At a steel production base of Shandong Iron & Steel Group, glowing red billets emerge from the reheating furnace as hundreds of tons of rolling mill equipment rumble into action.
The billets are precisely gripped by rollers, stretched, and thinned until they become sheets of steel.
Five kilometers away, in an intelligent control center, computer screens display real-time data on more than 100 metrics from the blast furnace, including temperature, flame intensity, oxygen lance vibration amplitude, and flue gas composition analysis.
Powered by the Shangzhi artificial intelligence platform, the system monitors equipment operating indicators around the clock and automatically alerts staff whenever anomalies occur.
Today, the steel industry is in a period of contraction and adjustment, with companies shifting their focus from expanding production capacity to improving efficiency.
Where is the room for cost reduction and efficiency gains? Artificial intelligence is providing new answers.
Sun Wei, deputy general manager of Shanding Yunke Information Technology Co., Ltd., said that all application scenarios of the Shangzhi AI platform generate more than 100 million yuan in annual revenue, serving as strong support for cost reduction and efficiency improvement at Shandong Iron & Steel Group Co., Ltd.
Cutting costs and boosting efficiency: working out three key accounts
How does Shangzhi save that hundred-million-yuan per-ton steel account? The first saving comes inside the steelmaking furnace.
In the past, the molten pool state at the end of converter smelting was invisible, and the furnace chief relied entirely on experience, judging the endpoint by "watching the flame color." Experience varied, operations fluctuated, and a slight misjudgment in timing drove up costs.
The dynamic smelting intelligent agent of the Shangzhi metallurgical large model integrates flue gas, video, audio, and process data, and the system automatically generates and executes optimal operating instructions: the converter endpoint temperature hit rate remains stable above 95%, the endpoint carbon content hit rate exceeds 90%, a single furnace smelting cycle is shortened by 2 to 3 minutes, consumption of auxiliary materials such as lime and alloys drops by 2% to 5%, and comprehensive per-ton steel production costs fall by 3 to 6 yuan.
The second saving comes in the raw material yard.
The bulk of steel costs lies in raw materials. Coal blending and ore blending used to rely on process personnel estimating by experience, and a calculation sheet took half a day or even a full day.
The coal blending and ore blending intelligent agent integrates multi-objective optimization algorithms, compressing a single calculation to three minutes, raising the approval rate of blending schemes from 60% to 95%, and saving another 0.5 to 2 yuan in per-ton steel raw material costs.
The third saving comes from efficiency and people.
At the 4,300 production line of the Rizhao base, steel billets at over 900 degrees Celsius pass between rolling mills, and the AI vision system automatically completes steel turning, saving an average of two seconds per turn and reducing operator labor intensity by more than 80%.
In office operations, AI intelligent agents reduce internal repetitive work by more than 80%, and automatic contract signature and seal verification blocks the high-frequency risk of incorrect or missed seals; in the machinery plant, the automatic filing robot takes heated pipe fittings from the furnace, positions them, and processes them in one continuous flow, reducing the per-piece cycle time by 20% to 30% compared with manual work and expected to generate about 1 million yuan in annual comprehensive value creation.
Turning experience into models so AI understands metallurgy
Industrial large models are everywhere, so why has only Shangzhi been able to save this money? The difficulty of cost reduction and efficiency improvement reflects a long-standing problem in the steel industry's digital transformation: the lack of high-quality standardized data assets and insufficient accumulation of tacit process experience.
Veteran workers' judgment of flame color, sensitivity to abnormal equipment sounds, and memory of abnormal quality cases once existed mainly in individual minds, and after staff turnover, experience inheritance can easily break down.
Shanding Yunke was spun off from Shandong Iron & Steel Group's automation and informatization team and has more than 40 years of accumulated metallurgical industry experience.
Drawing on this foundation, the company sorts frontline process experience sentence by sentence, organizes it into a knowledge base, and then transforms it into computable, iterable digital process assets.
The Shangzhi AI platform adopts a dual-engine architecture of "general large models plus industry small models," supporting unified management of access to multiple large models such as DeepSeek and Alibaba Bailian while also accumulating four types of datasets downward: operations, production, industry knowledge, and multimodal data, paired with a low-code intelligent agent development platform.
In response to the data security concerns most sensitive for steel enterprises, Shangzhi AI Platform 2.0 uses the Zhilian system to unify industrial protocols, securely enabling data interaction between internal and external networks so that production data does not leave the plant area, providing a security foundation for industrial large models to enter core production links.
General models provide basic capabilities, industry models understand metallurgical processes, the knowledge base carries enterprise experience, and intelligent agents connect specific scenarios.
Models, data, and applications continuously cycle on the production line, forming a closed loop from identifying problems to providing solutions and then verifying results.
With models, data, and scenarios integrated as one, bottlenecks become momentum.
Expanding scenarios to amplify intelligent benefits
At present, the Shangzhi AI platform has incubated 128 typical scenario applications, covering the full process of production, energy, logistics, supply chain, and customers, of which 105 scenarios have already been deployed and are running at Shandong Iron & Steel Group.
In 2026, Shandong Iron & Steel Group launched an artificial intelligence "100-day action," proposing a "ten, hundred, thousand" scenario multiplication plan: completing the deployment of 100 "AI+" scenarios within the year, reaching 1,000 by the end of the 15th Five-Year Plan, while advancing full-stack Xinchuang transformation of the computing power foundation.
In September 2026, the Shangzhi AI platform passed relevant evaluations by the China Academy of Industrial Internet at the Global Industrial Internet Conference, earning excellent ratings in three core areas: the large model industrial application index, metallurgical safety knowledge Q&A, and metallurgical production knowledge Q&A.
Previously, the Shangzhi metallurgical large model was included in "Research and Trend Analysis of China's Fifty Industrial Large Models (2025)" and became one of the typical cases of industrial large models in Shandong Province.
When "watching the flame color" becomes "calculating the flame conditions," and when experience becomes iterable algorithms, Shandong Steel is forging a smarter, more efficient, and more sustainable new path on the road of high-quality development.
Comments