Emanuel Kashi Thienpont, VDEh-Betriebsforschungsinstitut GmbH, Germany

Co-Author:
Tobias Kleinert, RWTH Aachen University
Elmar Schuster, voestalpine Stahl Donawitz GmbH
Kerstin Walter, DK Recycling und Roheisen GmbH
Thorsten Hauck, VDEh-Betriebsforschungsinstitut GmbH
Stefan Wienströer, thyssenkrupp Steel Europe AG
Ralf Schwalbe, thyssenkrupp Steel Europe AG

Abstract:
Sinter with high and consistent quality, produced with low costs and emissions is very important for iron making. Transport and storage degrade sinter quality, generating fines and segregation effects. Conventional sinter quality monitoring is insufficient as it is slow and expensive. Therefore, short-term fluctuations of sinter quality can only be represented inadequately. It is therefore difficult to draw conclusions about the actual process parameters of sinter production or they are based on empirical experience. Consequently, also a data analysis between sinter quality and its effects on daily blast furnace operation is extremely non-transparent and in need of optimization. In this work, a new approach will be introduced to strengthen the data base that describes the sinter quality in short term periods at two different sinter plants from VASD and DK. New on-line measurements will be established, combined, and analysed with Big Data technologies and should allow short term conclusions of the produced sinter. Based on this comprehensive data compilation, Machine Learning (ML) algorithms are trained to predict sinter quality indices. In this paper, the development of the novel measurement systems, the data preparation, feature selection and first results of the predicted indices are presented. This break-through in continuous high resolution quality monitoring should help to get continuous quality indices for sinter and will support the combined optimization of sinter plant and blast furnace.

Sudhanshu Kuthe, KTH Royal Institute of Technology, Sweden

Co-Author:
Andrey Karasev, KTH Royal Institute of Technology
Rössler Roman, voestalpine Stahl GmbH
Izaskun Alonso Oña, Sidenor Investigación y Desarrollo
Björn Glaser, KTH Royal Institute of Technology

Abstract:
Considering the advantage of achieving rapid and efficient process control with artificial intelligence (AI), the use of applied data-driven strategies is in demand among today's steelmakers. One important application is the online optimization of material additions during ladle refining. As every steel grade is unique, developing distinctly trained machine learning (ML) models seem significantly demanding. A data-centric ML approach can solve this problem by focusing on what AI systems must learn from unique datasets generated by specific steel grade production. Moreover, such data-driven process models may be simpler to integrate into existing production servers compare to the physics-driven models. In the proposed work, an operator-assisted decision support system (DSS) was developed using applied data-centric ML to optimize calcium additions required for producing ultra-clean low alloyed calcium-treated steels. Real industrial process parameters were collected before and after calcium additions for three steel grades. The collected process data were examined to design a base algorithm by continuously monitoring the characteristics of non-metallic inclusions at various stages of ladle refining. The predictive result obtained using the proposed DSS, and the response obtained from the actual process were compared and verified for each studied steel grade. Special efforts were taken to generate self-adapting output for each produced "heat" of a specific steel grade to avoid any degradation in predictive performance. The optimizer named "ClogCalc" was introduced based on the programming architecture of DSS for utilization and integration inside the industrial environment. Initial attempts were made to deploy this novel optimizer inside the production environment. The authors believe that this proposed approach could support operators in making dynamic decisions to optimize calcium additions. Keywords: Artificial intelligence, Industry 4.0, Online-monitoring, Steelmaking, Machine learning

Emanuele Trucillo, Danieli & C. Officine Meccaniche s.p.a., Italy

Co-Author:
Stefano Pantarotto, Marcegaglia s.p.a.
Claudio Rossi, Marcegaglia s.p.a.
Alessandro Ferraiuolo, Marcegaglia s.p.a
Nobile Matteo, Danieli & C. Officine Meccaniche s.p.a.
Luciano Vignolo, Danieli & C. Officine Meccaniche s.p.a.
Alessandra Primavera, Danieli & C. Officine Meccaniche s.p.a.

Abstract:
Coating Finishing processes, such as Color Coating or galvanizing, require the product to be accurately pickled. The most common method to accomplish this is acid pickling. Danieli has a long history of proposing the Turboflo solution, which allows for acid pickling speeds of up to 400 m/min. A common problem in pickling plants, however, is the scarce level of process automation and the heavy reliance on on-the-spot decision. These decisions not only regard pickling speed, but also the un coiling speed of the input coil and its coiling speed at the output. One of the reasons for this reliance on spur of the moment decisions is the difficulty of pinpointing the actual duration of “dead times” such as welding and cutting phases. Therefore, a set of innovative sensors to characterize scale and quantify overpickling, feeding an MPC control architecture core are presented here, which attempt to optimize the three main “levers” for the process, which are process speed, uncoiling speed and coiling speed, together with other secondary levers such as bath temperature, recirculation speed and fresh acid flow. The architecture is coupled with statistical evaluations of dead times through modern Data Science approaches.

Matthias Schops, Aperam Genk, Belgium

Co-Author:
Emmanuel Placier, AMI Automation
Saul Gonzalez, AMI Automation
Thierry Koeger, AMI Automation
Bastien Soete, APERAM Genk
Matthias Schops, APERAM Genk

Abstract:
The complexity of the stainless-steel production process requires precise control of the electrical and chemical energy input to avoid material or energy losses and deviation from the required bath chemistry. Facing this challenge, an agreement was reached between Aperam Genk and AMI in 2022 to install the SmartFurnace EAF optimization system including the DigitARC PX3 Electrode Regulator and SmartARC for electrical energy optimization, and the Oxygen Module for chemical energy optimization. Aperam Genk in Belgium is a steel plant dedicated to the production of high-quality stainless-steel grades for the worldwide market. The AC Electric Arc Furnace with 120 tons capacity and 80 MVA transformer has also significant chemical energy available for assisting the scrap melting. The SmartFurnace system and the SmartKnB platform recently developed by AMI integrates a wide range of technologies including real-time data acquisition and analytics, dynamic control based on complex process logic, and machine learning models all in the same user-friendly environment. The capabilities to follow the process from the raw material intake, analyzing its characteristics in advance to optimize the melting and final steel composition and continuously evaluating correlations between the process and usage of consumables to find the most favorable operating point are some of the functionalities implemented in this platform. Using the available data from the Aperam Genk process, the AMI system AI algorithms take real-time decisions for the control of the electric power input, and the flow of gas and oxygen given the operation goals and requirements of every heat. After the successful approval of this project, received in November 2022, the next stage in this collaboration is the Slag Module development, optimizing the slag practices using AMI algorithms. Details of the installed system are presented in this paper, as well as the reported results.

Giovanni Bavestrelli , Tenova S.p.A., Italy

Co-Author:
Christian Leoni, Tenova S.p.A.

Abstract:
Metal scrap is a strategic raw material for the steel industry, accounting for a demand of nearly 30% of the metallic charge required for the global crude steel production, with the share that is foreseen to increase in the mid-term future. In scrap recycling steel mills, the scrap is typically loaded into an Electric Arc Furnace (EAF) in a controlled manner, tracking what goes in the charge mix and relating it to the quality of the liquid steel. Accurate tracking of scrap material from the time it enters the plant to the time it exits the furnace as liquid steel requires multiple technologies. The presentation describes the machine learning applications implemented in Ori Martin’s Steel Mill in Brescia, Italy, as part of the Lighthouse Plant “Acciaio_4.0” project in collaboration with Tenova. The project was selected by the Italian Smart Factory Cluster (Cluster Fabbrica Intelligente), on behalf of the Italian Ministry of Economic Development (MISE). The project created a Smart Factory in Ori Martin Steel Plant by integrating the enabling technologies of Industry 4.0 in the steelmaking process. The presentation focuses on the following solutions: • Automatic metal scrap classification • Identification of bulky material on Consteel® to prevent damaging EAF electrodes • Finding correlation between loaded scrap material and tramp elements in liquid steel • Optimization of charge mix with strict requirements on residual elements content • Rating of scrap suppliers based on production results The solutions involve the use of convolutional neural networks for image classification and various machine learning algorithms for process and sensor data.

Hyoungkeun Choi, Hyundai Steel Co. , Korea, Republic of

Co-Author:
Kyunghwan Lim, Hyundai Steel Co.

Abstract:
In the steelmaking process, the input of ferroalloy, especially Mn, causes a lot of heat loss. In this situation, the importance of the ladle furnace is gradually increasing as the production of high-grade steels containing a lot of alloys increases. However, due to the characteristics of LF operation, it is difficult to standardize operation, so automation is being developed later than other secondary refining facilities such as RH. However, with the recent development of machine learning techniques, it is possible to develop models with high accuracy, and it is also easy to connect the calculation results of these models with systems such as PLCs. As a result, demand for automation system development is increasing in LF as well. Our company developed a high-accuracy model for temperature prediction and completed a field test, and also developed a linear programming-based model for calculating the ferroalloy input. Based on the operator's operation performance, the pattern model according to the situation was configured as a rule-based system, and the automation system was developed by combining the previously developed temperature and ferroalloy model. As a result of applying an automated system that minimizes operator intervention to the operation, it was confirmed that more standardized and less variance operations were possible.

Christine Gruber, K1-MET GmbH, Austria

Co-Author:
Maria Thumfart, K1-MET GmbH
Johannes Wachlmayr, K1-MET GmbH
Roman Rössler, voestalpine Stahl GmbH
Birgit Palm, VDEh-Betriebsforschungsinstitut GmbH
Bernd Kleimt, VDEh-Betriebsforschungsinstitut GmbH
Sudhanshu Kuthe, KTH Royal Institute of Technology
Björn Glaser, KTH Royal Institute of Technology
Izaskun Alonso Oña, Sidenor Investigación y Desarrollo
Vito Logar, University of Ljubljana
Dejan Gradisar, Institute Jozef Stefan
Miha Glavan, Institute Jozef Stefan
Mojca Loncnar, SIJ Acroni d.o.o.
Pavel Ettler, Compureg Plzeň S.r.o.
Matjaz Demsar, Siemens Trgovsko in storitveno podjetje, d.o.o.
Zdravko Smolej, SIJ Acroni d.o.o.

Abstract:
The INEVITABLE project applies digital technologies for an optimized and improved performance of different metalmaking processes with focus on steelmaking but also for nonferrous alloy casting. The aim is the development of high-level supervisory and control systems for different production plants and their demonstration in operational environments to enable an optimized operation of the processes, going hand in hand with a reduction of resource consumption and CO2 emissions. The digital transformation and upgrade of the processes include data acquisition, processing and analytics of datastreams, standardization of relevant data interfaces and storage, and application of the functionalities of smart sensor technologies, cognitive control and Industry 4.0 concepts. The INEVITABLE project revolved around various steelmaking processes, ranging from electric arc furnace (EAF) and secondary metallurgy up to the cold rolling mill. This talk gives an overview over the cognitive control solutions developed in the project. For the EAF operations, process models and optimization framework have been developed, based on both theoretical and data-driven approaches using operational data. Their aim is to allow continuous online estimation of the bath temperature and oxygen level, offline process simulation for scenario testing, and optimization of the energy consumption via improved EAF inputs. In secondary metallurgy, ladle furnace and vacuum degassing processes have been considered. Based on vibration sensors and image data, the stirring behavior has been monitored. Together with the evaluation of other process data, model-based advisory systems for process control and decision support as well as predictive models for cleanliness and castability of liquid steel have been developed. Furthermore, a system for supervision, optimisation, and condition monitoring of cold rolling mills has been developed. Strip speed sensors and an x-ray thickness measurement have been introduced, and an overall upgrade of the databases has been implemented, including communication interfaces between sensors, edge devices and cloud database.

Michele Roveda, Tenova S.p.A., Italy

Co-Author:
Stefano Moroni, Tenova S.p.A.

Abstract:
Furnace Digital Packages are Tenova’s Industry 4.0 solution to shift the furnaces from traditional equipment to more efficient and technological production units. They aim to improve: • Process and product quality • Overall efficiency • Plant availability Tenova Furnace Digital Packages comprise both AI / ML dedicated algorithms and dedicated counters, alerts and alarms. The former easily allow field data analysis and new insights to the production team, the latter give an effective help to the maintenance team by keeping monitored the wearable and critical parts. The platform services allows for packages synergy with the Tenova Dynamic Furnace which is the digital twin of the real furnace: a dedicated tool specifically designed to improve the production schedule management and for the optimization heating practices. The tool helps the process and metallurgical engineers in developing new heating strategies and get evaluation about furnace behaviour (e.g. consumption, set-points, etc.) as well as final slab/billet/bloom heating quality. The deployment of ML models and packages started in 2020, first figures from field about real benefits are already available while development of features is still ongoing.

Martin Fein, Andritz AG, Austria

Co-Author:
Andreas Argards, SSAB AB
Lubomir Slapak, Allisnet S.r.o.

Abstract:
SSAB intended to improve the quality and productivity for their existing Continuous Annealing Line (CAL) in Borlänge, Sweden, so they chose ANDRITZ Advanced Furnace Control (AFC). A new generation of AFC was first time applied on SSAB’s CAL. It consists of a modular design with a newly developed cooling model and a brand new multi model predictive control. The precise physical model for all heating and cooling sections acts as virtual sensors, which provides real time trends of the temperature distribution within the furnace and acts as a huge source for Big Data. The multi model predictive control ensures collaboration between all models to enable constant high quality heat treatment by assuring maximum productivity. It runs in fully automatic furnace operation and optimizes line speed, heating power and blower speed for all furnace sections. The predictive capability of the model opens a lot of new opportunities for online and offline improvements of the production. Some examples, that have been developed together with SSAB, are the online compensation of the varying strip emissivity over the coil length and the optimization of the heat cycle recipes. Furthermore, the visualization of the prediction also facilitates operators to have a better situational awareness of the furnace operation. This paper reports on details of the installation of AFC at a CAL of SSAB and the resulting increase in quality and productivity.

Berend Brasjen, Tata Steel IJmuiden B.V., Netherlands

Abstract:
At Tata Steel's European production lines for continuous annealing and hot-dip galvanising, strip temperature is controlled by an in-house developed, non-linear model predictive controller. Every incoming strip is assigned its own set of target temperatures, commonly referred to as the temperature recipe, based on its chemical composition, dimensions and special requirements. Over the years, the controller has demonstrated its value by maximising speed of production whilst minimising deviations from the recipe, especially during transitions between strips with different recipes. This ensures maximum uniformity of the time-temperature path over the length of each strip, which translates to maximum homogeneity of product properties (mechanical, chemical or otherwise) if these are uniform at the start of the annealing process. However, if the initial product properties vary between several strips within the same order, or even along the length of a single strip, these heterogeneities will not be fully alleviated by keeping the process temperatures close to recipe values. In order to improve the uniformity of the final product properties after annealing, we have developed an application capable of determining a tailored set of temperature targets, based on incoming strip properties. The application uses material models to determine which target temperature adjustments are expected for a specific strip, so as to provide optimum homogeneity of final product properties. The resulting tailored recipe is then fed back to the temperature controller to be assigned for this specific strip. In this way, maximum use can be made of the annealing line capabilities to achieve our customers’ target material properties with tighter tolerances.