Severstal Improves Mill 5000 Reliability with Machine Learning

2025

Implementation of ML model to prevent emergency downtime due to motor failures

At Mill 5000 of the Kolpino production site of Cherepovets Steel Mill, a machine learning model has been introduced that helps to avoid emergency downtime of the unit due to breakage of electric motors. The solution was developed by the specialists of Severstal Digital and Severstal-Infokom together with the technologists of the sheet rolling shop and experts from the center for technological development of pipe rolling production. Severstal announced this on August 7, 2025.

𝓲

Фото: Severstal

The rolling process is carried out in the horizontal reversing stand of Quatro "5000." Its operation is provided by two electric motors for which the torque load must be calculated. The motor force moment, or torque, is a physical quantity that characterizes the rotational force generated by the motor. It determines its ability to overcome resistance and rotate a shaft or mechanism. The higher the moment of force, the greater the load that the engine can overcome at a given rotational speed. However, at critical values, the risk of breakage of the shaft lines of the main drive of the cage increases, which leads to its failure and downtime of the entire unit. Previously, the operator set the settings of the cage, relying on his own experience, which was not optimal. For August 2025, the moment of force indicator is calculated using a machine learning model.

Before starting rolling, the control station operator sets the deformation mode parameters, and then clicks the Calculate Mode button. Based on these data, the program calculates clearances and compression along the passes, after which they, together with the parameters of the deformation mode, are transferred to the model. The model predicts the average and maximum moments of power of the cage motors and displays the results in the user interface as a value in kilonewton meters (kNm) with color indication. If the calculated motor moment exceeds the specified limit, it is highlighted in yellow or red. In this case, there is a possibility of damage to the equipment. This means that the operator must change the parameters of the deformation mode so that the design moment of the cage motors does not exceed the permissible threshold, or take other compensating measures to reduce torque loads during actual rolling.

To predict the moments of force of the Stan 5000 cage motors, a gradient boosting model was used, as well as separate models for forecasting on final and rough passes. Models are trained on actual rental data.

Our solution provides accurate adaptation to changing loads, increasing energy efficiency and reducing equipment wear. Already in the first half of the year, this made it possible to reduce the emergency downtime of the 5000 camp by 3.9 hours with an economic effect of 4.53 million rubles. At the same time, the system has an intuitive interface familiar to the operator, - said Svetlana Potapova, director of Severstal Digital.

Implementation of metal surface inspection systems

Severstal continues to scale computer vision models that automate the quality control of rolled metal. As of March 2025, 12 systems were implemented at the Cherepovets Steel Mill site. The solutions were developed by the Severstal Digital team (Artificial Intelligence cluster), together with the Technical Development and Quality Directorate and the Severstal Repair Directorate. The company announced this on March 26, 2025.

Both neural network systems based on previously installed video surveillance cameras and new surface inspection systems (PIS) based on the company's own designs have been introduced at the Cherepovets Steel Mill site.

So, at the end of 2024, four CIPs were put into operation: they work on the units of polymer coatings of metal No. 1,2,3 and on the unit of longitudinal cutting No. 8. These solutions also work on continuous hot galvanizing units No. 1 and No. 4 and on longitudinal cutting unit No. 4. They are capable of recognizing 40 to 100 classes of defects in color images, making it easier to classify deviations. The systems use a set of computer vision models, each trained to detect defects critical to a particular type of metal. When changing the type of products, for example, from cold-rolled steel to dynamic steel, the algorithm automatically switches to the desired model, and additional rules for classifying defects are flexibly adapted to the type of rolled metal.

In addition, the developed software provides adaptive lighting adjustment depending on the time of day and color tone of the coating, which allows you to obtain a better image for further recognition of defects by the neural network model. This is especially important for systems on polymer coating units, where up to several dozen color versions of polymer rolled metal are produced.

Computer vision solutions are an important part of the digital product qualification system. Due to their accuracy and ability to work 24/7, Severstal customers receive better metal products, and unit operators can switch from routine tasks to those that require greater involvement. The combination of multimodel algorithms, adaptive technologies and import-independent solutions not only improves product quality, but also forms a benchmark for the metallurgical industry. Over the past two years, we have increased the accuracy of solutions by more than 30%. In the future, we plan to introduce inspection systems at other key units, transferring entire production lines under digital control, "commented Svetlana Potapova, Director of Severstal Digital.

2023: Increasing Mill 2000 Productivity with Machine Learning

On May 11, 2023, Severstal announced the introduction of a software complex for managing the rate of rolling and issuing slabs from furnaces based on machine learning models. The solution, called Autotemp 2.0, was implemented at Cherepovets Steel Mill Mill 2000 (a key asset of Severstal). The developers were specialists from Severstal Digital and Severstal-Infocom together with experts from the Center for Technological Development of Downstream Rolling Plants and technologists from the production of rolled steel at Cherepovets Steel Mill.

"Autotemp 2.0" allows you to calculate and adjust the optimal pause before delivering slabs from the heating furnace of the mill and thereby increase its productivity. Earlier, the operator calculated the necessary time for removing the slab on his own, which could cause unproductive pauses in rolling. In addition, the solution is integrated with the metal heating model, which improves the energy efficiency of the heating furnace area and the quality of slab heating.

Autotemp 2.0 is based on a model using a gradient boosting algorithm that allows you to analyze tabular heterogeneous data and calculate the rolling time of metal in the mill with high accuracy. For three months of the solution, the savings in rolling time due to the optimization of pauses amounted to 27 hours, which made it possible to additionally produce 24 thousand tons of rolled metal.

"Autotemp 2.0" is Severstal's own development, which used current technologies and tools for building digital solutions. They allow you to develop the product in the future and not depend on foreign suppliers of software or components. The solution completely controls the issuance of slab, due to which we eliminate the human factor and reduce the likelihood of downtime of the unit. In addition, we are introducing a digital twin at the heating furnaces site in order to determine, based on its forecast, which of the existing furnaces is best loaded at a certain moment for the best productivity of the 2000 mill, "commented Evgeny Vinogradov, CEO of Severstal Russian Steel and resource assets.

The level of digitalization of Severstal's production sites is steadily growing, and Autotemp 2.0 is another successful example of how a digital solution helps improve the productivity of industrial equipment. Our developments cover increasingly large and powerful units, including Camp 2000, one of the key units in the process chain of the plant, "commented Svetlana Potapova, Director of Severstal Digital LLC.