Nornickel Teaches Computers to See Molten Metal
Not long ago, industrial computer vision was used mainly to check whether workers were wearing hard hats or to detect defects in finished products. Today, algorithms are taking on more complex challenges: learning to monitor high-temperature metallurgical processes in real time.
A striking example is a joint project by Nornickel and Russian IT company Reksoft. The partners have completed development and deployed a converter-tapping monitoring system at a metallurgical facility in Norilsk.
A High-Temperature Challenge
In metallurgy, one key process is converting. Inside a massive converter, molten metal boils at extremely high temperatures. The contents are periodically tapped in different batches: slag, a metal-rich matte, or blister copper. The main challenge is knowing exactly when one liquid phase ends and another begins. Previously, an operator had to make that call. If the switch came too late, valuable copper could end up in the waste stream.
Now, two specialized cameras are installed opposite the opening through which the molten material flows. They detect infrared radiation, or heat. Slag, matte, and copper each have different temperatures and distinct infrared signatures. Using algorithms, the machine-vision system identifies the material’s visual and thermal characteristics in real time. It tracks the stream, the ladle, and the converter mouth, then alerts the operator when, for example, slag is about to flow and a changeover is approaching.
The entire process is monitored without physical contact. Conventional sensors would melt or fail at such extreme temperatures, but infrared cameras can operate reliably under these conditions. Just as importantly for an operating plant, the system does not require a major retrofit of the converters themselves. The result is more copper recovered and less loss of valuable raw material.
A Digital Assistant for Metallurgists
The metals industry began adopting digital technologies several years ago.
In 2021, Nornickel introduced a system at its Kola Mining and Metallurgical Company facilities to monitor the use of personal protective equipment (PPE). The company installed 19 standard, non-infrared cameras across the site. They identified workers and checked for specific items, including hard hats, safety glasses, masks, and high-visibility vests. If anything was missing, the system immediately flagged a violation.
In 2022, EVRAZ applied video analytics directly to a production process, using it to measure the length of metal billets. Cameras monitored a moving, red-hot workpiece, while computer-vision software identified its edges and calculated its length from the image, quickly and without contact. Once the billet reached the required dimensions, the system issued a cutting command. This improved manufacturing accuracy and reduced material losses.
In 2023, the company introduced an intelligent metal-defect detection system on a rolling mill. Defects that go unnoticed can move further down the production line and compromise the finished product. Cameras capture images of billet surfaces, and a neural network analyzes them to automatically identify cracks, scratches, and other flaws. On a single rolling mill, the system generated more than 20 million rubles in economic benefits (about $238,000).
Nornickel and Reksoft later developed a flotation control system, which helps separate valuable metals from waste rock. Machine-learning and AI algorithms analyze camera images to assess the appearance of the froth, the movement of the flow, and conditions inside the flotation cell. Based on that information, the system can recommend adjustments or even change process parameters automatically. The companies have also introduced a contactless system for weighing bulk materials on a conveyor belt. Without touching the material, the software calculates its volume on the belt. That capability matters because accurate material accounting is essential for controlling the process effectively.
Nornickel’s “digital assistant for metallurgists” continues to expand its capabilities. Previously separate technologies, including infrared video analytics, spectrometry, and mathematical modeling, are being integrated into a unified system built around a single production process.
More Precise Control, Higher Profits
Today, Nornickel’s smart technologies monitor 80% of its core equipment, from mining operations through metallurgy, generating more than 10 billion rubles in annual economic benefits (about $119 million). By 2030, that figure is expected to reach 50 billion rubles (about $596 million).
The new system is expected to increase copper recovery into blister copper by 0.6 percentage points while reducing the amount of nickel and cobalt reporting to converter slag. At the scale of a major industrial operation, that translates into millions of rubles in additional profit, not from mining more raw material but from controlling the process more precisely. Video analytics can already provide information not only about the current phase of the molten material, but also about the condition of the converter mouth, the type of operation underway, and the amount of product being tapped. These data could serve as the foundation for a real-time digital picture of operations across the entire converter shop.
The developers emphasize that the technology is not limited to the Norilsk site. More than 10 copper and nickel converting operations in Russia and other CIS countries could benefit from a similar approach.
