Engineers at Severstal Digital, a high-tech subsidiary of Severstal, one of Russia’s leading metallurgical and mining businesses, earlier this summer introduced advanced artificial intelligence at one of its flagship steel-making assets, Cherepovets Steel Mill about 375km north of Moscow (part of Severstal Russian Steel Division).

A digital model that controls the speed of a local continuous pickling line (NTA-3) was enhanced with a sophisticated smart agent called Ruban, based on reinforcement learning (RL) algorithms.

The new intelligent agent differs from classical machine learning models in its ability to learn all on its own instead of feeding on historical data. The system explores a digital twin of NTA-3 and learns from a combination of different parameters specifically created for it by a generative adversarial network (two neural networks that generate new data).

Severstal Russian Steel Division CEO Evgeny Vinogradov believes the innovation has already lived up to the company’s expectations, boosting NTA-3 productivity by more than 5%.