1. Background and Issues of the Research
The steel industry is a foundational sector supporting economic development in China and is an energy-intensive industry. The carbon emissions and energy consumption generated during the blast furnace ironmaking process account for over 80% and 70% of the entire steel production process, respectively. With the implementation and continuous advancement of China’s dual carbon policy, low-carbon and efficient smelting in blast furnaces has become a key process in promoting energy conservation and emission reduction in the steel industry. In the blast furnace ironmaking process, the preparation of iron ore raw materials and the smelting process are the main links in energy consumption and carbon emissions.
With the extensive consumption and depletion of high-quality iron ore resources worldwide, the quality of the mainstream raw materials used by global steel enterprises has shown a trend of complexity and diversification. Approximately 80% of the iron ore in China’s steel industry is imported each year, placing China at a disadvantage in political and economic positions during ore import trade negotiations, which has long constrained iron ore imports. Therefore, the efficient utilization of iron ore raw materials is urgent. The efficient utilization of iron ore raw materials in China mainly faces the following four difficult bottleneck issues: ① the contradiction between “low-cost blending” and “high-quality sintered ore”; ② the matching relationship between raw material usage schemes and slag properties; ③ the efficient sintering of special ore resources; ④ the efficient, low-consumption, and low-emission use of fuel. For a long time, these four issues have restricted the economic, efficient, and low-carbon development of iron ore raw materials in China’s ironmaking process.
Blast furnace smelting is a “black box” operation, with unpredictable processes that overly rely on production experience, which limits the efficient operation of the blast furnace ironmaking process. Since the 1950s, researchers have begun to study and develop various mathematical models and “expert systems” for simulating and modeling the conditions of blast furnaces. Representative examples include the model integration system developed for Japan’s Keihin No. 1 blast furnace and the “Blast Furnace Automatic Control Expert System” at the Rautaruukki Steel Plant in Finland. Since the 1990s, domestic advancements in basic automation of blast furnaces have been promoted, predicting and simulating blast furnace parameters to identify abnormal changes in operating conditions. However, these models have not achieved the desired results, leading to the following issues in practical applications: ① poor interpretability of models, making it difficult for blast furnace operators and domain experts to accept and trust model outputs, thereby limiting their acceptability in practical applications; ② model complexity, making design, debugging, and interpretation difficult; ③ strong data dependency of models. When the operating environment of the blast furnace changes or new events occur, the performance of machine learning models declines or fails to adapt to new situations. The emergence of these issues has posed challenges for the application of big data technology, and the blast furnace has still not escaped the characteristics of a “black box”.
2. Solutions and Technical Approaches
In January 2014, Hebei Iron and Steel Group collaborated with research institutions such as Beijing University of Science and Technology, Northeastern University, and North China University of Science and Technology, aiming to utilize the ironmaking equipment and technical resources owned by Hebei Iron and Steel Group, combined with the scientific research strength of various research institutions, to solve the efficient utilization of iron ore resources and achieve low-carbon, intelligent, and efficient development of blast furnace ironmaking using big data and the fundamental theory of steel metallurgy.
1. Overall Approach
Addressing the issues faced by traditional blast furnace ironmaking processes in raw material production, such as complex raw materials and high costs, and the excessive reliance on production experience and low levels of automation and intelligence in the smelting process, greatly limits the intelligent, efficient, and low-carbon development of blast furnace ironmaking processes. Therefore, this project is based on the fundamental theory of steel metallurgy and fully applies computer and big data technologies. On one hand, it focuses on improving the efficiency and quality of raw material production and reducing energy consumption and carbon emissions in raw material production from the raw material end of blast furnace ironmaking. On the other hand, it conducts research on blast furnace smelting, aiming to improve smelting efficiency, enhance refined and intelligent control of blast furnaces, and achieve low-carbon, intelligent, and efficient smelting in blast furnaces.
2. Basic Content
To achieve efficient low-carbon smelting in blast furnace ironmaking, based on the concepts of source control and process enhancement, the project fully utilizes the fundamental theories of the sintering process and blast furnace smelting, organically integrating big data and artificial intelligence with front-end computer technologies. It conducts a comprehensive, multi-faceted system analysis of the entire process from iron ore raw material preparation to molten iron smelting in the blast furnace, developing efficient low-carbon ironmaking technology based on big data intelligence, which mainly includes:
① Development of low-cost, high-quality production technology for ordinary sintered ore;
② Low-carbon and efficient production technology development for vanadium-titanium sintered ore;
③ Development of a blast furnace status prediction and monitoring system based on big data technology;
④ Development of an intelligent analysis and operational status evaluation system for blast furnace parameters, based on data mining and expert experience.
3. Major Innovative Achievements
1. Established a database of the sintering basic characteristics of over 220 types of ores, and based on this, developed a quality-cost coupling blending model and expert system; using the magnesium-aluminum ratio of blast furnace slag as a pointer, achieved efficient utilization of MgO additives under low silicon and low alkalinity conditions, forming a low-cost, high-quality sintered ore production technology system.
(1) Developed a multifunctional coupling blending model and expert system, solving the problems of low-cost blending and high-quality production in sintering, achieving the goal of rational utilization of iron ore resources and flexibility in responding to changes in the iron ore resource market, thereby reducing ironmaking costs.
This project has established a database of sintering performance for over 220 types of iron ores, using the basic principles of linear programming to construct a blending model for sintering. Starting from the normal temperature sintering characteristics and high-temperature sintering characteristics of a single ore, and based on the sintering basic characteristics of the mixed ore during the baseline period on the production site, it derives the optimal blending scheme that meets the quality requirements for sintered ore. Combined with the expert system for sintering blending, a multifunctional raw material usage model that integrates the blending model and expert system has been developed. Under the premise of ensuring the quality of sintered ore, it breaks through the technical bottleneck of producing high-quality sintered ore from low-cost raw materials according to changes in iron ore market prices, as shown in the design and application effects of the sintering blending expert system in Figure 2.
(2) Established a raw material usage scheme based on the magnesium-aluminum ratio of blast furnace slag, breaking through the technical bottleneck of matching raw material usage schemes with slag properties, achieving optimization of iron ore raw materials and metallurgical performance of blast furnace slag, while reducing fuel consumption and smelting costs.
Using the optimal magnesium-aluminum ratio of blast furnace slag as a pointer, a systematic study was conducted on the influence patterns and mechanisms of blast furnace slag composition on the entire process of sintering-blast furnace smelting. Guided by the blending model and expert system, a core theory for the efficient utilization of MgO in the iron ore agglomeration process was established, and a series of modified MgO additive technologies for sintering were developed, achieving the production of high-quality sintered ore. First, based on thermodynamic analysis and kinetic calculations of interface diffusion, a theory of material migration and diffusion of MgO in the iron ore sintering process was established. Guided by this theory, efficient modified MgO additive technologies for sintering were developed, optimizing sintering process parameters to achieve the goal of producing high-quality sintered ore with low MgO, good RDI, and high metallurgical performance, as shown in Figure 3. In practical applications, the consumption of MgO flux in the sintering process was reduced from 1.85% to 1.75%, with solid fuel consumption reduced by 2-3 kg/t, and the cost per ton of ore reduced by 2.25 yuan.
At Tangshan Steel, through optimizing blending and resource reuse in the sintering process, the cost per ton of iron was reduced by 17.56 yuan. At Handan Steel, the MgO content of sintered ore was reduced from 1.85% to 1.75%, the cost per ton of ore was reduced by 2.25 yuan; the grade of materials entering the furnace increased from 57.96% in 2016 to over 58.3%, and the slag quantity decreased from 342 kg/tHM to below 330 kg/tHM. The Al2O3 content of the slag stabilized between 16.0%-16.5%, and the MgO/Al2O3 ratio was controlled at around 0.50.
2. Improved the sintering theory of vanadium-titanium magnetite and developed a real-time monitoring and forecasting system for the end point status of sintering, forming a complex ore low-carbon efficient agglomeration technology system.
(1) Explored the common issues and inherent characteristics of vanadium-titanium ore sintering through thermodynamic calculations and phase diagram analysis, improved the sintering theory of vanadium-titanium ores, and developed a full-process integrated technology to enhance the strength of vanadium-titanium sintered ore and reduce solid fuel consumption, effectively improving the indicators of vanadium-titanium ore sintering and ensuring the vanadium content in molten iron from the blast furnace and overall economic benefits.
Research was conducted on the existence forms of titanium-containing magnetite Fe2+(1+x)Fe3+(2-2x)TixO4 and titanium-containing hematite Fe2+xFe3+(2-2x)TixO3, analyzing the oxidation characteristics of vanadium-titanium ore sintering based on mineral structure morphology. Subsequently, through thermodynamic calculations and phase diagram analysis, the generation mechanisms of perovskite were studied from both solid-phase reactions and liquid-phase precipitations, examining the influence patterns and mechanisms of Ti on different minerals and liquid-phase properties (Figure 4). Based on this foundation, the sintering characteristics of vanadium-titanium ores were summarized, improving the fundamental theory of vanadium-titanium ore sintering, providing theoretical guidance for rationally formulating sintering process parameters; developing a full-process integrated technology to enhance the strength of vanadium-titanium sintered ore and reduce the return ore rate, with the strength of sintered ore increased by 1.8%, and the return ore rate reduced from 16.06% to 12.86%, a reduction of 3.2%.
(2) By integrating expert process experience and big data mining technology, established a real-time monitoring and forecasting model for the end point status of sintering, solving the issue of poor detection accuracy for the end point status of the sintering process.
Data from the entire sintering production process was collected, cleaned, and integrated to establish a data warehouse for the entire sintering process. By integrating expert process experience and big data mining technology, methods such as Pearson correlation coefficients, stability selection, and recursive feature elimination were used to select feature variables for the forecasting model. An ensemble learning algorithm was used to establish the end point status forecasting model, as shown in the technical roadmap of the model in Figure 5. The model output results were integrated with expert rules, using decision trees and optimization methods to achieve accurate forecasting of the end point status of the sintering model. The comparison of sintered ore quality before and after the model application over five months is shown in Table 1.
The stability rate of vanadium-titanium sintered ore TFe increased from 83.38% to 87.62%, the return ore rate decreased from 16.06% to 12.86%, a reduction of 3.2%, and the solid fuel consumption for sintering decreased from 50.55 kg/t to 47.95 kg/t, a reduction of 2.6 kg/t, with the utilization coefficient of the sintering machine increased to 1.3 t/m2·h.
3. Formed blast furnace status prediction and control technology based on big data analysis, achieving predictions and control of blast furnace pressure differential, material layer distribution, furnace throat erosion, vanadium content in molten iron, and temperature, realizing real-time predictions of blast furnace operating status.
(1) Based on blast furnace operating and monitoring data, using Keras algorithm, LSTM, and neural network technology, developed a prediction system for parameters related to molten iron quality, including molten iron temperature and vanadium content.
Traditional prediction models rely on a large amount of mechanistic reaction data and have poor prediction effects for molten iron temperature. To overcome this difficulty, a recurrent neural network (RNN) temperature prediction model based on the LSTM algorithm was established using a self-learning approach, with operational effects shown in Table 2. By collecting on-site process parameters and conducting big data analysis, a prediction and decision-making system for vanadium content in molten iron was established. After the system went online, operational effects are shown in Figure 6. The prediction error for vanadium content in molten iron was within ±0.025, achieving a rate of 95.01% for stable and rapid increases in vanadium content.
(2) Through deep exploration of multi-source heterogeneous data from raw materials, operations, and indicators, established a real-time monitoring system for the operating status of blast furnace pressure differential, material layer distribution, and furnace throat activity using data model fusion modeling.
Based on a deep neural network (DNN) prediction model, a prediction system for blast furnace pressure differential was established to overcome gradient disappearance and avoid local optimal traps, achieving a prediction accuracy of ±3kPa for 92.2%, realizing real-time monitoring and accurate prediction of blast furnace pressure differential. An intelligent monitoring system for the entire blast furnace material layer distribution was constructed, with several different charging regimes simulated as shown in Figure 7. By online collecting material parameters, tracking and simulating the charging process in the upper part of the blast furnace was achieved, forming a visual display. After each batch of material is charged, the site can view the distribution status and quantitative distribution data of the ore layer and coke layer in real-time, breaking open the black box of the blast furnace and achieving real-time tracking of the material status in the blast furnace.
By combining big data technology, blast furnace ironmaking theory, and expert experience, a working evaluation system for the blast furnace for smelting vanadium-titanium ore was established, comparing the predicted and actual erosion states of the furnace throat as shown in Figure 8. The monitoring effects for the furnace throat activity are shown in Table 3, achieving real-time monitoring and accurate prediction of different states and parameters in the furnace throat area.
After the system went online, the trend accuracy rate for molten iron temperature changes reached 95%; the prediction accuracy for molten iron temperature reached 91.2%-93.5%; the prediction accuracy for vanadium content in molten iron reached 95.01%; and the prediction accuracy for furnace throat activity exceeded 95%.
4. Established a blast furnace smelting knowledge base based on big data mining and expert experience, constructing a blast furnace operation analysis and optimization system and a comprehensive scoring evaluation system for blast furnace operating status, providing scientific evaluation and suggestions for blast furnace operations.
(1) Based on statistical correlation algorithms, determined the appropriate range for optimizing and adjusting key operational parameters of the blast furnace, combining actual production conditions to establish a model for optimizing and adjusting key operational parameters of the blast furnace, achieving the goal of quickly adjusting the blast furnace to the optimal production state.
By comprehensively considering indicator selection, indicator range division, indicator weight assignment, score calculation and evaluation, and parameter discrimination, a comprehensive scoring evaluation system for blast furnace operating conditions was established using big data combined with blast furnace ironmaking theory and expert experience. This system accurately displays the operating state of the blast furnace, quickly judges the short-term operational state of the blast furnace; using a stepwise adjustment intelligent strategy, the blast furnace conditions are quickly adjusted to the optimal state, forming a new system for judging and adjusting the operating state of the blast furnace, providing technical support for intelligent smelting in the blast furnace, with application cases shown in Figure 9.
2. Combined data mining and expert experience to establish a blast furnace knowledge base, developing an intelligent analysis and optimization system for blast furnace operational parameters, which can provide reasonable adjustment suggestions for parameters such as coal injection, wind pressure, and coke ratio based on the operating status of the blast furnace.
By combining the dynamic production process of the blast furnace, a real-time calculation model was established, extracting characteristic data of the furnace conditions, integrating expert knowledge base, fuzzy logic, and temporal reasoning, and developing an intelligent analysis and optimization system for blast furnace operational parameters. Based on inference results, suggestions for adjusting parameters such as blast volume, coal volume, wind temperature, and oxygen enrichment were provided. The application effects are shown in Figure 10, with the direction of operational parameter adjustments matching the predictions of experienced foremen or experts exceeding 90%.
After this innovation was applied at Chenggang, the system provided adjustment direction suggestions for operational parameters that matched expert predictions over 90%. With the support of the blast furnace operation analysis and optimization system and the comprehensive scoring evaluation system for blast furnace operating status, the utilization coefficient of the 2500m3 blast furnace increased from 2.5 t/(m3·d) to 3.12 t/(m3·d), with daily pig iron production reaching 7800 t.
5. Application Status and Effects
This project has been under research since 2014 and began application in January 2020 across various subsidiaries of Hebei Iron and Steel Group, involving blast furnaces ranging from 1000 m3 to 3000 m3, with raw material conditions including local iron powder, conventional imported iron powder, Australian high-alumina iron powder, alkali metal-containing iron powder, and vanadium-titanium magnetite concentrate. The results have been scaled up for application across various levels of blast furnaces and various raw material conditions within Hebei Iron and Steel Group. In the past three years, through guiding ironmaking production, costs have been reduced by 750 million yuan, with overall economic benefits exceeding 1.4 billion yuan.
In 2020, the endpoint prediction for sintering and vanadium-titanium ore sintering theory was used to guide the production of vanadium-titanium sintered ore at Chenggang. The blast furnace status prediction and intelligent analysis and evaluation system were used to guide the smelting of vanadium-titanium magnetite at Chenggang. Through continuous improvement and upgrades, the utilization coefficient of the vanadium-titanium ore blast furnace at Chenggang can reach up to 3.12 t/(m3·d), leading the international standards.
Hebei Iron and Steel Group Co., Ltd. Handan Branch adopted the key technology of the “Efficient Low-Carbon Ironmaking Technology Research and Application Based on Big Data Intelligence” project, specifically the “raw material usage scheme based on the magnesium-aluminum ratio of blast furnace slag” to guide sintered ore production, resulting in an average reduction of MgO content in sintered ore and slag to 1.58% and 7.32%, respectively, entering the optimal magnesium-aluminum ratio three-section control theoretical range (when the slag Al2O3=15%-17%, MgO/Al2O3=0.45-0.50). The MgO/Al2O3 ratio of the slag was reduced from 0.63 to the range of 0.45-0.50, while optimizing the raw material structure, resulting in a reduction of the blast furnace slag ratio from 342 kg/tHM to 313.40 kg/tHM, achieving good technical results.
This project has authorized 9 invention patents, 6 utility model patents, accepted 2 invention patents, and published 16 papers, including 7 included in SCI, with over 100 citations.
Source of Information: Hebei Iron and Steel Materials Technology Research Institute
To view all articles on new technological advancements, please click “Read Original”.