GOALS AND OBJECTIVES
To improve the efficiency of the rolling mill at NLMK’s plant.
To develop and implement a new, machine learning-based (ML-based) recommendation service.
- A mathematical ML-model was used to create very significant efficiencies.
This service, which was developed by Jet Infosystems specialists, is based on a machine learning model (ML-model) which analyzes data from several sources. The model was trained using a data array collected over the past 2.5 years. This data included: historical temperature, pressure and speed sensor data, as well as data from other recording equipment. Machine learning algorithms simultaneously take into account the types and grades of alloys in the semi-finished products, while excluding parameters that do not have a significant impact on the speed of slab movement along the rolling mill.
The new service processes a set of changing parameters (steel composition, heating temperature, characteristics of finished products, etc.) and provides rolling mill operators with recommended optimal slab feed intervals and advice on how to control rolling mill speed in real time. As a result, NLMK’s rolling mill output has been increased by an average of 19.5 hours per year.
All Mill 2000 operations data, as well as other production information, is stored on a Unified Digital Platform (EDP) implemented by NLMK Group specialists; ML-models are retrained on a DSML platform.
Time over which the data used to train the ML-model was accumulated
19.5 hours per year
Increase in rolling mill output
30 million rubles
Economies garnered from the applied solution