Roller screens are our specialty, mainly in the area of mineral (iron ore) processing. Thanks to WooDEM and custom-developed pellet contact model, we can capture results from these complex machines quantitatively.

## Screen types

Screens segregate feed into 2 or 3 resulting streams:

**Classifier**-type screens are used immediately after pelletizing (balling) discs and separate the**feed**into 3 streams:**fine**,**product**and**coarse**.**Feeder**-type screens are used to separate**fine**pellets from**product**, often used before green pellets go into furnace for burning. These screens sometimes have more than one deck.

Some screens can have a more complicated geometry, e.g. this one:

## Dynamics

The screen is transporting pellets which undergo complex interactions between themselves and the rollers; rollers and gravity move pellets downward on the sieve.

This video shows differently sized pellets u**N**dresize in red, **O**nsize in green and o**V**ersize in blue moving on a rather saturated screen:

The dynamics leads to some degree of separation in thick layers on the sieve:

The complexity fo the task requires some structure to describe what is happening − the contamination nomenclature.

## Contamination quantification

### Classifier-type screens

Novel terminology is used to described the sieving process in detail. Desired pellet size is defined by diameter range (usually d₀=9mm…d₁=16mm).

- based on pellet
**diameter**, we distinguish**uNdersize**(d<d₀),**Onsize**(d₀≤d≤d₁) and**oVersize**(d₁<d); these fractions are abbreviated as**N**,**O**,**V**. - based on pellet
**location,**we distinguish**Fine**(the stream which ideally takes out the undersize fraction),**Product**(the stream which ideally takes out the onsize fraction) and**Coarse**(the stream which ideally takes out the oversize fraction). These streams are abbreviated as**F, P, C**.

The**feed**stram is abbreviated as**I**(for Input).

As any pellet diameter can theoretically end up in any location, we have 9 combinations and note them with respective letter combination (**NF** = u**N**dersize in **F**ines, **OF** = **O**nsize in **F**ines, …):

Ideal, 100% accurate screen (of classifier-type) will send pellets from the feed based on their diameter u**N**dresize → **F**ine, **O**nsize → **P**roduct, o**V**ersize → **C**oarse, thus the diagonal entries (NF, OP, VC) will be maximized and non-diagonal entries will be zero.

Real screens do suffer from inaccuracies, thus the non-diagonal entries, called **contaminations**, will not be zero.

Contaminations far from the diagonal, namely **VF** (oVersize pellets in Fines) and **NC** (uNdersize pellets in Coarse) are highly unlikely to occur, save in serious misconfiguration (process parameters out of range, machine fault, simulation problem)

Only **OF **(Onsize in Fines), **VP **(oVersize in Product), **NP** (uNdersize in Product) and **OC** (Onsize in Coarse) contaminations are relevant.

**VP** & **NP** **degrade** **quality** of the product (thus we call them degrading contaminations) whereas **OF** & **OC** **diminish quantity **of the product (thus we call them lossy contaminations). These two groups of contaminations are balancing each other − it is easy to have high quality but abysmal quantity (e.g. by hand-picking and measuring suitable pellets); the challenge is to have acceptable quantity and acceptable quality.

**OF** &** VP **(under the diagonal) occur when pellets are screened **earlier** than they should, i.e. in smaller apertures. The cause is pellets being rolled flat by the rollers due to high vertical pressure from the pellet bed or when staying too long in the gap due to weak lateral transport (low inclination).

**NP** &** OC **(above the diagonal) occur when pellets are screened **later** than they should. This is often caused by excessive feed rate (oversaturation of the sieve, percolation), screen being too short, or lateral transport being too fast (high inclination).

Contaminations are specific points on the PSD curves of **I**nput, **F**ine, **P**roduct, **C**oarse:

Material flow can be shown graphically using Sankey diagrams; this type of presentation is only useful for qualitative assessment, not for quantitative comparison, where the differences are too small to be distinguishable:

### Feeder-type screens

Feeder-type screens are simpler in that they only separate **F**ines and **P**roduct (with pellet diameters being u**N**dersize and **O**nsize); they can be seen as a degenerated version of the classifier:

The contamination matrix is only 2×2 here, the relationship between PSD curves and contaminations is shown here:

## Optimizing screen parameters

The screens have many tunable parameters; they have been tuned in the practice over last 100 years, but this involves halting the production and doing physical changes of the machine. Changing a parameter in the simulation is as easy as typing a new number. These include

- number of rollers: total length of screen, number of rollers for different zones (fines, product);
- gap sizes in different zones of the screen;
- inclination of the screen (or even using non-linear cylinder arrangement, such as broken slope);
- roller angular velocity;
- feed rate;
- feed PSD;
- …

Contamination nomenclature introduced above makes it easy to plot trends depending on the parameter being changed, or even multiple parameters. Knowing trends and understanding them is a necessary step for process optimization. Below are some examples.

Simulation results are stored in the standard HDF5 format for easy and automated processing with Python or Matlab or any other scientific package.

This is a plot of contaminations when number of gaps in the fine section changes (on the *x*-axis):

This plot shows the PSD of feed and all sections when altering angular velocity of the rollers:

Feeder-type screen have only 2×2 contamination matrix, stacking the various atop each other (they cumulatively give 100% of the feed rate) is visually appealing and informative at once.

For multi-dimensional variation, visual maps can be built; in this case, number of rollers against **NP** contamination was plotted; each grid point represents one simulation:

## Detail analysis

With WooDEM, we can not only report global flow statistics, but also analyze pellet flow down to single pellet, e.g. time pellets spend in a gap before falling through it and similar.