AI models to Optimize Hydrogen Consumption in DRI Process (Green Steel)

Welcome to the next edition of my newsletter! The focus this month is on DRI.

?

What is DRI?

DRI is a methodology in iron making process where iron ore is directly reduced to produce iron using reducing gases (typically without melting the ore). It is one of the process employed in Green steel making. The advantages of using DRI are,

-??????? Energy Efficiency

-??????? Flexibility in reducing agents

-??????? Lower carbon emissions

-??????? High Quality Iron

?

Challenges:

There are many challenges in the DRI process. One of those is the hydrogen consumption. Additionally it is very expensive and therefore it is important to optimize the consumption.

In this edition of my newsletter I have to tried to use AI models in DRI process that can help in optimizing the hydrogen consumption.

?

Problems Definition to build the model:

In the DRI process, hydrogen replaces carbon to reduce iron ore (typically iron oxide, Fe?O?) into iron (Fe) in the form of Direct Reduced Iron (DRI). Hydrogen gas reacts with iron ore at high temperatures (700°C–1000°C) to produce iron and water vapour instead of CO?.

The challenge:

• Hydrogen is expensive and its consumption needs to be optimized to make the process economically viable and environmentally friendly.
• Variations in ore quality, temperature fluctuations, and changes in feed material can affect hydrogen efficiency.

Dat inputs for AI model:

Raw Material Characteristics – Ore grade, Size distribution and moisture content (used these 3)

Operation conditions- Temperature, Pressure, Hydrogen flow rate, reactor volume (text book case)

Metric- Optimize hydrogen consumption per ton of iron production

Sensor data- Real time (Hydrogen flow rate, temperature, pressure)- modelled DRI process in the same software that I used for EAF and blast furnace.

Model Components:

ML Model- Regression model was used, training was given on historical data (simulated+experience) based on various operating parameters

Reinforcement learning- RL algorithms was used to dynamically optimize hydrogen flow and process parameters by simulating the effects of different actions, thereby helping to learn the most effective strategies over time.

Performance after the implementation:

-??????? Hydrogen efficiency: 15 to 20% more efficient as the AI system adapts the hydrogen flow rate to match the actual requirements of the process.

-??????? Hydrogen Cost: Significant reduction in hydrogen usage per ton of DRI production.

Comparison:

Before AI Model:

• Hydrogen consumption per ton of DRI: 50 kg of hydrogen per ton of iron.
• Emissions: CO? equivalent emissions from inefficient hydrogen use (if hydrogen is produced using fossil fuels). (This data was based on both on simulations and through my past experience in DRI. Did the weighted average)

After AI Model:

• Hydrogen consumption per ton of DRI: Reduced to 40 kg of hydrogen per ton of iron.
• Emissions: Lower emissions due to better hydrogen efficiency, contributing to the green steel initiative. (Though the 2nd point needs further analysis)

Advantages that can derived out of this analysis:

-??????? Cost Efficiency: The model optimizes hydrogen usage, leading to lower operational costs.

-??????? Sustainability: By using less hydrogen, the plant reduces its carbon footprint and operational waste, contributing to greener steelmaking.

-??????? Real-Time Adaptation: The AI model adapts to varying process conditions in real time, ensuring consistent product quality and optimal hydrogen use.

Conclusion:

Implementing an AI model to optimize hydrogen usage in the DRI process enables steelmakers to significantly reduce hydrogen consumption, lower costs, and improve environmental sustainability. By integrating real-time sensor data the AI model makes the process more dynamic and efficient, driving the transition toward greener steel production.

-??????? Hydrogen Consumption in DRI Process:

The first bar chart shows a reduction in hydrogen consumption from 50 kg per ton of DRI (Before AI) to 40 kg per ton of DRI (After AI), highlighting the efficiency gains from AI optimization.

-??????? Emissions Reduction (CO? Equivalent):

The second chart illustrates the reduction in emissions (CO? equivalent), dropping from 5 kg CO? per ton of DRI (Before AI) to 4 kg CO? per ton of DRI (After AI), due to improved hydrogen efficiency.

Happy to discuss and learn!

?

Wishing everyone a VERY MERRY CHRISTMAS and a VERY HAPPY NEW YEAR!