Jet Infosystems using artificial intelligence were raised by efficiency of the rolling mill at NLMK plant

2020: Development and deployment of service for acceleration of work of a camp of hot rolling

On June 22, 2020 IT- the company "Jet Infosystems" announced development and deployment on Novolipetsk Metallurgical Combine service for acceleration of work of a camp of hot rolling. The mathematical model machine learning on which service is constructed helps operators to optimize process of rolling of metal on Camp 2000. The solution will help the enterprise to gain about 30 million rubles additional income a year.

The rolling mill in metallurgical production is an equipment which using rolling rolls rolls heated metal and turns it into sheet rolling. On the rolling mill give alloys of different chemical composition for steel production of different brands. The camp works continuously: day and night and without days off. That slabs did not face and did not connect, they should move with certain intervals. Than they it is less, those can roll more slabs for unit of time and the performance of all camp is higher. But at too small gaps between slabs the probability of defects and an abnormal stop of a camp therefore operators traditionally calculate distance with a stock sharply increases.

Recommendatory service based on machine learning (ML) processes a set of the changing parameters (composition of steel, heating temperature, characteristics of finished goods, etc.) and in real time provides to operators of the rolling mill recommendations about an optimal interval of giving of slabs and speed control of their movement. At the expense of it development of the rolling mill increases on average for 3.5 hours a month.

For training of a mathematical model which became a basis of recommendatory service specialists Jet Infosystems processed huge data arrays for the last 2.5 years. Experts investigated historical data from temperature sensors, pressure, motion speed and other recording equipment, and excluded parameters which have no significant impact on the motion speed of slabs on the rolling mill. In addition to indicators of the camp algorithms of machine learning consider in recommendations types and grades of alloys in procurements. All data on work of Camp 2000, as well as other information production, are stored in Data Lake which is also implemented by Jet Infosystems company.

Metallurgists of high qualification work at the NLMK plant. Traditionally they determined the rate of the movement of slabs on Camp 2000, based on experience and an intuition. And permanently high rates of performance of the plant speak about the level of their qualification. — Evgeny Kolesnikov, the director of the center of machine learning tells Jet Infosystems. — During the project the staff of NLMK shared with us subtleties of technology processes of the rolling mill. Having studied all processes, we created service which will help metallurgists to calculate the optimal rate of rolling even more precisely.

2019: Data Lake creation

On September 5, 2019 it became known that experts Jet Infosystems developed an analysis system of data and modeling for NLMK company (we SIT DOWN) which is responsible for availability and quality of data and also provides tools for calculations and the analysis.

In workshops and on units of the enterprise tens of systems which accumulate are operated and process information on production and technology processes. Besides, in premises tens of thousands of sensors which continuously collect are installed and transfer data. The convenient tool for work with the obtained data under the developed models of machine learning (ML) was required for division of Data Science of NLMK.

For the solution of a task the Center of software solutions of Jet Infosystems IT company together with a command of NLMK was created by Data Lake with a capacity of 300 terabyte based on a domestic distribution kit of the distributed platform of storage Hadoop. For solving of tasks of collecting, transfer, transformation and data storage such services as Apache Kafka, Apache NiFi, Apache Hive were used.

Within the project specialists started regular data loading in Data Lake from more than 70 sources (sensors and also MES and an APCS), loaded historical data for the last few years of work of the enterprise and developed medium maps technology and production processes of separate workshops.

Within the project the command Jet Infosystems developed for NLMK model of the unified data mart and also loading processes in it, implemented management of metadata of means of Apache Atlas (tagging, search, etc.), configured the centralized role model and its integration page Active Directory. It gave the chance date-sayntistam of NLMK more quickly to get access to data necessary to them in Data Lake.

For control of work of Data Lake complex condition monitoring of services of a system in Zabbix was also configured and also autotests are developed for control of integrity and completeness of data. For especially important and vulnerable data the possibility of backup was created: i.e., in case of inadvertent destruction of data by the user they can be recovered.

Creation of the lake of data which really brings economic benefit to the customer it is a difficult task both from technical, and from the organizational point of view. And here matter not only in iron, integration and programming. Readiness of production to collect and transfer qualitative data, and then to use analysis results of these data for adoption of production solutions is very important. In cooperation with colleagues from NLMK we managed to create quickly the solution which made production really digital emphasizes the director of development and deployment of the software Jet Infosystems Molodykh Vladimir

The implemented system already allows NLMK Group more effectively than a message development of digital services and models of machine learning for optimization of production processes^[1].

Notes