Phosphorus Content in BOF Steelmaking Process. Archives of Advanced Engineering Science (2025).
3. Research on prediction model of converter temperature and carbon content based on spectral feature extraction. Scientific Reports (2023).
4. Machine Learning-Based Tap Temperature Prediction and Control for Optimized Power Consumption in Stainless Electric Arc Furnaces (EAF) of Steel Plants. Sustainability (2023).
5. Gaussian Process-Based Hybrid Model for Predicting Oxygen Consumption in the Converter Steelmaking Process. Processes (2019).
6. Prediction model of BOF end-point P and O contents based on PCA–GA–BP neural network. High Temperature Materials and Processes (2022).