How machine learning helps steelworkers. Case NLMK and Jet Infosystems

IT solutions based on artificial intelligence are actively penetrating various areas of industrial production. For example, at the Novolipetsk Metallurgical Plant (NLMK), the AI workplace is next to the steel mill that smelts steel.

Specialists from Jet Infosystems joined the task of turning artificial intelligence into a steelworker. They literally trained the software system - it works on the basis of computer-based training.

Why do artificial intelligence understand steel brands

To obtain steel grades with improved characteristics, for example, designed for specific operating conditions, additional chemical elements are introduced into its composition during smelting. The most common method is to introduce special materials in the form of an iron alloy with one or more chemical elements (silicon, manganese, chromium, etc.) into the liquid metal melt. They are introduced in different periods of steel smelting and processing. For example, ferronickel is introduced in the first period - oxidizing, due to the fact that nickel is not oxidized in the oxygen converter, but contains hydrogen, which, when heated, turns into gas and is removed in the second stage during the boiling of steel. But ferroniobium and ferrovanadium are well oxidized, so they are introduced at the final stages of steel processing.

Steel brands differ in chemical composition, for which in each case there is a clear "recipe." However, the processes occurring in the smelting and processing of steel are so complex that the addition of a certain amount of ferroalloy does not guarantee accurate entry into the range of permissible chemical parameters. There are many reasons why it is not possible from the first time to get into the "gate" of tolerances, for example, there may be too much oxygen, etc. This process is somewhat akin to art, only instead of creative inspiration experience here. If an experienced specialist works in this section of metallurgical production, who meticulously studied all the smallest details of the processes, he more accurately takes into account the smelting parameters and gets the desired result faster than a beginner who only learns the profession - he has to undergo several iterations, adding ferroalloy and conducting chemical analysis of the resulting metal.

NLMK Converter Shop

The task that the plant set is understandable: to determine the minimum required amount of ferroalloy in order to fall into the given interval in terms of chemical composition, to use the minimum amount of the materials themselves, and if you can choose, then apply the cheapest material, "says Yevgeny Kolesnikov, director of the Machine Training Center of Jet Infosystems.

How to optimize the steel smelting process

From the point of view of digital solutions, this is a classic decision support system that is designed to tighten all workers to a single level of skill and at the same time optimize production processes, "explains Yevgeny Kolesnikov. - Of course, at any enterprise there are real talents, venerable professionals - they are able to work more accurately computer than the system. But on average, machine algorithm offers more optimal solutions, as a person tends to be reinsured.

The smelting optimization solution at the first stage of the project was tested at the site of the first converter shop NLMK on a limited steel grade. The solution has two key components:

Chemical composition prediction system. The "heart" of the system is a mathematical model based on machine learning algorithms (ML), which predicts what the chemical composition will be if certain materials are added at a particular time.

This made it possible to create a model that works with great accuracy in the forecast mode "what if," says Evgeny Kolesnikov.

NLMK stalwart is helped by smart computer programs

From the point of view of mathematical statistics, this allows us to achieve higher accuracy of the forecast, "the expert continues.

With this digital tool, a large number of different ferroalloys can be quickly searched, and those combinations can be selected to produce the greatest savings.

At the first stage of the project, we conducted a deep study of production processes, "says Yevgeny Kolesnikov. - We learned the target chemical composition of steel and ferroalloys, their interchangeability, melting route. Received data on the value of each ferroalloy, and the balance of materials in the warehouse. Based on all this data, we taught the system to select a combination of ferroalloy components that has the lowest cost, but gives the desired chemical composition.

Technical implementation

The advisory service is implemented as a steelworker's AWS. The recommendations are displayed on the monitors located in the control rooms of the various units of the workshop. As part of the project, the new system has been integrated with external data sources, including SAP and APCS systems.

A key element of integration mechanisms - the connection with a single enterprise repository of "raw" Data Lake data of the entire mill - it stores all the historical data of NLMK.

In addition, last year the company launched its own data analysis and modeling system (SADiM). The architectural solution of SADiM is based on the Data Lake structure, which allows you to obtain data in a "raw" form and save it for further processing in the simplest and most economical way.

Control of steel smelting at NLMK resembles the work of the situation center

The IT infrastructure already created at NLMK made it possible to implement the project of introducing the recommendation service as a regular system integration service, no special problems arose, "emphasizes Yevgeny Kolesnikov.

In 2019, the system operating in converter shop No. 1 was put into industrial operation. NLMK has already begun to develop a similar recommendation service at the second site - in the converter workshop No. 2. According to the estimates of the plant management, the expected economic effect when expanding the service by the maximum volume of grade in two converter production workshops may amount to 100 million rubles. per year.

The company "Infosystems Jet" says that the service-assistant developed for NLMK in the optimization of smelting has very good prospects for replication to other iron and steel enterprises. RUSSIAN FEDERATION According to company experts, about 100 metallurgical companies in the country are used in the production of ferroalloy.

Physicochemical smelting processes do not depend on specific production regulations, nor on the form of ownership, nor on the size of the plant, "emphasizes Yevgeny Kolesnikov. - Even a small steel production is a billion-dollar turnover, and the savings on ferroalloys are very tangible for it. This opens up good prospects for introducing such an IT solution, both at domestic steel enterprises and abroad.