AI That Learns and Evolves Like Skilled Operators Reduces Iron Loss Rate [AI Autonomous Manufacturing, Opening the Future]

Removing Impurities and Adjusting Sulfur Content
AI Takes Over the "Pretreatment Process"
Real-Time Analysis and Control of Field Data
Still Short of Top Expert Skills
Plant Manager Yoon Jungkyun: "Continuously Learning and Evolving"

The pretreatment control room is located on one side of the 3rd Steelmaking Plant at POSCO Pohang Steelworks. Instead of molten metal, sparks, noise, and vibration, the site was filled with a calmness akin to absolute silence. Inside the control room, unfamiliar electronic devices and monitors connected to the field lined the walls, creating an atmosphere reminiscent of a CCTV control center. On one monitor, molten metal shimmered inside a ladle?a large vessel for holding molten metal?while a skimmer (an impurity removal device) scraped away the impurities floating on the surface. The skimmer would stop and then move again, tracing a path like that of an excavator. The tension of this complex process, where molten metal and impurities are intertwined, was palpable even through the monitor.

Inside the pretreatment process of POSCO Pohang Steelworks 3rd Steelmaking Plant. An automatic skimmer is operating on the surface of the ladle containing molten metal at high temperature to remove impurities. Provided by POSCO

The pretreatment process is the stage where impurities are removed from the molten metal coming from the blast furnace and the sulfur (S) content?which determines the quality of steel?is adjusted. At this site, the process is also referred to as "baejae" (impurity removal). Until now, operators had to manually control the skimmer using a console. Even for skilled workers, this is a highly demanding task that requires more than 400 repetitive operations. Since last month, artificial intelligence (AI) has taken over the controls that were once guided by the operator's intuition. This is why the pretreatment control room has become so quiet.

AI Mimics Skilled Operators' Senses, Reduces Loss Rate

AI has succeeded in completing the impurity removal process more quickly and stably than the average operator. In fact, AI completes the pre-impurity removal process in an average of 5 minutes and 30 seconds, and the post-impurity removal in 10 minutes, whereas human operators typically require about 6 minutes and 15 minutes, respectively. By keeping the process time consistent, the loss rate of iron (Fe) has also been reduced by 2?3%. Although this figure may seem small, site officials explain that it makes a significant difference in a steel mill where the process is repeated hundreds of times a day.

Currently, AI analyzes field data in real time at the main console in the control room. The control screen dynamically displays the ladle number containing the molten metal, the composition of the molten metal, and the position and operating status of the pretreatment equipment. A POSCO official said, "During the tilting stage, which tips the ladle, AI predicts the timing of the molten metal flow in advance, so the pretreatment equipment moves accordingly." On the day of the reporter's visit, when a message reading "impurity removal rate 90%" appeared on the screen during operation, the system immediately proceeded to the next step and stopped the skimmer. AI independently judged the situation and initiated the subsequent process.

Yoon Jungkyun, Head of the 3rd Steelmaking Plant at POSCO Pohang Steelworks, is explaining the AI autonomous process during an interview with Asia Economy. Photo by POSCO

However, the current capabilities of AI do not yet match the highest skill level of expert operators. If the impurity removal device is inserted too deeply, molten metal can be lost; if it is too shallow, impurities may remain and affect the quality of the molten metal. For this reason, the on-site AI is being advanced by learning thousands of operational patterns to mimic human senses. Since each operator has a slightly different way of removing impurities, it has taken a long time to collect and average these differences as data. By increasing camera resolution and recording the skimmer's trajectory, AI is gradually moving more like a human and improving itself. Yoon Jungkyun, the plant manager, emphasized, "There is no such thing as a completely automated process," adding, "Continuous learning and evolution are the essence of autonomy."

Even Small Errors Pose Major Accident Risks... The Answer Lies in 'Algorithms'

The most technically challenging stage is tilting the ladle containing the molten metal. If the angle is even slightly misjudged, molten metal can overflow, leading to a major accident that halts the entire process. To prevent this, POSCO has built a "redundant judgment algorithm" by combining high-resolution cameras with tilt angle data. AI checks field data through multiple channels and automatically stops operation if any abnormal signs are detected. Plant manager Yoon said, "If a person makes a mistake and molten metal overflows, the entire building could collapse. That's why we've introduced AI."

Looking at the screen in front of the console, you can see signals flowing in real time between the pieces of equipment. When the converter facility notifies the pretreatment equipment that molten metal will be ready in a few minutes, the pretreatment equipment waits, and the skimmer automatically prepares to remove impurities. All of these processes are possible without any radio communication from humans.