With our research, it was demonstrated that machine learning can be used to discover correlations between
many processing parameters within nonlinear and complex systems, such as melting steel scrap in an
electric arc furnace. Using a data-centric approach with machine learning analysis, we found the
relationship between the most influential parameters and the electricity consumption of melting steel
scrap in electric arc furnace. In order to manage such data-centric models, expert knowledge is essential,
because data management can better increase model accuracy than model tuning alone. This concept has been
proven in terms of energy consumption, steel utilization, melting time, and it can also be implemented in
terms of CO2 emissions and product quality.
Operators would use the online software to choose scrap
materials for charging the furnace (of which there are more than 14 types), and set key process parameters
that affect energy consumption, steel quality, and carbon dioxide emissions. Data would be collected in
real-time and sent to the server so that the model could be further improved. Furthermore, we found that
the addition of new sensors will improve the model's weaknesses by gathering new data. Deploying such
sensors can be simple to install in existing infrastructure and useful for improving model performance.
FIND OUT MORE
ABOUT OUR SOLUTION
SERVICE
Expert analysis of the current state of data collection, with the goal of improving data infrastructure and obtaining more data for our models based on our knowledge. More data improves model accuracy and opens up new possibilities, such as generating a CO2 report for each process stage and identifying manufacturing weak points.
- Analysis of the current state
- Standardization of all data types
- Formation of the main data collection server
- Expert analysis of the data from a metallurgical point of view
PRODUCT
Online software based on data-driven machine learning methods for predicting important process parameters,
reducing energy consumption and CO2 emissions, improving product quality, and making products more
competitive in large steel industries.
Online software for controllingimportant process
parameters.
Software that looks for the influence of impurities and the quality of the input raw materials in
the material on the quality of the product.
For process monitoring and control, we use cutting-edge data-driven machine-learning methods. Our service
is
founded on a multidisciplinary team and scientific principles, and includes the following components:
1. Client consultation regarding data infrastructure adaptation;
2. Creation of a model unique to
each
customer(because different customers have a combination of different equipment);
3. Real-time
monitoring of process parameters with indicators for the main items essential to the client (energy
efficiency, product quality, saving raw materials, CO2 emission report). Also, our model links parameters
between different processing units, so that causal parameters can be found (e.g., chemical composition
with
tensile strength).
THE METHOD
We use an agile method to develop our software:
- We take data from the industry, process the data and connect the most important parameters for the set goal. We look for connections between data, clean data, remove outliers.
- We build a model to predict the use of the most suitable machine learning model.
- We influence the process in real time, check the model, install new sensors to collect data.
Our multidisciplinary team does not view machine learning as a black box, but carefully analyzes the relationships between parameters from a technological, economic, and environmental point of view. Once the MVP is established (online software), our team can successfully incorporate new user requirements in an agile way, which implies, a circular process of: data processing, model tuning, and analysis of results. In this way, the software would make the best possible combination of steel scrap to enhance the steel making process for energy conservation and production efficiency and obtain steel of better quality.
Production data is very scattered and difficult to obtain correlation using conventional methods.
We find the connection of the influence of certain parameters on the target parameters in the process.
We predict the target parameter with high accuracy (and low mean absolute error).
TEAM OF EXPERTS AND KNOWLADGE
This multidisciplinary but complementary team consists of an assistant professor who specializes in
metallurgy
and who cooperates with Metalfer, which is where the idea for the software that is the subject of this
project
came from. Vaso Manojlovic is a participant in a large number of projects of the Ministry of Science and
Technological Development. A concrete confirmation of the subject innovation is the published work on the
same
topic in a top scientific journal. This also indicates the novelty in this sector of industry. Jelena dealt
with waste and scrap iron, has 3 years of work experience in a managerial position in the company with the
main focus on efficient management of waste streams, environmental methods of waste recycling, studies.
Jelena is also a PhD student at the Faculty of Technology and Metallurgy with the topic of circular
economics
in the metallurgical industry. Jelena and Vaso won second place in the competition for the best
technological
innovation in Serbia in 2016, so the team members got to know each other's work.
Milan Dotlić, PhD,
specialist in applied mathematics and machine learning. For many years he worked at the Jaroslav Černi
Institute, participated in a large number of science projects and published a large number of scientific
papers in the field. What drives us are the preliminary data obtained for the model as well as their initial
confirmation in practice, ie the model follows a real system. The technical side of the story is led by
assistant professor Vaso Manojlovic, and each member of the team has done so far in his segment: problem,
solution, simulation model, negotiations with users, testing. What will keep the team together is the common
faith in the future of machine learning and artificial intelligence in large systems, the great savings that
are being made, the environmental side of the story and the encouragement of such initiatives in the
Republic
of Serbia.
MILAN DOTLIĆ
Machine learning scientist
VASO MANOJLOVIĆ
Team leader
JELENA IVANOVIĆ
Manager
What drives us are the preliminary data for the model as well as their initial confirmation in practice, ie the model follows a real system.
References
METALFER Steel Mill is group of companies involved
inmetallurgy, energy and trading. Metalfer steel mill owns and operates EAF basedsteel mill producing
construction steel products. Metalfer invest is responsible for development of new projects concentrating
on energy generation, environmental projects and new technologies.
Research article:
Manojlović Vaso, Željko Kamberović, Marija Korać, and Milan Dotlić. "Machine learning analysis of electric
arc furnace process for the evaluation of energy efficiency parameters." Applied Energy 307 (2022):
118209.
Other projects: MATERIAL DESIGN. Currently we are working on design of biocompatible Titan
alloys using machine learning methods.
We have one Ph.D. researcher working on this issue. We found the influence of certain elements on the modulus of elasticity in Titanium alloys.