"AI+ Steel" Driving Steel Industry Modernization

Wang Guodong, Academician of Chinese Academy of Engineering, Professor at Northeastern University of China

In today's world, the wave of digitalization is sweeping across the globe, placing the steel industry at a pivotal juncture unseen in a century. The research paradigm in materials science has shifted from traditional methods driven by experience, theory, and computation to a modern approach driven by data and artificial intelligence (AI). As a key technology in the digital era, AI is vital to overcoming development bottlenecks and fostering high-quality growth in the steel sector. Launching the "AI + Steel" initiative is now more crucial than ever.

A look back at the development of the "AI+ Steel" technology system

From a global perspective, some developed countries have made modest strides in the digitalization and intelligent transformation of steel production. While there have been reports of attempts by POSCO (South Korea) and JFE Steel (Japan) to use deep learning for intelligent forecasting and operation control in blast furnaces, further advancements remain elusive, highlighting the substantial challenges in achieving end-to-end intelligence.

In contrast, China has made significant progress in advancing "AI+ Steel" through the collaboration among industry, academia, and research. Relevant explorations can be traced back to the 1970s when Professor C.M. Sellars from the University of Sheffield in the UK pioneered studies on predicting microstructure evolution and performance during hot rolling of steel. Building on this foundation, Northeastern University of China, collaborated with Shanghai Meishan Iron & Steel Co., Ltd. (Meisteel for short) to utilize Bayesian neural networks for predicting the microstructure evolution and performance. This significantly advanced large-scale customized production of hot-rolled steel.

In August 2014, the State Key Laboratory of Rolling and Automation at Northeastern University held a seminar on leveraging big data for integrated microstructure and performance control in steel making, continuous casting, rolling, and heat treatment. This event accelerated the adoption of big data in the steel industry. In 2017, as part of the national intelligent manufacturing initiative, a proposal was put forward to establish a cyber-physical system for the steel industry.

Since 2019, the Collaborative Innovation Center of Steel Technology at Northeastern University has spearheaded research on AI integration across various stages of the steel manufacturing process. Leveraging extensive real-world data from each stage, the center drives the seamless fusion of human expertise with AI, incorporating human insights to develop a hot rolling microstructure and performance prediction model. This model operates online continuously in real-time, utilizing vast datasets of historical production operations to intelligently predict material microstructure evolution with high fidelity. The automation team at the center integrates AI technology with their deep-rooted control theories and experience to tackle critical control challenges. Among these are challenges involving hot rolling force parameters and shape dimensions, earning the team numerous provincial and ministerial-level science and technology awards in recent years.

Professor Chu Mansheng's team from Northeastern University partnered with Meisteel and other companies to develop an intelligent blast furnace system for Meisteel. By combining big data, AI, and metallurgical principles with experiential knowledge, they created a highly efficient, cost-effective, and high-fidelity intelligent blast furnace model. This model addresses key technical challenges like data governance, condition monitoring, trend forecasting, and optimization decision-making in blast furnace ironmaking, creating a unique and economical intelligent ironmaking process for Meigang.

Through the concerted efforts of various specialized teams at the Northeastern University's Collaborative Innovation Center of Steel Technology, humanoid intelligence (HI) systems have been successfully implemented in all major units of the entire steel production process.

Standardized technical system for the "AI+ Steel" industry

Through our practice and exploration in the steel industry, we have gradually developed a standardized technical system for "AI+ Steel". This system consists of six interdependent key technological components:

(1) Data collection and governance are the foundation of the entire system. The core value of data lies in discovering hidden patterns, trends, and correlations in data to support decision-making, prediction, and innovation. The accuracy, integrity, and fidelity of data collection, as well as data extraction, transformation, and load (ETL) governance, are the most basic requirements for scientifically utilizing data.

(2) In terms of model system construction, we combine artificial intelligence generated content (AIGC) with HI to develop a data-driven industrial control model system. This system consists of three layers: a digital twin model for edge black-box systems, an unsupervised machine learning model deployed in the cloud, and a large language interpretation model.

(3) When configuring computational resources, special attention must be paid to the real-time demands of steel industry control. Despite the use of personalized data from a company's data pool and relatively low processing volumes, the industrial control model system requires low latency and high fidelity. Experience has shown that standard computational resource setups generally suffice to meet these stringent requirements.

In selecting algorithms, an end-to-end approach is recommended. IT technology can be used to establish a direct prediction model between input variables and output variables, eliminating the need to simulate complex material changes. This results in a system that is simpler, more efficient, and easier to maintain.

(4) For system architecture design, we suggest adopting a "one network, three platforms, and four functions" (1-3-4) digital foundation solution. Utilizing 5G Industrial Internet as the network backbone, the solution features a three-tier architecture: an underlying physical device platform (device), a digital twin core platform (edge), and a resource allocation and management cloud platform (cloud). The entire system employs a streamlined dual-layer IT architecture to facilitate machine learning and the development of cyber-physical systems.

(5) Industrial software development places particular importance on the modernization and enhancement of existing systems. To achieve this, specialized end-to-end process control and data governance software must be developed to meet the unique intelligent modernization needs of the steel industry.

Building "AI+ Steel" with Chinese characteristics

In advancing "AI+ Steel," it is essential to tailor our efforts to China's specific conditions and forge a path with distinct Chinese features. This involves focusing on the steel production line as the central axis and addressing the complexities throughout the entire process. By leveraging "AI+ Steel" technology, we aim to create Robot Steel (RS) for the industry. Our approach employs the "1-3-4" digital foundation with a two-tier flattened framework and integrates HI with AIGC to form hybrid human-machine intelligence, advancing the intelligent transformation of the steel industry.

Simultaneously, we should fully leverage the most advanced computer hardware systems and operating environments from the steel sector to achieve data-driven and software-defined IT transformations. This will pave a low-cost, efficient, scalable, and risk-free modernization pathway for the "AI+ Steel" initiative in China.

The "AI+ Steel" standardized technical system, while primarily designed for the steel industry, can also be extended to process control in other material sectors such as non-ferrous metals, chemicals, and building materials.

Suggestions for the "AI+ Steel" initiative

To further advance the "AI+ Steel" initiative, we should balance immediate actions with long-term vision, systematically planning our development path. First, we should establish clear development goals. These should focus on improving product quality and streamlining production processes through technological innovation, building a robust ecosystem to drive innovation, and enhancing the overall core competitiveness of the steel industry. In terms of R&D, I recommend prioritizing three key areas: deepening the integration of big data and machine learning, advancing the development of human-machine hybrid intelligent systems, and exploring multi-agent collaborative optimization technologies.

The initiative will unfold in three distinct phases:

Phase I (2025–2026)

Focus on setting an example by building 10 or more integrated "AI+ Steel" production lines covering all ironmaking, steel production, casting, and rolling processes. It will also involve establishing and refining relevant standards and evaluation criteria.

Phase II (2027–2030)

It will involve scaling up efforts to complete over 30 end-to-end intelligent production lines.

Phase III (2031–2035)

Aims to achieve widespread adoption across the industry, effectively transforming the steel sector through the "AI+ Steel" initiative.

By strategically implementing the "AI + Steel" strategy and harnessing China's industrial practices and intelligent technology advantages, we will lead the way in the modernization of China's steel industry with unique Chinese characteristics. This will pave the way for high-quality development in the sector and establish a benchmark amid the global trend of intelligence in the steel industry.