Steel Industry Smart Manufacturing: Lessons from Automation History

The steel industry has long been at the forefront of digitalization, achieving remarkable feats in automation and process control decades before terms like Industry 4.0 became buzzwords. Yet, as new technologies emerge, many steelmakers find themselves at a crossroads, grappling with how to apply modern smart manufacturing concepts in a sector where the fundamentals differ sharply from discrete manufacturing. This article delves into the historical context, current challenges, and strategic directions for intelligent manufacturing in steel, emphasizing the need for tailored approaches that prioritize economic value over technological novelty.

Historical Milestones: Steel’s Digital Pioneering

In 1978, China’s total steel output was merely 31.78 million tons, with per-worker productivity below 15 tons annually. Fast forward to today, and the country consistently produces around 1 billion tons per year, with per-worker output exceeding 600 tons—some advanced mills even reach 2,000 tons. This leap would have been impossible without automation. Beyond direct labor savings, automation acted as a catalyst for other technological advances: continuous casting replaced ingot casting, basic oxygen furnaces supplanted open-hearth furnaces, and equipment scale grew dramatically, all enabled by digital control systems.

As early as the 1960s, digital technologies like PLCs and DCS—the hallmarks of Industry 3.0—were being integrated into steel plants. When construction began on a major Chinese steelworks in 1978, these systems were already extensively deployed. In steel, PLCs and DCS form Level 1 (L1) systems, while Level 2 (L2) process control systems functioned as early precursors to today’s industrial internet platforms, running mathematical models that are essentially industrial apps. These systems dramatically improved control precision, enhancing steel purity, rolling tolerances, and energy efficiency.

Consider the environmental impact: in the early reform era, producing one ton of steel consumed 1.2–1.5 tons of coal and 30 tons of water. Without automation-driven improvements, today’s output would require 1.2–1.5 billion tons of coal annually and generate over 30 billion tons of CO₂, while water consumption would equal the entire annual flow of the Yellow River into the Bohai Sea. Automation made these gains possible.

High control precision also enabled product diversification and customization. Advanced mills pioneered “make-to-order” production, delivering steel tailored to exact customer specifications—a practice that boosted margins by hundreds of yuan per ton. However, this introduced immense complexity: with small batch sizes and varied requirements, manual production management became untenable. A Japanese recollection noted that at 2 million tons annual output, manual management barely sufficed; at 6 million tons, it was impossible.

To cope, early digital integration was essential. In hot rolling, where a new coil is produced every 1–2 minutes with different parameters, computer-controlled parameter switching and real-time tracking of every slab and contract were critical. This required linking management and control computers across processes—a concept dubbed “data never touches the ground,” ensuring data integrity without human intervention. Ingenious solutions using RS232, modems, and even floppy disks achieved this long before modern industrial internet platforms existed. By the late 1990s, massive investments in mainframe computers integrated production, supply, and sales, enabling weekly delivery commitments and winning top honors in national manufacturing informatization rankings.

New Era Challenges: Why Steel Must Chart Its Own Course

Despite this head start, many steel enterprises entered a period of confusion in the 21st century. The reason is subtle: steel’s digital maturity means many Industry 4.0 concepts, born in discrete manufacturing, don’t translate directly. For instance, mass customization on assembly lines—a hallmark of Industrie 4.0—has existed in steel for decades as make-to-order production, because product switching mainly involves parameter changes easily automated. With automation levels already high, further gains in blue-collar productivity are marginal; the real opportunity lies in boosting white-collar efficiency.

Different nations have pursued distinct paths. In the US, where labor costs are high (GDP per capita around $80,000), the focus is on maximizing tons per worker. One US mill achieved nearly 4,000 tons per worker, double that of some Asian peers, but at the cost of product diversity and quality. In contrast, South Korea’s steelmakers, renowned for quality, aim for unmanned smart factories to eliminate human error and achieve unbeatable cost and quality competitiveness. For Chinese steel, with relatively lower labor costs, the priority should be enhancing white-collar productivity and leveraging digital tools for resource sharing, knowledge reuse, and management improvement, rather than blindly pursuing automation for its own sake.

China’s steel sector employs about 1.6 million directly, but supporting industries employ four times that number—a vast area where automation can yield significant value. Moreover, with half the world’s steel production, China has immense potential for personalized customer services and energy conservation. While energy efficiency gains may approach physical limits, pollution reduction offers endless opportunities. Digital platforms can also integrate resources across enterprises, simplifying product mixes at individual plants and enabling best practices to be replicated at near-zero marginal cost—a key advantage of knowledge products like software and management models.

Strategic Directions: AI, IIoT, and the Service Transformation

The industrial internet is ushering in a management revolution. Where Japanese management once tried to make people work like machines, digital systems can now assist, replace, manage, and supervise human workers, compensating for weaknesses in execution. Many industrial internet competition entries focus on energy management, safety, quality, and logistics—essentially elevating management practices. The Japanese Industrial Value Chain Reference Architecture (IVRA) frames this as PDCA for resource management, enabled by digital means.

For steel, the most promising AI applications lie not on the factory floor but in service and market-facing functions. Large language models excel in open-ended, personalized problem-solving—exactly what’s needed to serve diverse downstream customers. Instead of chasing autonomous production, steelmakers should build knowledge bases that AI can leverage to provide expert-level advice to thousands of small and medium enterprises. This “manufacturing to service” shift, long envisioned but hampered by a shortage of human experts, now becomes feasible with AI.

Data analytics is another high-impact area. “There is gold in data,” but manual analysis is slow, costly, and often unsuccessful. AI-driven analytics can dramatically improve efficiency, as demonstrated by companies like Palantir. Combining large models with data analysis tools can empower analysts to uncover insights faster, turning data into actionable intelligence.

Avoiding Pitfalls: Common Missteps in Steel Digitalization

A frequent mistake is equating smart manufacturing with deploying robots. Automation should create value, not just replace labor at any cost. If a robot’s capital cost exceeds the wages it saves while also eliminating jobs, it’s a net loss to society. Similarly, many digital dashboards display data without context—showing output but not whether it’s high or low, energy consumption without trend analysis. Information must drive action to have value; otherwise, it’s waste.

Applying technologies in unsuitable scenarios is another trap. Large AI models thrive in open environments with unpredictable inputs, like customer inquiries or road conditions. But in closed-loop industrial control, systems are designed to reject disturbances and operate within narrow, predictable ranges. Here, simpler models often suffice, and introducing large models may add risk without benefit. The role of AI in production should be as a strategist, not a frontline soldier—supporting decisions, not executing real-time control.

Digital twins and cyber-physical systems (CPS) are powerful concepts, but forcing them into steel without understanding the underlying logic leads to superficial implementations—like replacing simple icons with 3D animations, violating the industrial principle of simplicity. The real value of CPS in steel lies in managing the complexity of digital factories, which may contain hundreds of millions of lines of code. Properly structured digital twins can ease software maintenance, but this doesn’t require flashy graphics.

The divergence between discrete and process manufacturing stems from fundamental differences in how opportunities arise. Moore’s Law enabled the storage and processing of massive design data, revolutionizing discrete industries through generic CAD/CAE tools. In steel, however, process models rely heavily on empirical parameters, making universal simulation software less effective. The challenge is not computation but accurate parameterization—a scientific, not purely mathematical, problem.

The steel industry’s digital journey is far from over, but it requires clear-eyed strategy. By learning from its own pioneering past, focusing on genuine needs, and adapting new technologies to its specific context, steel can continue to thrive in the era of smart manufacturing.