Coke Rate Optimisation in Blast Furnace using Particle Filters

Welcome to the next edition of my newsletter. A big thanks to all of you for great support until now.

In my last edition I had discussed on the Kalman filter and the benefits of using this filter in steel industry. However it has its limitation,

-????????? It is less efficient for non linear / high dimensional applications.

In such case we have an alternate PARTICLE FILTER, these filters are one of the best available filters for non linear and high dimensional applications, or in physics we call them non gaussian systems.

?

Now, how do they work; they use a set of particles or samples to represent the probability distribution of the state. Each particle has a weight, which is updated based on the measurement (likelihood).

Based on the literature (available in google), in a steel industry they can be used for

-????????? Temperature and Heat distribution estimation

-????????? Ream time monitoring of steel slab thickness

-????????? Defects tracking in steel sheets

-????????? State estimation in continuous casting

-????????? Emissions monitoring

?

As stated in my inaugural edition of my newsletter, I want to keep it as practical as possible to check and reap the benefits of gen AI/ML models dedicated to steel industries. Unfortunately I don’t have the needed infrastructure/tools to validate the usage on the above mentioned examples. The closest I could have done was on temperature and heat distribution estimation. However with the set up I have access to has the limitation by which I somehow am unable to reap the benefits.

I was having a discussion with one of my friends in R&D employed at a global steel giant and he was telling me the importance on coke rate optimization. This discussion happened during my first key note address at my alma matter. (Another proof things learnt :) )

So, in this edition we are drilling down into how we can optimize the coke rate in a blast furnace.

Objective of the simulation – To check if amount of coke required can be minimized while maintaining productivity.

Set up-:

Sensors- Simulated to track the coke rate (input), weight sensors to monitor the total stock level

Particle Filter Optimization-Particles represent the distribution of possible states for iron ore reduction efficiency based on current coke rates and ore input.

Model- The filter uses an energy model that considers coke as the primary energy source and includes parameters for reaction rates, heat transfer, and furnace efficiency.

The working methodology-

1)????? Particles predict the reduction efficiency based on the amount of coke and other reducing agents added.

2)????? As the furnace operates, the particle filter updates weights based on measured temperatures and gas compositions, which reflect reduction efficiency.

3)????? If the efficiency estimate indicates that the reduction is sufficient, the system can reduce coke input without sacrificing productivity.

?

The Overview of Scenario Simulation-

In a blast furnace, coke is added at a certain rate, typically measured in kilograms per ton of hot metal produced (kg/THM). The goal is to optimize the coke rate to achieve efficient reduction of iron ore, high productivity, and reduced operating costs, while maintaining the quality of the hot metal output.

With the particle filter, I could continuously estimate unmeasurable furnace states like temperature distribution and gas composition, allowing me to dynamically adjust the coke rate. In the absence of the particle filter, the coke rate was typically set conservatively to ensure furnace stability, which may have lead to excessive coke consumption.

?

Comparison-

In order to check the effectiveness both conventional approach and the approach with particle filter had to be compared. I tried both, here are the results,

Conventional Approach:

1)????? Coke rate set point average 500kg/THM

2)????? A fixed coke rate was used regardless of variation in furnace dynamics

3)????? Temperature or reduction efficiency deviations are managed manually, often lagging behind the actual changes in the process.

When simulations done it doesn’t come without challenges, some of the challenges observed were,

1)????? Higher overall coke consumption due to a conservative approach

2)????? Less flexibility in responding to real-time furnace condition changes

3)????? Higher variability in temperature and gas composition, leading to periods of inefficiency and increased emissions

Optimized Approach using Particle Filter:

1)????? Coke rate – started at 500kg/THM

2)????? The particle filter adjusted the coke rate based on real time estimates of furnace temperature, gas composition and reduction efficiency

Advantages observed:

1)????? Real-time adaptation to changes in temperature and gas flow dynamics.

2)????? ?Lower coke consumption by reducing the coke rate when the filter estimates sufficient reduction efficiency.

3)????? Enhanced stability in furnace conditions, resulting in consistent product quality and energy efficiency.

?

Analysis:

To illustrate I simulated coke rate data for both scenarios over a 10-hour shift in a blast furnace, plotting the coke rate as well as reduction efficiency over time.

?

Graph 1: Coke Rate Over Time

• X-axis: Time (hours)
• Y-axis: Coke Rate (kg/THM)

With the particle filter, the coke rate starts high but gradually decreases as the filter’s state estimates indicate stable reduction efficiency. Without the particle filter, the coke rate is held at a fixed level.

Graph 2: Reduction Efficiency Over Time

• X-axis: Time (hours)
• Y-axis: Reduction Efficiency (%)

The particle filter enables more consistent reduction efficiency by dynamically adjusting the coke rate, resulting in fewer fluctuations. Without the filter, the fixed coke rate leads to periodic over- or under-reduction, which affects efficiency.

I'll now simulate and generate these graphs.

These graphs illustrate the benefits of using a particle filter in optimizing the coke rate in a blast furnace:

1. Coke Rate Over Time: Without Particle Filter: The coke rate remains fixed at 500 kg/THM, as adjustments are conservative and static. With Particle Filter: The coke rate fluctuates dynamically, reducing gradually as the filter estimates stable internal conditions, achieving an average reduction in coke usage.
2. Reduction Efficiency Over Time: Without Particle Filter: Reduction efficiency fluctuates more widely, sometimes dropping due to inefficient coke utilization or lagging responses to furnace condition changes. With Particle Filter: Efficiency is more stable and consistently higher, thanks to the real-time adjustments that the particle filter provides.

By using particle filters, the blast furnace achieves a more efficient, adaptive coke usage strategy, maintaining high reduction efficiency while reducing operational costs and emissions.

Happy to discuss and learn.