DEM simulation of charge distribution in a blast furnace hopper

Share this

Published on: November 25, 2019

When Acciaieria Arvedi ran into performance problems with its coking and sintering plant in Servola, Italy, engineer Nicola Petronelli turned to the University of Trieste to help research the issues. I was fortunate to be working on an academic thesis there at the time, under the guidance of professors Lucia Parussini, Carlo Poloni and Marco Boscolo. Our task was to optimize material distribution inside a blast furnace, which, in this case, melts coke and iron ore into either cast iron or ingredients for industrial concrete production.

I was excited about this opportunity to apply discrete element modeling (DEM), which I had not been exposed to in my studies or work experience. Previous engineering courses introduced me to modeFrontier software (the university’s go-to optimization package); it easily integrates with third-party software. For this multidisciplinary study, our team used Onshape to develop the 3-D CAD geometry, Rocky DEM to analyze particles inside the blast furnace, and modeFrontier to both manage the engineering design process and optimize the furnace hopper’s distribution device.

Chemical reactions occur throughout a blast furnace when material is superheated as it falls (Figure 1). Our work centered on the loading process, filling (and emptying) the hopper with alternate layers of combustible coke and iron ore. We knew that performance issues could occur if the layers were not uniform. The hopper incorporates an internal two-pitched distribution device, called the deflector. Real-time photography showed that this component divides material into two different-sized large piles, and two depressions form under the deflector and opposite it — far from uniform.

DEM simultarion for Particle Modeling

It became clear that we needed to develop a better deflector component to improve layering. Ultimately, we used Rocky DEM to perform a charge simulation that predicted behavior of granular material, composed of a large number of particles that, together, assume characteristic behaviors of both fluids and of solids. What differentiates this process from other physics simulation methods (Figure 2) is that no domain discretization grids are used and continuous motion equations are not resolved. Instead, a certain quantity of solid particles is simulated, and, for each one, the equations of motion integrated in time are resolved numerically. Acting forces are determined based on contact between particles, dependent on specific external forces, such as gravity, electrostatic forces, etc. Each particle interacts with others based on parameters like static and dynamic friction, which the user must choose to make the simulated material behave like the real one.

Our study employed spherical-shaped particles, which simulate non-sphericity of the real material using a rolling resistance parameter (one of our objects of calibration). The time-step was in the order of 10-5 seconds, and air resistance was considered negligible.

Material Calibration/Validation

To determine integrated-simulation parameters at the outset, we performed a material calibration by validating the simulated material and comparing it to the real one with respect to the angle of repose (Figure 3). This value represents the angle formed by a pile of material arranged in the most-conical shape possible. Our validation determined the parameter set that minimized the difference between the real-material angle of repose and simulated one.

We ran a modeFrontier–Rocky study to simulate the charge, with the goal to validate the 3-D model and plant parameters.

This simulation set included hopper and conveyor CAD geometry during the charging phase with speed, inclination and dimension variables. Flow rates and durations of iron and coke charges were set: two minutes for iron and one minute for coke. The coke simulation took 5.5 calculation hours to converge ~280,000 particles; iron took ~25 hours for ~1,400,000, both significantly speeded up by GPU (RTX-6000).

Simulation results (Figure 4) agreed with the original real-time photographs: two depressions under the deflector, two piles of material, one pile higher than the other for coke coal, and a greater presence of fine materials in one pile compared to the other — validating the approach.

Optimizing the Deflector

Videos 1 and 2 show that the original deflector divided the flow coming from the conveyor belt into two piles, accumulating in two different sections and throwing larger spheres farther away in the hopper. Because the coke’s jet is thicker than iron ore’s, the ironworks’ piles are the same height.

With a goal of greater charge homogeneity along with coke and iron ore overlap, the ideal deflector needed to divide material flow by four, not two, forming four piles. Our team created two geometries (Figure 5) to study: one that includes flat surfaces and one that is curved.

To optimize the device candidates, we set up a workflow within modeFrontier that included input geometric variables and the Rocky DEM data set. We reduce time to simulate to evaluate an adequate number of designs (only the first 10 seconds of charge were resolved). Two target parameters were chosen, relating to weight distribution within radial sectors that were created virtually inside the hopper.

The integrated modeFrontier–Rocky DEM optimization study showed a clear improvement for both designs: the depression areas were filled below the deflector, and four piles of material formed. In the ironworks simulation using the curved-surface design, Video 3 shows four distinct material flows following collision with the deflector. In particular, there is a marked improvement in two objective parameters. The weight variance between current and concept-device scenarios was cut in half, from 10 percent to 5 percent, respectively. Homogeneity between coke and iron ore (the variance of the difference between the percent weights) improved from 0.5 percent to 0.05 percent, the equivalent of one tenth. Improvements like this can be translated into dollar savings for foundries, a result of reduced furnace wall corrosion and raw material waste as well as improved efficiencies.

Summary

The study yielded some interesting results. We successfully identified a calibration method suitable for any granular material using DEM software, specifically integrating Rocky DEM and modeFrontier. The simulation produced a great deal of insight about how material flows through the hopper. Finally, integrating geometry parameters, Rocky, and modeFrontier can readily be applied to other applications — in our case, identifying two different devices that improve charge distribution in a blast furnace hopper. This led us to perform further optimization of a new geometry, giving the approach more time to explore the solution spaces. And finally, an academic investigation discovered benefits that can be shared with industry.

The team acknowledges Acciaieria Arvedi Trieste for its help throughout this research process.Look for my thesis when it is published, Design of a Device for Optimizing Charge Distribution in the Blast Furnace Hopper with Discrete Element Method.

Gabriele Degrassi

Member of Engineering and Support team in Esteco

Gabriele Degrassi holds a master’s degree in mechanical engineering from the University of Trieste; his thesis involved research on coupling Rocky DEM and modeFrontier. While completing his academic studies, he worked as a quality control inspector at Mangiarotti-Westinghouse. Currently, Degrassi is a structural engineer at ITS Studio.

Leave a comment

Get Fresh Updates on Email

We'll never share your email address, and you can opt out at any time, we promise.