How Artificial Intelligence and operational data analysis are transforming environmental monitoring and efficiency in steel plants
By Ana Rocha – Energy Efficiency Analyst at Vetta
NOx is a term used to describe a group of gases formed by nitrogen and oxygen. Among them, the most common are nitric oxide (NO), nitrogen dioxide ($NO_2$), and nitrous oxide ($N_2O$). These gases emerge as byproducts of combustion processes when fuel reacts with air under specific conditions, primarily at high temperatures. The formation of NOx is not a simple occurrence; it is a process influenced by various operational variables, such as temperature and the thermal energy sources utilized.
In a steel plant, one of the primary sources of NOx generation is the Electric Arc Furnace (EAF). The environmental impact of NOx is significant, contributing to acid rain, photochemical smog (a type of air pollution resulting from complex chemical reactions between pollutants and sunlight, appearing as a brownish-orange haze over cities), and severe impacts on the greenhouse effect.
Study Scope
This study focuses on the NOx generated in a specific plant consisting of an Electric Arc Furnace (EAF) and a Ladle Furnace (LF), which are connected to a common exhaust gas line. Typically, these two units share the same gas treatment system and a single stack. The total amount of NOx emitted into the atmosphere is the sum of the contributions from both the EAF and the LF.
Development and Methodology
To address the lack of predictability in plants that do not have dedicated analyzers in the stack, Vetta developed a prediction model based on real operational data. The methodology integrated:
Key Variables and Results
The model utilizes day-to-day operational variables to estimate NOx concentration. These include furnace power (EAF and LF), exhaust gas temperature, oxygen and coal injection, the foaming slag index, and electrode position. Validation showed consistent performance: the model achieved an $R^2$ of 0.71 and an adjusted MAPE of 14.3%, indicating that the results are reliable.
But what do these indicators mean?
The $R^2$, or coefficient of determination, shows how much of the variation in NOx emissions the model can explain. This value ranges from 0 to 1. In this case, 0.71 means that 71% of the emission behavior is explained by the input variables, such as furnace power and temperature. In an environment like a steel plant—full of interference and difficult-to-control variables—this level of explanation is considered solid, as it shows the model captured the core logic of the process without relying on generic averages.
The MAPE (Mean Absolute Percentage Error) measures the size of the prediction error in percentage terms. The value of 14.3% indicates that, on average, the model's prediction deviates by this percentage from the actual measured value. The term "adjusted" indicates that the calculation was treated to avoid common distortions when actual values are very low, which is frequent in intermittent emissions. In practice, an average error in this range is low for software-based monitoring without a direct physical sensor in the stack. This allows the data to be used confidently in environmental reports and operational decisions.
Ultimately, the two indicators complement each other: $R^2$ shows the model understands the process behavior, and MAPE shows that its predictions are close to the real values. Together, they prove the tool is a reliable substitute or complement to expensive physical sensors.
Conclusion
This study reinforces Vetta's capability to develop dedicated models that explain the relationship between process variables and pollutant formation. The solution becomes a valuable tool for environmental monitoring, trend analysis, and operational optimization, ensuring greater sustainability and intelligence for our clients.