Severstal for the first time implemented the solution on the basis of AI at "Cherepovets iron and steel works"

On June 19, 2020 it became known that specialists "Severstal Didzhital" together with the Center of technology development, workshop of etching of production of flat-rolled products of "Cherepovets iron and steel works" improved the model managing the speed of continuous and etching unit No. 3 (NTA-3). The intelligent agent at the heart of which the algorithm of deep machine learning with a reinforcement (reinforcement learning, RL) performing functions of artificial intelligence was connected to it.

Ruban works along with the digital Adelina model which was set on NTA-3 earlier. Adelina calculates the speed of control of the unit, and the agent adjusts it for achievement of optimal result. It ensures to production process big flexibility and safety as model and the agent are capable to adjust unit speed every second and to react instantly to unforeseen situations.

The intelligent agent of Ruban differs from classical models of machine learning in what studies not according to historical data, and independently, investigating the circle of the digital double of NTA-3. Speed of management of a technology part of the unit in many respects depends on parameters of the passing steel band – length, width and thickness of a roll, a steel grade, temperature of reeling of metal and others. Ruban studies on combinations of parameters which are created especially for it by generative and competitive network. Also it sets the plan of rolling and creates situations which were not in the history. For more effective training in an intelligent agent the system of encouragement and penalties was appointed. By experiments of Ruban looks for the solution at which the amount of encouragement as much as possible exceeds the amount of penalties.

The machine learning technology with a reinforcement for control of units was not widely adopted in the industry and, in particular, metallurgy yet. The artificial intelligence started on NTA-3 can become the first similar case to the Russian practice. The performance improvement recorded on NTA-3 after implementation of digital instruments proves that data driven of approach have a big future in the industry, and we move in the right direction, – the CEO "Severstal Didzhital" Boris Voskresensky commented[1]

Notes