The heating process of coke ovens belongs to a complex thermal process characterized by “intermittent operation of individual combustion chambers and continuous operation of the entire oven”, which is subject to multiple interfering factors. The traditional manual adjustment mode for setting coke oven temperatures, relying on empirical experience, suffers from long temperature measurement cycles, crude target temperature adjustments, poor stability, and high energy consumption. This paper aims to replace manual temperature adjustment with automated intelligent heating control technology, achieving enhanced production stability, improved coke quality, and reduced energy consumption. To achieve these objectives, the study proposes three core components: a flue temperature prediction model integrating STL time series decomposition and Transformer algorithm (STL-Transformer model),a target flue temperature setting model based on particle swarm optimization algorithm, a flue temperature control model. These models have been implemented in an intelligent coke oven heating control system. Experimental results using real operational data from Nanjing Iron and Steel demonstrate: the flue temperature prediction model achieved a mean absolute error of 1.82 ℃, outperforming comparable algorithms. The target temperature setting model reduced average errors to 2.49 ℃ (machine side) and 2.55 ℃ (coke side), representing 42.23% and 40.56% improvements over manual control respectively. After system implementation at Nanjing Iron and Steel, the intelligent heating control system delivered: 3% reduction in overall coke oven energy consumption, 0.2% improvement in post-reaction coke strength. The system has significantly contributed to production stability enhancement, coke quality improvement, and energy efficiency optimization.

In the production process of continuous casting slab hot delivery and hot charge, the charging temperature is one of the key factors leading to cracks on the surface. The traditional single-point measurement method is susceptible to the interference of oxidized skin, with low measurement accuracy and poor stability, for this reason, this paper proposes a measurement method based on the full-field temperature of the slab surface. For the problems of temperature field deformation and slab sticking, a combination of automatic thresholding and nonlinear interpolation was proposed on the basis of spatial coordinate transformation, which realizes the dynamic tracking of the temperature field and slab segmentation; for the interference problem of iron oxide skin, a morphological expansion method was adopted to restore the temperature of the interfered area of the slab; in order to express the distribution law of the temperature of the surface of the slab, a partitioned characterization of the full-field temperature was proposed, and a specific analysis method was established for the sensitive area that generates quality problems. In order to express the temperature distribution law on the surface of the slab, the full-field temperature zoning characterization was proposed, and the specific analysis method was established for the sensitive areas that produce quality problems. The above methods have been applied in production, and the statistical data show that the temperature measurement accuracy and stability have been improved by about 20-50 ℃; the temperature of slab in the furnace is concentrated at 500-600 ℃, which is in line with the requirements of hot loading temperature; the temperature of 1/4 of the slab is higher than the center temperature, which is a sensitive area for quality defects; and the preliminary finding is that the trend of the temperature of the slab in the furnace has an effect on cracks. The above results provide an important support for optimizing the furnace entry process system and heating furnace temperature control.

The injection of hydrogen-rich fuel in blast furnace is an important measure for blast furnace to achieve green and low-carbon transformation and practice the dual carbon strategy. In order to accurately evaluate the effect of Shougang 2 650 m3 blast furnace injection coke oven gas on the reduction behavior of the comprehensive charge, this paper studies the effects of the hydrocarbon ratio, charge structure, charging mode and reduction temperature on the hydrogen-rich reduction behavior of the blast furnace according to the on-site production data and the national standard iron ore reduction method. The results show that with the increase of H2 ratio from 7% to 23%, the reduction degree of pellets and lumps increases significantly to 94.48% and 94.47%, while the reduction degree of sinter decreases to 88.36% due to carbon separation. The change of the comprehensive charge ratio has little effect on the reduction degree, and the overall reduction degree is stable at about 88%. With the temperature from 700 ℃ to 1 100 ℃, the degree of reduction was significantly improved, among which pellets performed the best (76.57% to 91.09%), followed by sinter (75.83% to 90.23%), lump ore was slightly lower (75.4% to 90%), and the reduction degree of comprehensive charge was the highest at 91.7%. The alkaline pellets have stable structure and excellent reduction performance at 900 ℃, and the surface begins to bond at 1 000 ℃, and the bond and crack intensification at 1 100 ℃ lead to a significant decrease in strength and performance. The reduction efficiency of the coke ore charging method was the highest, and the reduction degree of the comprehensive charge was 92.43%, which was significantly better than that of coke ore (86.61%) and coke ore blending (89.76%). The results of this study provide some data reference for the study of the influence of hydrogen enrichment in Shougang 2 650 m3 blast furnace on the reduction behavior of comprehensive charge.

The ironmaking-steelmaking interface, as a crucial link in the steel production process, has a direct impact on the stability of production rhythm and resource utilization efficiency through dynamic scheduling and optimization. Iron transport scheduling is essential for the smooth and efficient operation of the iron-steel interface, with scheduling efficiency and rationality directly affecting key indicators such as molten iron temperature drop. This paper proposes an iron transport scheduling decision-making technology tailored to the “one-ladle-to-the-end” model adopted in Tangsteel New Area. The technology includes optimization of molten iron ladle grouping and transport planning, locomotive task allocation rules, and transport task path planning. It aims to ensure tapping safety and meet steelmaking time requirements by optimizing ladle grouping and task allocation while considering locomotive load and real-time positioning constraints. Additionally, it features dynamic adjustment capabilities under abnormal conditions, allowing for real-time optimization of transport plans in response to equipment failures, schedule changes, and other unexpected situations, thereby ensuring production continuity. Practical applications in Tangsteel New Area have demonstrated that this technology significantly improves molten iron transport efficiency, path conflict avoidance rates, and overall scheduling performance, providing practical support for the intelligent and green development of the steel industry.

The online shear optimization method for medium and thick plates is an important factor affecting shear quality and production efficiency. The traditional method uses a single shape feature index or a fitting reference line to determine the shape of the steel plate and provide a cutting strategy. Usually, only a single solution was calculated or a solution was misjudged as no solution, resulting in low tolerance.The shear optimization method based on Support Vector Machine (SVM) considers the contour points on both sides of the rolled plate with head and tail removed as two categories of SVM, and then calculates hyperplane and maximum interval. The hyperplane and the head-tail shear lines form a maximum shearable parallelogram. By using analytical methods, calculate the shear optimization solutions within all parallelograms to form a solution space, and then select typical solutions in accordance with order requirements. Finally, a comparative experiment was conducted with 419 orders. The proposed method reduced invalid solutions to zero and decreased the proportion of unsolvable cases by over 5%, verifying the effectiveness and accuracy of the approach.

In the steelmaking production of metallurgical enterprises, plant operational efficiency directly affects the smoothness of production logistics and the coordination between processes. This study aims to improve the efficiency of overhead crane scheduling in steelmaking production of metallurgical enterprises, and innovatively proposes a scheduling method that integrates Deep Reinforcement Learning (DRL) and simulation. By constructing a simulation model to replicate the crane operating environment and designing a DRL algorithm to process real-time spatial information, this study can formulate optimized scheduling plans for uncertain transportation tasks. Training with historical data enables the model to make optimal decisions under varying conditions. Through simulation modeling technology, dynamic optimization and real-time monitoring of scheduling strategies are realized. The reward function, as a key metric, is used to monitor on a temporal dimension, significantly enhancing the intelligence and operational efficiency of the scheduling process. Experimental results indicate that the A2C method proposed in this study shows significant advantages in scheduling efficiency and task completion time. The algorithm′s final cumulative reward gap value is significantly better than other reinforcement learning methods (gap=7.89%) and traditional methods (gap more than 100%).

This paper presents a method for predicting electricity consumption in the production process of steel enterprises using TabNet-XGBoost, aiming to improve prediction accuracy and optimization efficiency. Preprocessing techniques such as data augmentation and outlier handling are employed to enhance model performance. By leveraging the interpretability and feature selection capabilities of the TabNet model, this study identifies key influencing factors for feature selection. Furthermore, Bayesian optimization and grid search methods were applied to fine-tune the hyperparameters of eXtreme Gradient Boosting (XGBoost), thereby further enhancing the model′s effectiveness. Comparative experiments are also conducted using machine learning models including Support Vector Regression (SVR), Random Forest (RF), Gradient Boosting Decision Tree (GBDT), and Light Gradient Boosting Machine (LightGBM). The experimental results indicate that CO2 emissions, lagging reactive power, Number of Seconds from Midnight (NSM), weekly status, and specific days of the week are identified as critical predictors. These factors reflect indirect indicators of electricity usage, the efficiency of the power system, potential waste during non-working hours, and cyclical patterns of production activities. The predictive performance of the model was evaluated using Mean Absolute Error (MAE), Root Mean Squared Error (RMSE), and Mean Absolute Percentage Error (MAPE). The results show that the TabNet-XGBoost model achieves an MAE of 0.326, RMSE of 1.097, and MAPE of 1.032 on the test set, representing a noticeable improvement in prediction accuracy compared to traditional methods. In summary, the proposed model offers significant advantages in addressing the challenge of electricity consumption prediction in the steel industry′s production process, providing new research insights and technical solutions for related fields.

With the rapid development of digital transformation in China′s steel industry, steel enterprises have accumulated massive knowledge and data assets. How to efficiently unearth the value of knowledge and data assets and gradually transform from digitalization to intelligentization has become a challenging problem. Large model has entered the stage of large-scale application, and the industry large-scale model is the key to its deep penetration into vertical fields. As a typical process industry, the steel industry has abundant scene resources and data assets, and urgently needs to be empowered by industry big models to create a new business model driven by the integration of knowledge, data, and intelligence, in order to achieve intelligent upgrading and high-quality development. This article first proposes the architecture design concept of the steel industry′s large model, and studies the data architecture, platform architecture, and application architecture; Then, the application of knowledge engines, intelligent agents for deep knowledge insight reports, metallographic detection models, and embodied intelligence models were introduced, exploring the application modes of natural language models, visual models, and multimodal models; Finally, the future development of the steel big model was discussed from the aspects of industry data space, collaborative system of large and small models, and application security protection.

Post-rolling cooling is a crucial method for regulating the microstructure and properties of ho-rolled steel strips, where the heat transfer coefficient serves as a key parameter in temperature models, directly determining the accuracy of coiling temperature control. However, traditional mechanism models have limited accuracy in calculating the water-cooling heat transfer coefficient, making it difficult to meet the control requirements for low-temperature coiling of X80 pipeline steel, which results in significant temperature prediction deviations. To address this issue, a temperature prediction model integrating data-driven and mechanism approaches was proposed. Based on heat transfer theory, the model calculates the internal heat conduction of the steel strip by solving a one-dimensional heat conduction differential equation. At the same time, a data-driven model was constructed to predict the water-cooling heat transfer coefficient, improving the accuracy of surface heat transfer calculations. In developing the data-driven model, the effectiveness of artificial neural networks, gradient boosting decision trees, and support vector regression in predicting the heat transfer coefficient was compared, and the optimal machine learning algorithm was selected to construct the integrated model. Experimental results show that, compared to the mechanism model, the integrated model based on artificial neural network prediction of the heat transfer coefficient reduces the coiling temperature prediction error by 5 ℃, with the mean squared error and mean absolute percentage error decreasing by 76.8% and 49.5%, respectively, significantly improving the accuracy and stability of coiling temperature prediction. This integrated model effectively compensates for the shortcomings of the mechanism model in heat transfer coefficient prediction, providing a feasible solution for the precise control of coiling temperature in hot-rolled steel strips.

Against the backdrop of the intelligent transformation of the global steel industry, Digital Technology (DT) has become the core driving force for the intelligent upgrading of Hot Strip Mill (HSM) production lines. This paper systematically analyzes the application status and development trends of DT in wide strip Hot Strip Mill (HSM) production lines. By integrating key technologies such as the Industrial Internet of Things (IIoT), big data analysis, digital twin, and artificial intelligence, it explores the application paths and innovative achievements in process control optimization, intelligent equipment operation and maintenance, quality closedloop control, energy efficiency improvement, and integrated coordination of production and marketing. Taking the 1 780 mm HSM production line as the research object, this paper summarizes the role of digital technology in promoting the production efficiency, product quality, and cost control of HSM: the production efficiency of hotrolling technical personnel has increased by more than 30%, and the comprehensive yield has reached 97.02%. Finally, the challenges and future development directions of DT are proposed. The research shows that the deep integration of digital technology and HSM production lines will promote the hot strip rolling process towards intelligence and greenization.

In the steel continuous casting process, due to dynamic changes in production conditions, such as steel grade, casting speed, and temperature, single models exhibit poor adaptability in detecting surface defects of continuous casting slabs, and often fail to flexibly adapt to changes in different production environments, and tend to lose accuracy in complex practical operations, thereby leading to frequent false positives and missed detections, severely affecting the quality control of continuous casting slabs. To address these challenges, this study proposes a data-driven slab defect detection method based on multi-model fusion. First, a Gaussian Mixture Model was employed to cluster industrial process data, effectively distinguishing samples from different distributions. Next, local anomaly detection sub-models were constructed using autoencoders for each distribution. Control limits and reconstruction errors are determined for each sub-model. Finally, Bayesian inference was used to fuse the detection results of the local sub-models, enabling global anomaly detection in complex multi-condition environments. Using data from the continuous casting process of a certain steel manufacturing company, the proposed method is validated through comparative analyses of the detection performance of various models. The results demonstrate that the proposed approach achieves superior performance, effectively reducing both false negative and false positive rates. This method can be extended to other industrial data analysis and modeling scenarios, offering significant reference value for improving product quality through industrial data utilization.

This paper focuses on the innovative applications and development trends of digital twin technologies in the field of metallurgical engineering. As core components of Industry 4.0, digital twin demonstrates enormous application potential and broad prospects in the steel industry. Firstly, this paper reviews the definition, background, and core development technologies of digital twin technology, and expounds its significance in the metallurgical field. Secondly, based on the production process of the steel manufacturing system, it systematically combs the practical paths and application scenarios of digital twin technology in the metallurgical field. Finally, from a multi-dimensional perspective, it analyzes the technical bottlenecks in the industrial implementation process of steel digital twin, and based on the characteristics of the industry and technological development, prospects its future development direction. It is worth noticing that this article tracks the latest progress of cutting-edge technologies such as generative artificial intelligence algorithms, large language models, and intelligent agents, and analyzes the integration potential of such technologies with digital twin technology at different levels, providing a theoretical direction for the construction of the next-generation metallurgical digital twin system with intelligent cognition.

Aiming at the issues of low switching frequency and significant current harmonics in standalone medium-voltage high-power grid-tied converters, a coordinated control method for multiple grid connected inverters on the continuous rolling grid side based on variable control period digital phase-locked technology was proposed. The distributed control units of each grid-tied converter sample the utility grid voltage as the synchronization signal. Through a digital phase-locked loop with a variable control period, they divide each fundamental wave cycle of the grid voltage synchronization signal into several segments and assign numbers to them. Subsequently, based on the corresponding segmentation points within the fundamental wave cycle corresponding to the sequence number of the grid-tied converter, triangular carrier increment and decrement counting was activated. This facilitates uniform phase shifting of the triangular carriers for each grid-tied converter, eliminating the need for additional interconnection lines between controllers. The correctness and effectiveness of the proposed control strategy were verified based on the on-site operational results of a 1 850 mm strip hot rolling production line at a specific steel plant.

In industrial inspection, duplicate detection of X-ray weld images is crucial to prevent whole image or local area forgery and evaluate training sample consistency. Existing methods primarily address whole image duplication. However, it is difficult for them to detect tampering in specific regions. To address this problem, this paper propose a duplicate detection method based on segmentation of the weld bead and parent material regions in X-ray weld images. First, the gray-scale distribution characteristics of the weld bead and parent material regions were used to adaptively determine the boundary of the weld bead by comparing the actual gray-scale curve and the fitted gray-scale curve, and as a result, the weld bead and parent material regions were divided. Then, the perceptual hashing algorithm with high computational efficiency and robustness to small changes in brightness and contrast was used to calculate the perceptual hash values of the weld bead and the parent material regions respectively. Finally, the similarity between the regions was evaluated by calculating the Hamming distance of the perceptual hash values of the weld bead and the parent material regions of the two compared images, and the duplicate detection was determined by a given threshold. And the cross-area weighted fusion strategy can be also used to evaluate the whole image similarity. The proposed detection method was applied to test the tampered images in the duplicate detection experimental dataset constructed based on the GDXray dataset, with the detection rate serving as the evaluation metric. Experimental results demonstrate that the proposed method achieves superior detection accuracy, with an average detection rate of 95%, in scenarios such as region replacement, region modification, and composite tampering.

Continuous processing, inaccessible internal states, pronounced lag between actions and quality outcomes, and rare-yet-costly failure samples distinguish the manufacturing of metallic materials from discrete manufacturing and dictate that industrial AI for steel rolling cannot remain at the level of prediction and generation, but must advance toward a goal-driven closed loop of perception, inference, decision-making, execution, and feedback. Taking goal-drivenness as the central thread, this paper analyzes the capability leap of artificial intelligence from generative AI to agentic AI and Physical AI, and proposes explicitly introducing “goal-driven” into the “data-intensive-intelligence-emergent-human-machine collaborative” paradigm to form a new paradigm for industrial control. A Vision-Memory-Controller (VMC) functional loop is used to characterize the core capabilities of the world model, clarifying its relationship to and distinction from digital twins, embodied intelligence, and Physical AI; three technical routes—full-stack-supported, spatial-simulation-based, and abstract-prediction-based—are summarized together with their bottlenecks. On this basis, a rolling industrial AI system architecture is constructed, in which the cloud-based world model provides global “Knowing,” domain-specific agents together with deterministic control provide field-level “Acting,” and full-stack vertical integration provides the engineering foundation. Given the marked differences among microstructure/properties, geometric dimensions, and shape quality in measurability, mechanism maturity, and execution complexity, three differentiated modeling paths are proposed: “data first—knowledge empowered,” “mechanism led—data compensated,” and “problem driven—virtual-physical integrated.” Representative engineering cases show that the relevant technologies have formed verifiable local closed loops within individual domains and are accelerating toward cross-process, multi-objective coordinated full-process autonomous decision-making, providing a theoretical foundation and engineering pathway for the intelligent transformation of metal industries such as iron and steel, as well as other large-scale complex process industries.

Slab number is a critical identifier in steel manufacturing for process tracking and intelligent logistics. However, poor print quality and harsh industrial environments often damage image data, seriously reducing the accuracy of slab number recognition. To improve the recognition capability of damaged images, a novel algorithm—Feature Recognition Inference Network (FRI-Net)—which combines damaged region detection, contextual feature reasoning, and attention mechanism to achieve high-quality restoration and accurate recognition of degraded slab number images. FRI-Net adopts a modular architecture, introduces a feature feedback optimization mechanism and Knowledge Consistent Attention (KCA), and significantly enhances the restoration capability for complex defective regions. Experimental results on multiple public and industrial datasets demonstrate that FRI-Net outperforms existing mainstream methods in recognition accuracy and fault tolerance, effectively enhancing the stability and intelligent level of slab tracking systems.

To address the challenges of infrequent detection and prolonged measurement cycles for FeO content in iron ore sinter, this study proposes a soft-sensor model for online estimation of FeO content using routinely monitored parameters from the sintering process as input variables. The model development first employs the Random Forest (RF) algorithm to identify high-impact variables, subsequently constructs a Multilayer Perceptron (MLP) to capture the intricate nonlinear relationships between input variables and FeO content, and finally integrates the Wild Horse Optimizer (WHO) algorithm to optimize MLP hyperparameters, thereby enhancing both model fitting accuracy and generalization capabilities. Validation using real-world production data demonstrates that the proposed model achieves an accuracy of 91.26% within ±0.5% error margin. This framework provides actionable insights for industrial applications, effectively mitigating FeO content fluctuations and advancing operational precision in sintering production, with significant implications for optimizing process stability and elevating metallurgical manufacturing standards.

To address the issue of low thickness control accuracy during non-steady-state rolling processes, such as specification switching and roll change in hot continuous rolling, this paper proposes a data-driven model based on Stacking ensemble learning for thickness prediction during specification switching. The optimal hyperparameter configurations of both the base learners and the meta-learner are determined using a Bayesian optimization algorithm. The performance of the Stacking model is systematically compared with that of other mainstream models. Additionally, the SHAP (SHapley Additive exPlanations) method is introduced to interpret the Stacking model and evaluate the importance of input features. The results demonstrate that the proposed Stacking model effectively integrates the predictions of base learners, significantly enhancing the accuracy of thickness prediction. The performance evaluation metrics of the developed Stacking model are as follows: root mean squared error (RMSE) of 0.019 9, mean absolute error (MAE) of 0.014 9, and coefficient of determination (R2) of 0.999 9. The probability that the error between the predicted and actual thickness is within ±35 μm reaches 94.8%, while the probability within ±50 μm is 97.8%, achieving high-precision thickness control during the specification switching process in hot continuous rolling.

XWith the continued development of intelligent steel manufacturing, green and low-carbon transition, and industrial big data platforms, steel production is shifting from experience-driven operation and isolated automation toward intelligence driven by data perception, knowledge constraints, and closed-loop optimization. However, steel processes are characterized by long process routes, numerous variables, strong coupling, evident time delays, and delayed quality feedback, making multi-source heterogeneous data difficult to use directly for reliable modeling and on-site decision-making. From a knowledge-data dual-wheel perspective, this paper reviews recent progress in big data-based intelligent modeling, deployment, and applications in the steel industry. It analyzes the governance and fusion of time-series, image, text, material genealogy, and multi-source heterogeneous data; summarizes the role of industrial knowledge graphs in semantic alignment, knowledge-based reasoning, diagnostic interpretation, and quality traceability; and discusses the applicability and limitations of key techniques, including time-series prediction, visual inspection, industrial natural language processing, multimodal fusion, federated learning, continual learning, and cloud-edge collaboration. On this basis, typical applications such as real-time detection and diagnosis, key quality indicator prediction, process scheduling optimization, and full-process quality traceability are reviewed. Future directions are further discussed, including knowledge-enhanced modeling, trustworthy deployment, industrial foundation models, digital-twin closed loops, and green low-carbon decision-making. This review provides a reference for the evolution of steel industry intelligence from local model applications toward full-process trustworthy closed-loop systems.

The control of the final cooling temperature after medium plate rolling is a core technology for improving product quality and production efficiency. The accurate prediction and regulation of the temperature hit rate have attracted much attention from both the academic and industrial communities. Early research was dominated by mathematical analytical models, which constructed modified Newton′s cooling law models based on heat transfer theory. Although these models had high computational efficiency, their adaptability to complex working conditions was limited. With the development of computer technology, the finite difference method (FDM) and the finite element method (FEM) have been widely applied in temperature field simulation, enhancing prediction accuracy through discretization. However, they rely on a large amount of experimental data for parameter calibration and have high computational costs. In recent years, machine learning models, with their strong nonlinear mapping capabilities, have become a research hotspot. Algorithms such as BP neural networks and XGBoost models have shown significant advantages in predicting the final cooling temperature hit rate. For model optimization, scholars have proposed innovative methods such as adaptive tuning of hyperparameters. By integrating mechanism-based and data-driven strategies, these methods effectively address the bottlenecks of industrial data noise sensitivity and insufficient generalization ability.Although machine learning technologies have demonstrated significant application value in industrial production, the steel industry—as a quintessential traditional complex industrial sector—features production processes characterized by multi-stage coupling and high dynamism.This results in multiple adaptation challenges when introducing and implementing machine learning technologies, further constraining their large-scale application. Future research should focus on multi-modal coupling modeling, energy-saving cooling process optimization under low-carbon targets, and the application of explainable artificial intelligence in industrial decision-making, thereby promoting the leapfrog development of hot rolling cooling control towards intelligence and greenness.

The slag grinding process is essential for manufacturing slag powder, which is a green raw material. The quality of its indicators directly affects the economy and safety of the entire process. However, this process involves multiple controlled variables and features strong nonlinearity, large feedback delays, and frequent fluctuations in operating conditions, making it difficult to control the indicators of quality, throughput, and temperature. To address these issues, this paper designs an intelligent control system for slag grinding based on model predictive control. The system first performs intelligent optimization of the control targets through an existing platform, and then controls the quality, throughput, and temperature indicators using the model predictive control method. Practical operating results show that the application of this system reduces the unit gas consumption and unit power consumption from 35.7 m3/t and 40.4 kWh/t to 34.1 m3/t and 38.7 kWh/t, respectively. These results indicate that the proposed method can significantly reduce energy consumption while maintaining stable product quality, thus demonstrating its practicality and effectiveness.

Aming at the problems of rolling force calculation and process optimization in the hot continuous rolling process of high-strength non-oriented silicon steel, the influence of deformation temperature and deformation rate on the hot rolling deformation process of high-strength non-oriented silicon steel was studied through thermal simulation experiments and analysis of true stress-strain curves. A corresponding deformation resistance model was constructed, and the regression coefficient of the deformation resistance curve was solved. Based on the rolling force model, the influence of different rolling speeds and deformation temperatures on the difference in rolling force changes was analyzed. Furthermore, with the goal of balancing the equipment capacity of each stand rolling mill, improving product performance quality and production line efficiency, the objective functions of balancing the remaining proportion of rolling mill equipment capacity and controlling the temperature and speed of hot rolling process were established. The optimization technology of high-strength non-oriented silicon steel rolling process was developed, and the reasonable setting of inlet temperature and rolling speed was achieved, and significant application effects were achieved.

Abstract:A machine learning based method for predicting the amount of molten iron in blast furnaces was proposed to address the importance of predicting the amount of molten iron in ladle scheduling and improving production efficiency, as well as the problem of insufficient prediction accuracy of traditional mechanism models. Based on the blast furnace production data of a certain steel plant from January 2024 to March 2025, a high-quality dataset was constructed through missing value filling, outlier processing, and Z-Score standardization. Pearson correlation analysis and random forest feature importance evaluation were combined to screen 19 key parameters such as soft water pressure, gas utilization rate, and coal quantity. The performance of AdaBoost, random forest, support vector machine, neural network, and linear regression models were compared. The results showed that the AdaBoost model performed the best in predicting the amount of molten iron, with a fitting goodness of R2 of 0.78 and a Mean Square Error (MSE) of 13.59. The prediction accuracy reached 87.2% within the range of ±10 tons and 94.1% within the range of ±15 tons. The model using the Stacking integrated framework achieved a prediction accuracy of 100% within the range of ±15 tons, which can effectively support actual production scheduling needs. This method provides a feasible data-driven solution for accurate prediction of the amount of molten iron in blast furnaces.

In hot strip rolling mills, loopers are installed between finishing stands to ensure the balance of metal mass flow and the stability of strip tension during the rolling process. Therefore, the control accuracy and dynamic response characteristics of the looper system directly determine the quality of the final strip product. To address the control challenges caused by the multivariable strong coupling, nonlinear and time-varying characteristics of the hydraulic looper system, this paper designs a composite intelligent control strategy that combines an Improved Particle Swarm Optimization (IPSO) algorithm with Back Propagation Neural Networks (BPNN) Proportional-Integral-Derivative (PID). A feedforward compensation decoupling control loop is constructed to weaken the coupling effects in the system. An improved PSO algorithm with dynamic adjustment mechanisms for inertia weights and learning factors is developed to solve problems in traditional BP neural network PID, such as strong randomness in weight initialization, tendency to fall into local optima, and slow convergence speed. Simulation experiments based on the MATLAB/Simulink platform demonstrate that the proposed IPSO-BP-PID algorithm achieves better control performance compared to both BP neural network PID and traditional PID under unit step input signals, verifying the effectiveness of the IPSO-BP-PID algorithm.

Abstract:As a critical material in industry, hot-rolled strip steel has a direct impact on downstream manufacturing quality due to the accuracy of its mechanical property prediction. However, in practical applications, the predictive performance of models is difficult to improve due to insufficient data sample sizes. When considering collaborative modeling across multiple production lines or factories, dual challenges of data heterogeneity and privacy protection arise. To address these issues, this paper proposes a method for predicting the mechanical properties of hot-rolled strip based on federated learning. The method first achieves feature dimension alignment and enables multi-party collaborative prediction of tensile strength, yield strength, and elongation of hot-rolled strip under the premise of ensuring data privacy and security. Its results are compared with those of models trained on single production line data. Experimental results demonstrate that this method exhibits good performance in all mechanical property prediction tasks. Additionally, to further optimize the scheme, this paper introduces a federated dimension optimization method based on feature importance to collaboratively screen out key factors that significantly affect mechanical properties.

Longitudinal cracks are common and serious quality defects in slab production. Accurate prediction of longitudinal cracks is of great significance to improving slab quality and production efficiency. However, in actual production, the number of longitudinal fissure samples is far less than that of normal samples, resulting in extremely unbalanced data distribution, which affects the effect of model training and brings huge challenges to model construction. How to improve the generalization ability and prediction accuracy of the model has become a key problem. To this end, firstly based on a large amount of measured temperature data of thermocouples on the copper plate of the mold, the sliding window technology was applied to extract temperature samples of longitudinal cracks′ and other conditions. Then, a slab surface depression-type longitudinal crack prediction method based on grid search optimized Convolutional Neural Network (CNN) and Bidirectional long Short-Term Memory (BiLSTM) network was proposed. The samples were input into the CNN-BiLSTM network, CNN was used to obtain the local features of the time series, and BiLSTM was used to obtain the long-term dependency features. Finally, the slab longitudinal crack prediction output was performed through the fully connected layer. Experimental results show that the proposed grid search optimized CNN-BiLSTM model performs significantly better than other models on the test set, with a prediction hit rate of 99.29% for longitudinal crack temperature waveforms, a false alarm rate of 0.71%, and a Matthews Correlation Coefficient (MCC) as high as 0.96, and the training and prediction speeds of the model are relatively fast. The research results provide a reliable theoretical basis for the identification of longitudinal cracks on the slab surface.

The ironmaking front-end processes, encompassing the raw material yard, sintering, pelletizing, coking, and blast furnace ironmaking, constitute the most energy-intensive and carbon-emitting segment of steel manufacturing. Consequently, the level of intelligence achieved within these processes directly determines the overall competitiveness and sustainability of the entire production chain. Taking industrial big data empowerment as the central narrative, this paper systematically reviews the nearly fifty-year evolution of digital and intelligent technologies in ironmaking front-end processes. This evolution is delineated into four distinct stages: automation enlightenment, digital transformation, intelligent upgrading, and whole-process collaborative breakthrough. From this historical analysis, three overarching developmental trajectories are distilled: the hierarchical advancement of control architecture, the paradigm shift in modeling approaches, and the progressive evolution of decision-making modes. Building upon this foundation, the paper provides an in-depth examination of the current applications of industrial big data and artificial intelligence technologies across the five core unit operations, elucidating the fusion mechanisms between data-driven methods and mechanistic models under diverse operational scenarios. Through a systematic comparison of domestic and international technological pathways, the paper identifies existing limitations regarding the depth of data-mechanism-scenario integration and the breadth of cross-process coordination. Looking forward, future innovation directions are envisioned from four critical dimensions: industrial big data technologies, research on large foundation models, smart manufacturing technologies, and the synergistic integration of green and low-carbon strategies. This review aims to provide a systematic reference for both theoretical research and engineering practice in the intelligent manufacturing of ironmaking front-end processes, thereby supporting the high-end, intelligent, and green transformation of the steel industry.

:Iron molten steel transportation serves as a critical link connecting ironmaking and steelmaking in steel enterprises, with its efficiency constrained by factors such as solution quality and computation speed of multi-locomotive path planning algorithms at the iron-steel interface. To address existing bottlenecks including low computational efficiency and susceptibility to local optima, this paper proposes a Whale-optimized Dynamic Weight Time A* algorithm (WODWT-A*) for multi-locomotive path planning. Firstly, the algorithm introduces a dynamic weight mechanism that linearly combines cost ratios with difference functions to enhance the heuristic function. This innovation significantly improves temporal efficiency and pathfinding accuracy in dynamic environments, overcoming the efficiency limitations of traditional A* algorithms in complex scenarios. Furthermore, to enhance global optimization capability, the method incorporates the Whale Optimization Algorithm (WOA). By simulating humpback whales′ group predation strategies, it achieves dynamic optimization of initial solutions and adaptive parameter adjustment, thereby strengthening multidimensional exploration in complex solution spaces and effectively avoiding local optima traps. This synergistic mechanism between global search and local optimization enables WODWT-A* to significantly improve planning stability and solution quality. Practical case studies demonstrate that WODWT-A* exhibits high adaptability and reliability in multi-task concurrency, path congestion, and dynamic environments, providing an optimized solution for coordinated scheduling of multiple locomotives at the iron-steel interface.

To address the sub-optimal intelligence level in current defect prediction methods for continuous casting billets, this study proposes a gradient boosting decision tree (GBDT)-based model for predicting slag inclusion defects. The synthetic minority over-sampling technique (SMOTE) was employed to resolve data imbalance issues, while Bayesian optimization was applied to determine the model′s globally optimal hyper-parameters. Furthermore, the GBDT algorithm enabled the extraction of coupled process parameters governing slag inclusion, ranked by variable importance metrics. This research accomplishes slag inclusion prediction using continuous casting process parameters and quantifies the influence of individual parameters through Shapley additive explanations (SHAP). The results provide actionable insights for parameter adjustment in billet production. The proposed framework has been successfully implemented in steel plant, it not only enhances the scientific rigor of process control in continuous casting but also lays a foundation for subsequent process optimization and new technology development.

Linz-Donawitz Gas (LDG) is a vital secondary energy generated during the converter blowing process. In balancing its recovery into gas holders and distribution to end-users, traditional scheduling mainly relies on manual expertise, long plagued by bottlenecks such as uncoordinated operational interfaces and high full gas-holder venting rates. To enhance LDG utilization efficiency, this study proposes a methodology for constructing production-consumption prediction models and a coordinated optimization strategy targeting gas-holder level stabilization, through systematic analysis of LDG pipeline network dynamics and recovery-distribution processes. A full-process balancing and scheduling model is established, encompassing gas recovery prediction, multi-constraint adjustment for compressors, and optimized allocation for end-users. Furthermore, a closed-loop execution mechanism is designed to send model-generated commands to the industrial control system. Deployed under actual operating conditions at a steel plant, the developed intelligent regulation system automatically adjusts compressor delivery volumes and modulates gas consumption at adjustable users, significantly improving pipeline network stability and balancing capacity. Operational results demonstrate that post-implementation, the frequency of full gas-holder venting events decreased from an average of 4.39 times per day to 0.37 times under same operating conditions, with significant effects. This provides an implementable solution for intelligent dispatching of secondary energy in steel enterprises, actively responding to low-carbon and energy-efficient development requirements.

To address the issue of shape prediction in the roller quenching process, this paper proposes a novel method based on clustering analysis and an improved Bidirectional Gated Recurrent Unit (BiGRU) network. Key process parameters influencing plate shape during quenching are first analyzed, and K-means clustering is applied to categorize sample data, thereby identifying the distribution characteristics of different steel grades and specifications. For continuously produced steel plate batches, a BiGRU model is employed to extract high-dimensional temporal features, capturing the sequential dependencies between preceding and succeeding shape variations. By integrating Convolutional Neural Networks with bidirectional gated units, an enhanced BiGRU model is constructed for plate shape prediction. Experimental results demonstrate that the K-means-guided improved BiGRU model achieves accurate shape predictions within acceptable process error margins. This method provides a reliable foundation for plate shape control in steel quenching processes and contributes to intelligent control in advanced manufacturing.

The rolling mill is a key piece of equipment in steel production, and the health of its rotor directly affects equipment safety and production efficiency. However, in practice, fault data is scarce and class imbalance exists, limiting the performance of traditional diagnostic models. To address this, this paper proposes a data augmentation method based on SN-WGAN-MMD to improve the effectiveness of rotor fault diagnosis. The method combines Wasserstein loss with Maximum Mean Discrepancy (MMD) loss to generate realistic samples from both global distribution and high-dimensional feature perspectives, while spectral normalization (SN) is used to enhance training stability. The generated samples are added to the original dataset to alleviate class imbalance. Experiments conducted on two datasets show that the proposed method outperforms GAN, WGAN, and WGAN-GP in terms of sample similarity, class balance, and diagnostic accuracy. The highest accuracy reaches 97.3%, representing a 15% improvement over the original dataset, demonstrating its advantages in global alignment, local feature representation. This provides an efficient and novel data augmentation solution for rotor fault detection.

In the production process of hot-rolled wire rods, material (wire rod) position tracking is of great significance for accurately recording the production process parameters and achieving closed-loop control of wire rod production. Traditionally, dedicated hot metal detectors were installed to achieve material position tracking, which increases the cost and maintenance burden of production facilities. This paper uses the existing rolling mill current and radiation thermometer temperature measurements in production as signals for judging material arrival, achieving position tracking of the wire rods. It was found in the study that due to interference factors such as insulation cover obstruction and temperature measurement position deviation, using a fixed threshold algorithm will lead to misjudgment of position status when judging material arrival. To overcome the above problems, this paper designs an adaptive threshold algorithm based on signal jumps. In the initialization process, the initial position judgment threshold is obtained by normalizing the signal values; after median filtering of the original signal, the type of signal jump is determined. subsequently, the threshold is calculated based on Wien′s formula. Field test results demonstrate that the position tracking results obtained with this algorithm are accurate and reliable.

The AI-native steel plant architecture is designed to address prevalent industry pain points in the process of empowering intelligent manufacturing with large language models, such as fragmented architecture, bolted-on AI, and insufficient production safety constraints. The reference architecture comprises three layers from bottom to top: the Digital Asset Interface Layer (DAS), the Intelligent Core Layer, and the Application Agent Layer. Taking an industry-specific large model as the domain-wide intelligent engine, the architecture relies on the DLA digital twin ontology to accomplish structured modeling of production lines and activities, delineate the inference boundaries of the large model, verify generated outputs, and constrain model behavior, thereby achieving a dialectical unity between unleashing large model capabilities and enforcing industrial production safety constraints. This paper analyzes the positioning, core functions, and innovative value of each of the three layers, providing theoretical guidance and an implementation reference for the engineering deployment of AI-native steel plants.

Abstract:This study aims to address the difficulty of straightness control in the straightening of high-strength plates caused by insufficient prediction accuracy of traditional methods. To this end, a high-precision prediction model is developed, and a particle swarm optimization-BP neural network model (PSO-BP) is proposed based on the synergy between finite element simulation and intelligent algorithms. A 3D dynamic explicit finite element model of an 11-roll straightening machine is constructed using ABAQUS to simulate the stress-strain field evolution and flatness change during the entire straightening process of a high-strength plate under various process parameters. Subsequently, 2 000 sets of measured data are collected from the production line, combined with 500 supplementary finite element simulation samples, and a cross-source heterogeneous training dataset is established after standardization. The Particle Swarm Algorithm (PSO) is employed to optimize the initial weights and thresholds of the BP neural network, effectively overcoming the problems of slow convergence, gradient vanishing, and local minima associated with traditional BP neural networks. Experimental validation shows that the PSO-BP model has excellent predictive performance: the correlation coefficient R-value in the training set is 0.957 and in the test set it is 0.965; the Root Mean Square Error (RMSE) has been reduced to 0.027; and the key prediction error rate remains stable at between 3.7% and 8.57%. This is a significant improvement on the traditional Finite Element Method (FEM), which has an error rate ranging from 18.57% to 20%. This study combines PSO global optimization and BP local approximation to achieve a breakthrough in generalization performance under complex process conditions, with prediction results that are highly aligned with actual outcomes. In the future work, the data diversity needs to be expanded to enhance the model′s adaptability and migration capability.

Ultra-thin high-strength strip is known as the “crown jewel of the steel industry” due to its complex rolling process, poor applicability of traditional mechanism models, and strong coupling among multiple control parameters. To address these challenges, a high-precision model and an intelligent technology system integrating thickness, tension, and flatness control have been established, leading to the development of an independently controllable intelligent control system for 20-high precision rolling. This system meets the production demands for ultra-thin dimensions, higher accuracy, and greater stability. It overcomes the control difficulties in thickness, flatness, and high-speed stable rolling of ultra-thin high-strength strip, ensuring high-quality independent supply of key materials such as high-grade stainless steel, electrical steel, and precision alloys in China. The materials have been applied in national defense, aerospace, new energy vehicles, special transformer systems, and other key fields.

The high-temperature, high-pressure, and heavy-dust environment of metallurgical furnaces severely impacts the stability and imaging quality of imaging equipment. To address this, this paper designs a protection system and proposes corresponding structural parameter optimization methods for the core challenges of high-temperature and dust protection. First, a steady-state heat transfer model of the furnace wall is constructed to determine the external thermal load. Based on this, a cooling structure parameter adjustment strategy based on steady-state heat exchange was proposed, and a correlation model between cooling structure parameters and cooling medium properties was established to ensure a stable operating temperature for the imaging equipment. Second, to address the issue of dust accumulation and lens blockage, a conical dust-proof structure utilizing an air curtain for dust removal was designed. An optimization objective function incorporating hybrid constraint handling techniques and its constraints were defined, covering the dust-proof performance, cooling performance, field-of-view limitations, and safety penalties of the protection system. Finally, simulations and real-world production environment tests verify the effectiveness of the optimized protection system. The results show that the internal temperature of the protection system is stable and lower than that in the no-cooling state. The average temperature during the test period is 37.6 ℃. At the same time, it effectively avoids lens clogging. The difference between the image Laplace energy and autocorrelation during production and during blow-off period is less than 5%. The designed system has been stably operating on site for 7 months, effectively extending the service life of the instrument and ensuring high-quality imaging.

Traditional manual disassembly and assembly of continuous casting slide gate cylinders require direct exposure to harsh working environments such as high temperature, high noise, and diffuse dust, featuring high labor intensity and high safety risks. The developed automatic disassembly and assembly technology for continuous casting slide gate cylinders takes the upper computer as the control core, with PLC responsible for bottom-level data collection and logic control. High-performance industrial robots are selected as actuators, supplemented by a 3D vision inspection system that detects distance characteristic values to adjust the cylinder piston stroke to match installation requirements and provides real-time target poses to the robot, enabling fully automated operations throughout the entire process of slide gate cylinder positioning, disassembly, and installation. Verified by on-site tests, this technology has successfully enabled robots to completely replace manual disassembly and assembly operations by robots. All performance indicators meet or exceed the requirements of production processes, significantly improving production efficiency and safety, and providing strong support for the intelligent upgrading of continuous casting production.

To improve the product quality of Ti6411 heavy plates, a three-dimensional thermo-mechanical coupled model was established using ABAQUS software to simulate the multi-pass rolling process of Ti6411 heavy plates based on actual rolling production conditions. The simulated rolling force data were compared with field measurements, and error analysis was conducted to verify the model′s accuracy. Through finite element simulations, the plate profile of the titanium alloy under different rolling parameters was investigated. The results demonstrate that roll diameter, rolling speed, rolling temperature, slab thickness, slab width, pass reduction rate, and friction coefficient are the primary factors affecting the head-end abnormal zone length in Ti6411 heavy plates. The causes of variations in the head-end abnormal zone length were analyzed based on heavy plate rolling theory. Furthermore, principal component analysis (PCA) was employed to investigate the influence patterns and extent of various rolling parameters on the head-end abnormal zone length. These findings were compared with the influence coefficients obtained from FEM-based analysis of each parameter′s effect on the head-end abnormal zone length. The research outcomes provide a theoretical foundation for optimizing the profile control process of Ti6411 heavy plates.

As the core unit ensuring thickness precision in strip rolling, the hydraulic Automatic Gauge Control (AGC) system directly determines the quality of high-performance metal materials. This paper systematically reviews the technological evolution of hydraulic AGC, dividing it into three logical stages: servo execution based on analog regulation, calculation-based precision control grounded in theoretical models, and adaptive compensation leveraging intelligent algorithms. The analysis focuses on evolutionary characteristics in response speed, model accuracy, and nonlinear compensation at each stage. Addressing the current control bottleneck caused by strong coupling between hydraulic drive dynamics and mechanical systems, this paper proposes a novel predictive management and control architecture that deeply integrates physical mechanisms with real-time data. Four key technical pathways are elaborated: high-fidelity mechanistic modeling, variable-fidelity surrogate model reduction, cloud-edge-end collaborative regulation, and digital twin-enabled full lifecycle perception. This framework aims to enhance system robustness under complex operating conditions, providing theoretical support for the transformation of hydraulic AGC systems towards digital and intelligent and autonomous operation.