Hunan iron & steel group makes the steelmaking process smarter and more efficient with the first steel industry model applied

On April 28 at the 2024 Hunan Advanced Manufacturing Application Scenario Conference in Changsha, Xiangtan Iron & Steel Group (Xiangtan Steel), a subsidiary of Hunan Iron & Steel Group Co., Ltd., stood alongside China Mobile Hunan and Huawei as they launched a Pangu Large Model application for the steel industry. Their collective goal with this launch was to promote the integration of traditional industries and AI, and to set a new standard for the digital transformation of the global steel industry.

Two days later, on May 30, Xiangtan Steel was awarded for their application of AI model in the steel industry at the AI for Good Summit in Switzerland. The application was named by the International Telecommunication Union (ITU) as one of the Innovate for Impact Use Cases of the year, standing out among over 219 submissions by companies from 38 countries.

From manual calculation to model computing

Xiangtan Steel 's main factory stands about 30 meters high and houses several large pieces of metallurgical equipment, including three converters, three refining stations, five ladle furnaces, and five continuous casters. The air surrounding the equipment ripples from the heat rolling off them. Five cranes slowly transport 150-ton steel ladles that hold molten metal clocking in at 1000 °C, spanning a distance of 500 meters.

Coordinating the cranes is a major challenge for the steel mill. According to Wei Shuilian, an information engineer at Xiangtan Steel, it's like completing a series of high-risk math tests, in a boiling hot classroom. Traditionally, cranes are manually controlled, and only experienced workers can operate the cranes. However, due to the complexity of the steelmaking process, relying solely on the experience of the crane operators can lead to process issues unsmooth, unaligned practices.

At Xiangtan Steel, the transition from manual crane operations to using large models for coordination has solved many of these issues. Their new intelligent crane scheduling system, equipped with steel models, integrates a vast amount of data, including steel production plans, crane maintenance information, the real-time location of the steel ladles, and various business rules. This system uses algorithms to generate intelligent crane scheduling plans. What's even smarter is that if there are any sudden changes to the production plan, the system only needs a minute to "think" and issue a new scheduling plan covering the next 30 minutes. Now, thanks to AI, even novice crane operators can become "experienced operators." The status and trajectory of each piece of equipment is displayed on the screen in front of the operator's seat, making it easy for them to follow the prompts and complete the operation. With the support of the large model, the turnover rate of steel ladles has greatly improved, and the waiting times and energy consumption per batch have decreased, leading to a cost savings of 1.2 CNY per ton of steel.

Large model’s "eagle eyes" supervise the "last-mile" of steel production

Conveyors play a critical role in steel mills by transporting material in bulk. They typically operate for 24 hours a day at a speed of two meters per second, and their status directly affects the continuity of steel production.

Conveyors usually require experienced workers to conduct regular inspections. They check for misalignments, bulging, foreign objects, tearing, and a host of other issues. If a belt breaks, the blast furnace is forced to shut down, which can result in a potential loss of one ton of molten iron for each minute that it’s down. An alarm is triggered when the conveyor belt deviates by only five centimeters, and a severe alarm is triggered when it deviates by a decimeter. The belt is 1.6 meters wide, with a buffer of two decimeters on either side. The conveyor belt is made of a mixture of rubber and steel wire, and bulging can occur over time due to friction.

Experts from Huawei's Oil, Gas & Mining Business Unit stationed at the mill conducted a detailed field survey of these issues and decided to employ an intelligent conveyor supervision solution using 5G and computer vision technology. Based on regular sampling, the system can intelligently supervise the conveyor belt's operations around the clock, predict trends that could result in bulging, and issue early warnings. The scenario design and alarm processing logic have since been continuously improved based on feedback from mill workers. After a month of large model training, the accuracy of the intelligent supervision has increased to 98%, effectively reducing the frequency and intensity of manual inspections.

Data-driven production transforms steel mills into a "central kitchen" for new applications

The potential of the Pangu Large Model is obvious to those pushing for intelligent transformation within the steel industry. The "0 to 1" breakthrough made by Xiangtan Steel with the large model was just the first step. They are now leveraging internal development to go from "1 to 10," applying the model to diverse scenarios. In Xiangtan Steel's high-speed wire rod mill, production lines can each turn out a single coil of wire in just one minute. In the past, workers had to conduct rough sampling for inspections by flashlight. The high temperature of the product reduced inspection efficiency and made inspection a relatively unsafe job.

Now, eight industrial cameras are controlled by a cloud platform to perform real-time 180-degree quality inspections on every coil produced. Once the platform detects defects, workers can remotely review them on their computers. Xiangtan Iron & Steel's own software team developed the wire coil quality inspection system with the guidance of Huawei's technical team in just 10 days, marking a significant improvement in development efficiency compared with their previous development process, which could take more than a month.

Thanks to low-code and no-code development tools, technicians can now efficiently convert business knowledge into technical language, turning their office into a "central kitchen" that creates new AI applications via modular development. Modular application development not only reduces development thresholds but also lowers the cost of development. In the past, a scenario transformation could cost millions of yuan, but now it only two to three hundred thousand.

Whilst 5G's high bandwidth and low latency make possible the efficient collection and transmission of data from a variety of different devices, AI steel models enable data to command production and to make effective decisions. For example, the AI prediction and optimization system can predict the temperature and composition of molten steel in real time and intelligently control the power supply and feeding, so that the whole steelmaking process can be completed within a specified time, while reducing the workload of workers. The intelligent coal blending system can better control the proportions of raw materials used, combining the state-of-art coal blending methodology to accurately predict the quality of incoming coke.

As of April 2024, Xiangtan Steel’s deployment of the Pangu Large Model covered 23 intelligent application scenarios in nine specialties, including coking, sintering, ironmaking, steelmaking, rolling, and quality inspection, within two categories of visual and predictive applications. These applications are used for production optimization and quality control. The large model has also helped them reduce energy consumption by 10%.

In the future, the group plans to implement more than 60 innovative application scenarios to achieve further cost reductions, quality improvements, and efficiency enhancements. Xiangtan Iron & Steel is just one of many manufacturers in traditional industries that will be taking advantage of the productivity unlocked by this new phase of AI-powered development.