Research on the Application of Artificial Intelligence Technology in the Steel Industry

Abstract: With the development of information technology, the maturity of big data, the Internet of Things, and the improvement of computing power, artificial intelligence has begun to experience a third wave of academic research and industrial application. The steel industry is a complex process industry with intricate internal production processes and numerous influencing factors. Artificial intelligence has high application potential and value in the steel industry. With the advancement and implementation of the national integration policy, the level of informationization in China’s steel industry has gradually improved, laying a solid foundation for the application of artificial intelligence technology in steel enterprises. This article first explores the research areas of artificial intelligence technology, mainly including expert systems, neural networks, intelligent robots, machine learning, and intelligent optimization. It then studies the main application scenarios and results of these technologies in the steel field, and finally proposes how artificial intelligence technology can assist the high-quality development of the steel industry from production optimization to strategic management.

Keywords: Steel Industry; Artificial Intelligence; Production Optimization; Intelligent Manufacturing

The concept of artificial intelligence was born in the mid-20th century and has experienced two peaks during its development, both of which fell into valleys due to technological bottlenecks and application costs. Currently, with the development of a new generation of information technology, there has been a significant leap in data storage and computing capabilities, leading to a dramatic change in the environment for artificial intelligence. The research and industrial application of artificial intelligence technology have entered a new stage of development.

The Chinese steel industry has undergone years of rapid development and is currently transitioning from extensive development characterized by “high output, high costs, low prices, and low efficiency” to high-quality development. In terms of market demand, there is a trend towards small batches, diverse varieties, and customization. Chinese steel enterprises still face severe challenges, influenced by both upstream raw material industries and downstream steel product processing industries. Additionally, issues such as unreasonable product structure, low production management levels, high energy consumption, and poor product quality stability are significant internal factors affecting the competitiveness and sustainable development of enterprises. To promote the sustainable development of the steel industry, “Made in China 2025” proposes that the steel industry implement intelligent control and optimization of production and logistics, focusing on developing new information and intelligent technologies based on big data and cloud computing, achieving deep perception of enterprise information, intelligent optimization decision-making, and precise coordinated control.

Most processes in steel production have characteristics of multi-scale, multi-variable, non-linearity, and uncertainty, making it difficult to apply deterministic mathematical models to solve these problems. With the continuous development of the Internet and information technology, big data, cloud computing, and intelligent technology have emerged, undoubtedly providing a new way to solve these challenges. Furthermore, throughout the steel production process, enterprises have accumulated vast amounts of various data information; however, much of this data is currently only collected and stored, with its deeper value not fully explored. Based on the characteristics of the steel production process, utilizing the long-term accumulation and continuous collection and monitoring of massive heterogeneous data lays a solid foundation for the application of artificial intelligence technology and has become a hot topic of concern for current steel enterprises, significantly contributing to the efficient, collaborative, and intelligent operation of steel production, enhancing the level of informationization and intelligent manufacturing in the steel industry.

Artificial intelligence is a branch of computer science and, as a new technological science, it uses information technology to simulate and expand human intelligence. Its research areas mainly include speech recognition, natural language processing, expert systems, image recognition, neural networks, etc. Artificial intelligence technology involves multiple disciplines and has many categories, with the main applications in steel production including expert systems, neural networks, intelligent robots, machine learning, and intelligent optimization.

1.1 Expert Systems

In the process of artificial intelligence technology moving from theory to practical industrial application, expert systems have played a crucial role as an important research direction. By leveraging the expertise of professionals in a specific field, general knowledge reasoning strategies (if-else) can be achieved, which further elevates to specialized knowledge in a specific field to guide production processes. In industrial applications, expert systems typically simulate complex problems that can only be solved by domain experts through knowledge representation and reasoning.

Commonly used expert systems in industry can be divided into five categories: (1) Rule-based expert systems: using a series of rules to represent expert knowledge; (2) Frame-based expert systems: a natural extension of rule-based expert systems, utilizing object-oriented programming concepts to describe data structures; (3) Case-based expert systems: using previous cases to solve current problems; (4) Model-based expert systems: clearly defining concepts and standardizing knowledge bases through models; (5) Network-based expert systems: positioning human-machine interaction at the network level.

1.2 Neural Networks

Neural network systems emerged in the mid-20th century and are algorithmic mathematical models that mimic the characteristics of biological neural networks, using neurons and adjustable weights between connected neurons to perform large-scale distributed parallel processing. They can conduct parallel computation, distributed storage, and processing of large-scale data while possessing self-organization, adaptability, and self-learning capabilities. Neural networks can effectively address the many uncertainties and fuzzy information processing challenges encountered in industrial production, which are difficult to model with mathematical programming models. The classic structure of neural networks typically consists of three layers: input layer, hidden layer, and output layer.

1.3 Machine Learning

The core of artificial intelligence is to enable computers and other devices to possess intelligence, and machine learning is the fundamental approach to achieve this goal. Machine learning involves methods from probability theory and statistics, allowing computers to analyze and learn from human behavior to acquire new knowledge or skills, continuously improving their performance. With the accumulation of industry data, methods based on machine learning for in-depth analysis of massive data and more efficient utilization of hidden information in big data have become the main direction of current machine learning.

1.4 Intelligent Robots

Intelligent robots are machines that possess varying degrees of human-like intelligence, capable of executing a closed-loop workflow of “perception-decision-action-feedback,” assisting humans in production and service, and automatically performing various tasks. The automated tasks performed by robots can replace humans in completing heavy and dangerous work, improving work efficiency, enhancing work environments, and reducing errors caused by human subjectivity.

1.5 Intelligent Optimization

Intelligent optimization algorithms are a collective term for a class of evolutionary algorithms, primarily derived from simulating the biological evolution process in nature. They use different expressions to represent genetic traits and iterate through evolutionary operators (such as crossover and mutation operators) to obtain optimal or satisfactory solutions to problems. Compared to traditional mathematical programming algorithms and exhaustive algorithms, intelligent optimization algorithms are a globally adaptive evolutionary algorithm that effectively solves problems that traditional optimization algorithms struggle with through modeling and encoding.

For large steel enterprises, the long process steel production process primarily includes processes such as coking, sintering, pelletizing, and blast furnace ironmaking before iron production, and in the steelmaking stage, processes such as converter, refining furnace, continuous casting, and rolling, as well as subsequent processing stages of rolled products.

The steel production process involves a combination of continuous, discrete, and semi-continuous situations, making it complex. Additionally, production processes often involve semi-structured and unstructured problems, making optimization challenging. The combination of artificial intelligence technology with traditional methods provides a pathway to address these issues. Artificial intelligence technology has penetrated various aspects of the steel production process, including product design, molten iron quality prediction, process control, product quality assessment, equipment failure diagnosis, raw material procurement and batching optimization, production planning, and scheduling, as shown in Figure 1.

Figure 1 Application of Artificial Intelligence Technology in Various Aspects of Steel Production

2.1 Coking Coal Blending

Blending coal is a critical step in coking production, referring to the determination of the proportion of individual coals in the blended coal through intelligent optimization and neural network models to achieve high-quality coke production at a lower cost. A reasonable blending ratio can not only reduce production costs for coking enterprises but also conserve resources and reduce environmental pollution. The coking coal blending process is illustrated in Figure 2.

Expert systems, neural networks, and intelligent optimization have extensive applications in the optimization of coking coal blending (the dark sections in Figure 2). In the coking production process, the various coal components in the raw materials differ, and their participation in production affects the quality of the final coke. Obtaining blending ratios through small coke oven experiments does not meet the real-time requirements of blending. Therefore, based on the actual coking coal blending production, Guo Yinan and others adopted an intelligent optimization algorithm in two steps to calculate the coal ratio, first establishing a neural network model based on historical data to predict coke quality, introducing a genetic algorithm during the training process to optimize the weights, enhancing model adaptability, and overcoming the reliance on qualitative analysis methods based on knowledge. Tan Shaodong and others developed a blending optimization expert system based on the management of the coking coal blending plant, achieving significant results in production management, quality management, and coke quality prediction. In recent years, with the maturation of neural network computing, which requires lower adaptability, research on neural networks in coking coal blending has gradually progressed, with Huang Yonghui, Tian Yingqi, and Tao Wenhua successfully using improved artificial neural network models to predict coke quality.

Figure 2 Coking Coal Blending Process

2.2 Iron Procurement and Batching Optimization

In the steel production process, the quality of sintering batching and blast furnace batching schemes largely determines the cost of molten iron. A batching scheme that meets smelting performance requirements and has a high cost-performance ratio can not only guide procurement but also guide production, bringing significant cost reduction benefits to steel enterprises. Shi Cantao and Luan Zhiwei, addressing the procurement of iron raw materials and the batching issues in steel production, comprehensively considered the process requirements of sintering and blast furnace production, establishing a two-stage mathematical programming model with production cost and output iron quality as objectives, using intelligent optimization algorithms and operations research algorithms for solutions, and applying research results to Tianjin Rongcheng United Steel Group Co., Ltd. and Xingtai Delong Steel Co., Ltd., reducing raw material costs in the iron area.

2.3 Blast Furnace Ironmaking

The information systems involved in the blast furnace ironmaking process focus on basic automation and production planning management, and are largely disconnected from the intelligent control of the smelting process due to the characteristics of the blast furnace core unit, which operates at high temperatures, high pressures, and is a closed, continuous large black box, making internal diagnosis of working conditions difficult, with operations still primarily based on human experience and subjective judgment. The emergence of artificial intelligence provides a pathway to explore the internal rules and characteristics of blast furnace ironmaking, establishing models. The stability of the blast furnace’s operation is fundamental to continuously producing high-quality molten iron, and the furnace temperature is a guarantee for stable operation, also serving as an important indicator for real-time assessment of furnace conditions. In blast furnace smelting, the temperature of the furnace is typically determined by the silicon, sulfur content in the molten iron, and the slag condition, which overly relies on human experience and lacks precision. In recent years, Cui Guimei and others have utilized the vast historical data accumulated by steel mills to construct a neural network model targeting the management of furnace temperature in relation to silicon quality fractions and molten iron temperature, allowing for comprehensive and accurate predictions of furnace temperature to assist operators in decision-making. Wang Wenhui and others, considering the differing statistical characteristics of blast furnace production data, established support vector machine (SVM) and random forest (RF) prediction models to further explore temperature control and prediction during the blast furnace smelting process. In the visualization of ironmaking, imaging technology plays an important role in monitoring the blast furnace top imaging, blast furnace tuyeres imaging, and blast furnace slag monitoring, with Li Xinyu and others conducting a comprehensive review of the application of imaging technology in the ironmaking field, analyzing the current application status of imaging technology in various ironmaking systems and discussing future application directions. In terms of blast furnace charging, Ma Futao and others combined blast furnace charging theory with artificial intelligence algorithms to address the difficulties and defects of traditional charging models, developing a numerical simulation model for the distribution of material layers during the charging process, which is highly adaptable and closer to actual conditions.

2.4 Steelmaking Continuous Casting Stage

The continuous casting stage of steelmaking is the most complex stage of the steel production process, where molten iron transported from the blast furnace through torpedo cars and other equipment is processed through converters, refining furnaces, and casting machines to produce steel billets or slabs that meet rolling requirements. The temperature of molten iron transported from the blast furnace to the steelmaking area affects the production operation in the steelmaking area; higher temperatures are beneficial for stabilizing operations and automatic control of the converter. Ren Yanjun and others analyzed the factors affecting the temperature during the transportation of molten iron at the blast furnace-converter interface, quantifying these factors and designing parameters that influence the temperature drop during this process, establishing a prediction model based on BP neural networks to forecast the temperature of molten iron and its drop process, meeting the actual requirements of field production. In the steelmaking stage, the quality issues of continuous casting slabs are a significant concern for steel enterprises. Online, real-time assessment and feedback of continuous casting slab quality can guide operators in reasonably controlling the steel composition. Guo Xianli and others established a neural network model aimed at accurately determining the quality of continuous casting slabs online and analyzing the causes of quality incidents, developing an online diagnostic system for continuous casting slab quality to guide production. Chang Yunhe and others, based on an analysis and summary of the types of quality defects in cast slabs and their main influencing causes, established prediction models based on BP neural networks for three typical quality issues, such as intermediate cracks, center cracks, and center segregation, and based on the trained neural network models, further developed an online forecasting system to achieve real-time forecasts of slab quality. Xue Meisheng and others, considering the significant lag and inertia often encountered in the production of regenerative step-type heating furnaces, established a heating furnace temperature prediction model using wavelet neural networks, successfully predicting future output values of furnace temperature and constructing a temperature optimization controller based on secondary performance indicators, achieving good tracking of temperature changes and shorter adjustment cycles.

In the logistics field of molten iron and steel during the steelmaking stage, Baidu collaborated with domestic steel enterprises to develop an intelligent ladle recognition system based on IoT technology. Through intelligent upgrades of traditional ladles, real-time data on ladle operational temperature and pressure can be collected, and through thermal imaging and computer vision technology, real-time information on ladle operational status can be visually displayed, achieving intelligent perception of ladle operational conditions, and based on real-time information, utilizing big data analysis and artificial intelligence technology to achieve refined management of ladles.

2.5 Rolling Stage

Artificial intelligence has extensive applications in the rolling field, primarily in three types of uses: (1) online control and adjustment of rolling processes; (2) optimization calculations of rolling process parameter information; (3) analysis of collected rolling data. Liu Xianghua and others conducted a comprehensive review of these three types of applications of artificial intelligence in the rolling field; He Anrui and others elaborated on practices and results achieved in precise control of rolling processes from five aspects: precise control of slab heating, precise control of slab shape during free rolling, online precise assessment and analysis of product quality, precise grinding and management of rollers, and the new generation of rolling mathematical models. In the optimization of cold rolling rolling force, Gao Lei and others improved the accuracy of rolling force calculations by uncovering the underlying patterns hidden in actual field data regarding deformation resistance and friction factors.

2.6 Gas Balance Scheduling

Steel enterprises consume substantial energy and resources, with energy resource costs accounting for a significant proportion of total production costs. In the context of high environmental costs and energy consumption, along with increasing environmental pressure, improving energy efficiency has become an intrinsic need and inevitable requirement for the steel industry in the new era. Gas is an important secondary energy source generated during the production of steel products, accounting for about 30% of total energy consumption in the enterprise.

In steel enterprises, certain production processes generate gas, such as blast furnaces and converters, while others consume gas, such as heating furnaces and boilers. Gas forecasting and optimization scheduling involve coordinating the supply of gas-producing equipment and the consumption of gas-consuming equipment through gas storage facilities and transmission equipment, regulating the use of externally purchased energy to effectively avoid gas shortages or surpluses, stabilizing gas pipeline pressure, and reducing gas dissipation and consumption, thus improving gas utilization efficiency. The surplus of gas and the demand for steam and electricity vary under different operating conditions in steel enterprises. To achieve maximum economic benefits, real-time optimization scheduling of the gas-steam-electricity system is required based on different operating conditions. He Dongfeng and others established a coupling optimization scheduling model for the gas-steam-electricity system of steel enterprises under different operating conditions, with the objective function of minimizing energy costs for system operation based on linear programming. Li Hongjuan and others addressed the frequent fluctuations and imbalances in the production and consumption of by-product gas systems in steel enterprises, where the balance between supply and demand significantly impacts production costs and energy consumption. They utilized improved neural network algorithms and models, optimizing scheduling for by-product gas based on the energy utilization characteristics of consuming equipment and forecasting results. The functional structure of the gas forecasting and optimization system in steel enterprises is illustrated in Figure 3.

Figure 3 Functional Structure of Gas Forecasting and Optimization System in Steel Enterprises

2.7 Equipment Failure Diagnosis

The equipment processes involved in steel production are complex, and the close interconnection between processes means that a failure in any one link can affect the entire production line. Preventive maintenance and failure diagnosis of equipment in steel enterprises, along with early warning of failures, become particularly important.

Equipment failure diagnosis is a prerequisite and an important component of achieving lifecycle management of equipment. Wang Yinghong and Hu Hao introduced the online diagnosis system for equipment failures constructed by Hebei Steel Tangshan. The system collects vibration data, electrical instrument data, control data, and other information about the equipment, analyzing this data to understand the operational status of the equipment and reasonably formulate maintenance plans, thus preventing potential issues. For blast furnace production, Ma Zhuwu and others developed an intelligent diagnosis and decision support system based on expert systems, achieving functions such as production management, furnace condition diagnosis, data analysis, and mathematical modeling.

2.8 Production Planning and Scheduling

Production planning and scheduling are key to achieving intelligent manufacturing and management in modern steel industries, having significant practical implications for efficient coordinated production and energy conservation in steel enterprises. Although production scheduling research has formed a systematic theory and methodology over many years, complex steel production processes that integrate multiple equipment, multiple processes, and various product structures exhibit characteristics such as numerous processes, complex equipment, and many uncertainties and disturbances in the production process. The combination of artificial intelligence technology with steel production process rules effectively mitigates the impact of uncertainties in steel production. Luan Shaojun and Wu Xiuting established a four-level planning system consisting of capacity planning, order planning, batch planning, and production scheduling for production planning and scheduling issues in steel enterprises, designing a prototype of an advanced planning and scheduling system based on intelligent optimization algorithms. Addressing the specific circumstances of Tianjin Rongcheng United Steel Group Co., Ltd., they implemented research results in order planning, integrating with ERP systems and e-commerce systems to achieve unified management of orders, reducing human intervention, and in batch planning, improving hot loading and hot delivery rates through integrated steel rolling planning.

With the increasing level of digitalization, networking, and intelligence in steel mills, as well as advancements in artificial intelligence technology, the application scenarios in steel production are gradually enriching, showing a trend of development from local to overall, and from production level to strategic management level.

(1) Artificial intelligence is data-driven, evolving from real-time data acquisition of individual processes to real-time data acquisition of the entire process, with the fusion and analysis of multi-source heterogeneous data within these processes being a direction for further research and application.

(2) Improving the online learning capability of models is essential; currently, they are mostly applied to historical data to train models, which are then used for actual predictions or inferences. As production continues, new situations arise, necessitating continuous iteration of models based on new situations to adapt to future conditions.

(3) Applying artificial intelligence methods to optimize the supply chain, sales forecasting, and guidance for steel enterprises, providing strategic-level decision-making guidance for steel enterprises.

Acknowledgments: The author thanks the National Natural Science Foundation and the Joint Fund Project (U1660109) for their support.

Contact: Wang You

Consultation Phone: 010-65256519

Official Website: China Metallurgical Planning Network Website: www.mpi1972.com Address: 12th and 15th floors, E building, Global Trade Center, No. 36, North Third Ring East Road, Dongcheng District, Beijing

Authoritative Qualifications: The Metallurgical Industry Planning and Research Institute was established in 1972 and is a national-level industry planning research and engineering consulting institution, one of the first batch of first-class engineering consulting units in the country. Outstanding Achievements: Completed over 6000 projects; served over 200 government departments, industry associations, more than 400 domestic metallurgical enterprises, and over 50 international enterprises. Strong Talent Pool: 90% of staff hold master’s or doctoral degrees, 70% possess senior technical titles, and 50% are experts enjoying government special allowances, registered consulting engineers, or senior engineers with professorial titles.

At the end of the Thirteenth Five-Year Plan, the Metallurgical Planning Institute assisted steel enterprises such as Shagang, Yizhong Group, and Rongcheng Group in preparing their strategic development plans for the 14th Five-Year Plan.

In recent years, as the steel industry enters a new normal of development, the Metallurgical Planning Institute has particularly carried out research in new business areas such as green development planning, enterprise environmental diagnosis, “Belt and Road” and steel enterprises’ “going out” planning, product deep processing planning, multi-industry development planning, enterprise cost reduction and efficiency enhancement planning, logistics optimization, management improvement, integration of two technologies, intelligent manufacturing, and e-commerce development research, and human resource optimization, assisting numerous steel enterprises in completing their transformation and upgrading to enhance overall competitiveness.

The Metallurgical Industry Planning and Research Institute is an important supporting agency in the field of standardization under the State Administration for Market Regulation and is the pilot unit for the cultivation of metallurgical standardization consulting service offices approved by the National Standardization Committee. The institute has a dedicated standardization research center that carries out standardization-related work. At the national standard level, it has undertaken the leadership and secretariat work of six standardization working groups, including TC20/Metallurgical Energy Fundamentals and Management, TC207/Metallurgical Environmental Management, TC275/Metallurgical Environmental Protection Industry, TC442/Steel Water Conservation, TC415/Comprehensive Utilization of Metallurgical Resources, and TC269/Steel Logistics; at the industry standard level, it has undertaken the secretariat work of three standardization working groups for the steel industry under the Ministry of Industry and Information Technology concerning water conservation, energy conservation, and comprehensive utilization of resources; and at the group standard level, it has undertaken the secretariat work of various group standardization working committees, such as the China Special Steel Enterprises Association, China Circular Economy Association, China Metallurgical Mining Enterprises Association, and China Ferroalloy Industry Association. The China Special Steel Enterprises Association is a national-level group standard pilot, with multiple group standards listed as application demonstration projects by the Ministry of Industry and Information Technology. The institute can provide comprehensive and systematic standardization consulting services.