Artificial intelligence at Metinvest — Case Study

by Anastasia Liberman
Wednesday, October 20, 2021
Artificial intelligence at Metinvest — Case Study

"Metinvest", metallurgy enterprise Ukraine, improves the efficiency of blast furnaces with cloud technology, big data, and machine learning

The company has implemented a pilot project using cloud-based technologies Azure Data Factory and Azure Machine Learning to improve the efficiency of blast furnaces. This is achieved by reducing fuel consumption by improving the accuracy of predicting and controlling the silicon content in cast iron. Metinvest plans to roll out the solution for all of its blast furnaces.

Why is it important to control the silicon content in cast iron

At the end of 2018, Metinvest launched a large-scale program aimed at improving the efficiency of its enterprises. One of the main aspects was to improve fuel consumption in blast furnaces with an expected value added of more than $100 million.
The fuel consumption in a blast furnace depends on several factors, one of which is the silicon content in the pig iron. The higher the silicon content, the higher the heating and fuel consumption. And reducing the silicon content by 0.1% can save up to ten kilograms of coke.
However, reducing the silicon content requires an accurate approach, since if it is sharply reduced, there is a risk of a drop in temperature and a shutdown of the blast furnace. This means that the accuracy of assessing the future fuel balance of the blast furnace is important.
Metinvest has launched a pilot project using artificial intelligence and machine learning technologies to predict the silicon content in cast iron over a time horizon of up to nine hours.

How it works

"We use Azure Data Factory as our primary tool for organizing the data integration process. Data is loaded regularly, and the model we developed in Azure Machine Learning predicts the silicon content in cast iron. Azure Machine Learning pipelines then run. These pipelines help get the data and run Python —scripts, including those responsible for data preparation. Machine learning models use this data to predict silicon content. The prediction results are loaded into an Azure SQL database," explains Alexander Perkhun, head of data management at Metinvest Digital.

"We use a Power BI dashboard to visualize the data, which is updated every hour. Our production team has access to the dashboard and adapts the key metrics to manage the furnaces as predicted. This helps maintain silicon levels in the desired range," explains Volodymyr Kravchenko, business transformation expert at Metinvest holding.

The implementation of the solution was accompanied by comprehensive measures, from adapting the algorithms for controlling the thermal conditions of the blast furnace to training operators and revising the goals and motivational component of the shop workers.

As a result, by the end of the year, the variability of the silicon content decreased from the historical 0.16% to 0.1%, which made it possible to stabilize the silicon content and obtain the necessary coke savings. Shortly, Metinvest plans to deploy a completely turnkey solution for all furnaces at the group's enterprises.

You will be interested