In a targeted study, ABB developed and successfully tested an advanced artificial intelligence (AI) model that accurately predicts the temperature of the steel melt within an electric arc furnace (EAF) continuously from the rise in the cooling water. Further development of this innovation will allow the steel industry to better optimize their steelmaking process and promote sustainability, saving energy and lowering emissions.

Subscribe to ABB innovation and technology news

_{Subhashish Desgupta, Vishal Jana, Dinesh Patil ABB Corporate Research Process Automation Bangalore, India subhashish.dasgupta@in.abb.com, vishal.jana@in.abb.com, dinesh.patil@in.abb.com}

It is widely known that the production of steel is associated with high levels of greenhouse gas (GHG) emissions: steel mills globally produce 1.78 Gt of off-gases annually [1]. Without a doubt the carbon footprint of steel is high – in 2018, for instance, for every ton of steel produced, an average 1.85 tons of CO₂ was emitted; this equates to about 8 percent of global carbon dioxide emissions for that year [2]. This reality is in sharp contrast with the increasing customer demand for less carbon-intensive steel production, tightening of carbon emission regulations and the increasing investor and public interest in sustainability [2].

Thus, despite notable challenges, steel production needs to become greener; efforts toward decarbonization, optimizing energy consumption, and other actions are necessary. The application of advanced digital technologies could help resolve these challenges. With more than a century of experience, ABB focuses on electrification, automation and digital solutions that incorporate advanced data analytics, machine learning (ML) and other AI technologies to help mitigate the challenges faced by metal producers to reduce emissions, optimize energy consumption while improving productivity and lowering costs. Thus, ABB can enable the digital transformation of the steel industry to accelerate sustainability. In this work, ABB investigates the application of mechanistic and AI models, specifically Neural Network (NN) modeling, to provide the means to accurately estimate the steel melt temperature in an electric arc furnace (EAF) non invasively in an effort to reduce outgassing and energy consumption through optimization.

Choosing the right furnace with sustainability in mind

Pursuing a low carbon future in the steel industry begins with the choice of smelting furnace. Two main steelmaking technologies are in use: the conventional integrated blast furnace (BF)/basic oxygen furnace (BOF), the principal method used currently in Europe, and the increasingly popular EAF →01. While integrated blast furnaces produce steel from raw materials such as iron ore and need coal (or coke) to act as a reductant, EAFs can use a wide range of raw materials such as scrap steel and direct reduced iron (DRI), and hot metal – independent of coal. Not only does the flexibility provided by EAFs help producers to meet market demands more quickly whenever the availability of raw materials changes, EAFs are associated with lower carbon dioxide emissions, better resource efficiency, and recycling capabilities. ABB, with its focus on innovative technologies in the metals industry, is actively assessing and developing breakthrough technologies for EAFs, thereby providing technologies to improve operational efficiency and support the steel making industry’s pursuit of sustainable low-carbon steel production.

Promoting efficiency too

As with all batch processes, control of process variables, such as temperature, during each stage of the steelmaking process is crucial to ensure maximum productivity while reducing energy use and operational costs. The utilization of non-invasive temperature measurement can help reach these goals.

To promote EAF process efficiency and better control the production of off-gas, it is necessary to monitor the steel melt temperature accurately during the process but this is challenging. Given that the melt temperature can be as high as 1,630 °C or more, it is impossible to install a permanent temperature probe within the EAF. To measure melt temperature the existing practice is to introduce a sacrificial sensor into the melt, to record the temperature momentarily and then to discard the sensor. This process is performed once in every production run. The thermal sensor is inserted into the thermowell; the electrodes are powered off during the measurement process, thereby reducing efficiency, impacting costs and even safety.

To fine-tune the various operational parameters with the objective of saving energy and ensuring quality, it would be necessary to monitor temperature in real-time and not simply once per run. ABB set out to explore a method to do just that: estimate the melt temperature continuously using external temperature measurements and AI methodology at the Process Automation Hackathon 2022 – ABB’s annual event designed to showcase innovative ideas in the field of process automation.

EAF construction and smelting process

Prior to describing the methodologies applied in this investigation it is important to explain how the EAF →01 functions. Using the heat generated by arcing electrodes, steel, along with other feed materials, is melted within an enclosed chamber →02. The firebrick chamber is provided with a refractory lining within which a jacket of water pipes is installed for cooling purposes to protect the refractory lining →02. The gases generated during the cooling are ejected from an outlet at the top of the EAF.

Non-invasive temperature estimation

At the PA 2022 Hackathon, ABB and collaborators examined the idea of whether the melt temperature rise of the melt could be determined based on external manifestations, specifically the temperature rise of the cooling water that circulates in pipes within the chamber jacket. ABB performed an initial proof-of-concept study using measurement data supplied by the customer. The cooling water temperature rise, Tcw, determined by subtracting the outlet water temperature from the inlet temperature, was directly obtained from the plant →03a. Plant data from measurements of the arc power, oxygen flowrate, etc. were also supplied. The melt temperature, Tm, was estimated in real-time using a heat balance model performed in python →03b. Scaling up the Tcw with a mathematical non-linear function revealed a good correlation between the scaled Tcw and the Tm, after 1,000 s, approximately 16 minutes, of the process initiation. This positive study result led the team to investigate whether more advanced AI methods could be used to estimate melt temperature non-invasively.

Subsequently, a deep neural network (DNN) model was developed in Python to relate the cooling water temperature to the melt temperature →04a. The DNN model, trained on historical data, could predict the melt temperature from the cooling water temperature rise with an acceptable accuracy →04b. With further modifications to the analysis and improvements in data collection, better accuracy prediction is expected in subsequent investigations. Currently, ABB is working to be able to integrate the non-invasive temperature estimation code in its control platform as an offering for metal processing operations.

Unique advantages

In this study, ABB has shown that it is possible to satisfactorily predict process behavior non-invasively, eg, melt temperature, using AI modeling methods as long as relevant process conditions are known, the availability and quality of data is sufficient and domain experts are involved. In ABB’s case, the technologist will be able to non-invasively estimate internal parameters. This capability is not only gratifying, it also has extensive practical implications: Non-invasive melt temperature estimation will allow real-time temperature monitoring. This in turn provides an invaluable means to optimize the operational process: The melting process continues while the need to shut down the electrode power, even momentarily, and to discard electrodes, is obviated, thereby saving time, energy and costs. Importantly, the ability to monitor temperature in real-time allows for the control of off-gas emissions; this is a significant concern for the steel making industry not only in terms of operational efficiency but is also imperative for the industry’s efforts to reduce GHG emissions to address customer, investor and regulatory demands.

Outlook

While the primary idea of this targeted investigation was to examine whether cooling water temperature could be be used as a proxy to predict melt temperature, ABB proposes, based on the positive results to date, that other external measurements, such as the off-gas temperature, could be used to estimate the melt temperature using advanced DNN methodology. Such investigations are underway. In this way, ABB, with its strong focus on metals and digital technologies, is working to create value for the steelmaking industry through analytics-based innovations.

_{References

[1] J. Kleinikorda et al., “What Shall We Do with Steel Mill Off-Gas: Polygeneration Systems Minimizing Greenhouse Gas Emission”, Environmental Science & Technology, Vol. 56 No. 18, 2022, pp. 13,294 – 13,304.

[2] C. Hoffmann et al., “Decarbonization challenges for steel”, McKenzie & Company, June 3, 2020. Available: https://www.mckinsey.com/industries/metals-and-mining/our-insights/decarbonization-challenge-for-steel [Accessed June 5, 2024].}

Select region & language