Chapter 18 · Concentration and Separation

Dense medium and magnetic separation

A controlled medium and an apparent cut

Dense medium separation deliberately changes the fluid environment. In a bath at density ρₘ, light and heavy particles can move in different directions. In a cyclone, a particle moves in a rotating suspension and encounters other effects that shift the apparent cut away from the feed medium density. A bath density, a cyclone cut density and the particle density of a named component describe different quantities. The relevant curve maps particle density to the fraction reporting to Sinks at each size. The cut ρ₅₀ occurs at a Sinks partition of one half. The probable error Eₚ is half the difference between the densities giving 75% and 25% partition. A smaller Eₚ indicates a narrower transition on the same density basis.

A washability survey gives the dry mass and assay distribution by density and, ideally, by size. It describes what is available to be split. An equipment partition curve describes how effectively the machine sorts that population. To forecast product quality, apply the partition to that distribution and reconstruct the grade. A sharp partition at the wrong density can be commercially inferior to a broader partition at the proper cut, depending on the slope of the plant’s yield–quality curve.

Dense medium drum and cyclone variants

The calibrated Baguley dense medium drum represents bath conditions through apparent density and viscosity. A drag relation for a sphere gives signed terminal velocity, with positive direction toward Sinks. The DPSIM partition uses an adjusted velocity s·vₜ and a threshold vfull for full Sinks recovery. In the fitted size and velocity units, define a velocity shortfall Δv = max(0, vfull − s·vₜ).

Pₛ = [max(0, 1 − Δv²)]^(A/d² + B)

Here d is the numerical particle size in mm and the entered velocity scale is in m/s. The numerical form is empirical. Its coefficients must accompany the units and duty used in fitting. A neutrally buoyant particle has vₜ = 0, so it can still reach Sinks with a nonzero recovery. Under the same fitted curve, a lighter particle rises and a denser one sinks faster. Bath medium properties specify an environment for classification. They do not represent a medium recovery circuit on their own.

In the calibrated Dunglison dense medium cyclone, the ore cut rises toward smaller particle diameters, and probable error can broaden as particle size decreases. One useful way to express the implemented partition is

ρ₅₀(d) = ρ∞ + K/d
Eₚ(d) = Eₚ,₀ (d₀/d)ⁿ
Pₛ(ρ,d) = 1/{1 + exp[−ln(3)(ρ − ρ₅₀(d))/Eₚ(d)]}

The density and probable error are in t/m³, and the numeric value of K in kg/m² divided by d in mm gives a density increment in t/m³. The factor ln(3) sets 25% and 75% partition at one Eₚ below and above the cut. This property is a practical calibration check. The model estimates feed pressure using an inlet velocity and a fitted inlet discharge coefficient. The resulting pressure is separate from the ore partition coefficients. Geometry and cut should not be treated as universally interchangeable tuning controls.

The cyclone can identify one feed component as medium solids. Its size partition is independent of ore density classification. Let Fₘ be the specified fraction of combined medium solids and water volume that reports to Sinks, and dₘ the segregation reference size. In consistent size units:

Pₛ,ₘ(d) = 1 − (1 − Fₘ) exp[−(d/dₘ)²]
Qwater,sinks = Fₘ(Qwater,feed + Qmedium solids,feed) − Qmedium solids,sinks

This is a volume equation. One t/h of a material at 5 t/m³ occupies 0.2 m³/h. If the desired medium split leaves a negative water flow or a flow above the incoming water, the settings are inconsistent and the unit rejects the split. This is an important check that a model using separate recovery inputs can easily miss.

Worked example. Ore and medium need separate balances

Use a representative size of 0.20 mm. Feed 50 t/h light ore at 2.9 t/m³, 50 t/h heavy ore at 3.5 t/m³, 20 t/h medium solids at 5.0 t/m³ and 80 t/h water. Use ρ₅₀ = 3.2 t/m³ and Eₚ = 0.15 t/m³ at this size. The two ore densities differ from the cut by ±2Eₚ. Their Sinks fractions are 0.10 and 0.90. The ore Sinks therefore contains 5 + 45 = 50 t/h of ore with 90% heavy-ore grade on an ore-only dry basis.

Set Fₘ = 0.40 and dₘ = 0.20 mm. The medium solids recovery is 1 − 0.60/e = 0.779272. Thus 15.58545 t/h of medium solids occupy 3.11709 m³/h in Sinks. Total feed medium volume, defined here as water plus medium solids, is 84 m³/h. The Sinks target is 33.6 m³/h. Water to Sinks is the difference, 30.48291 m³/h. At water density 1 t/m³ this is 30.48291 t/h. The Sinks stream contains 65.58545 t/h total dry solids, so the heavy-ore grade including medium is 68.61%. Report which basis fulfills the performance criterion. A downstream medium-recovery step changes one basis relative to the other without changing the ore mass already recovered here.

For equivalent inlet diameter 0.100 m and discharge coefficient 0.50, the actual slurry volume is 115.52709 m³/h. The effective inlet velocity is slurry flow in m³/s divided by C times inlet area. The associated pressure ρₘu²/2 is about 39.75 kPa when the bulk medium density is 100/84 = 1.19048 t/m³. This pressure calculation is illustrative of the active fitted hydraulic law. A nominal capacity from specified pressure is a separate result from the material split at actual feed.

Magnetic response, grade and bypass

A magnetic separator classifies the response of particles to a field. A fine particle with a magnetically active mineral can still respond differently from a coarse particle with the same mineral. A composite particle needs either magnetic volume grade or a calibrated effective susceptibility. These quantities are not interchangeable. In addition, physical entrainment can send particles to the magnetic product whether or not they have intrinsic attraction to the magnetic field.

The King–Schneider Wet Low-Intensity Magnetic Separator uses a component magnetic volume grade g and a calibrated corrected response G. The sharpness ratio SI is g₂₅/g₇₅. The grade exponent n follows directly from demanding the two corrected response levels at their respective grades. Use a separate fines bypass B(d) decaying with representative size d in mm.

n = ln[ln(0.25)/ln(0.75)]/[−ln(SI)]
G = 1 − exp[−ln(2)(g/g₅₀)ⁿ]
B(d) = B₀ exp(−kbd)
Pmag = B(d) + [1 − B(d)]G

This mixing rule prevents counting the bypassed material twice. At g = g₅₀ the corrected response is 0.5, while actual partition is 0.5 + 0.5B. The water split is directly specified. A decreasing size bypass can strongly affect magnetic product grade when slimes carry little magnetic material. A higher dry yield in this case may mean poorer magnetic grade.

WHIMS (Dobby–Finch) uses magnetic susceptibility instead of grade. The model converts mass susceptibility χₘ in m³/kg to volumetric susceptibility with χᵥ = ρₚχₘ, with particle density in kg/m³. Its magnetic criterion compares B·min(B,Bsat) times a power of χᵥ and d^1.6 with slurry velocity^1.3 times matrix loading^1.1. The magnetic term is a fitted logarithmic response with 50% at its calibrated M₅₀. Negative magnetic capture is set to zero before physical capture is added. The total is bounded by unity. When field or susceptibility is zero, physical capture can remain positive. Magnetic flux density B uses tesla. Substituting a field unit or a susceptibility convention without converting changes the position of the fitted cut.

Worked example. Distinguishing corrected partition from contamination

In a wet low-intensity separation curve, take n = 2, equivalent to SI ≈ 0.45554. At half the cut volume grade, the corrected response is 1 − exp[−ln(2) × 0.25] = 0.15910. For d = 0.100 mm, B₀ = 0.20 and kb = 5/mm, bypass is 0.12131 and total recovery is 0.26111. Out of 50 t/h of such solids, 13.0555 t/h reports to the magnetic outlet. For a separate 10 t/h class with zero magnetic volume grade at the same size, 1.2131 t/h reports to the magnetic outlet solely through bypass. A field-strength change would not be the first correction to investigate for that non-magnetic contribution. Check the bypass calibration, feed fines and actual stream measurements.

Exercises and answers

  1. A cyclone sends 25% and 75% to Sinks at 2.90 and 3.30 t/m³. Find cut density and Eₚ for a symmetric logistic. Answer. The cut is 3.10 t/m³, Eₚ is 0.20 t/m³, and particles at 3.50 t/m³ have 90% partition.
  2. A medium carries 40 m³/h water and 15 m³/h solids. The medium volume recovery target is 20%, while 12 m³/h medium solids have already been assigned to Sinks by the solids segregation law. What is the implied water flow? Answer. Target medium is 11 m³/h, giving −1 m³/h water. The inputs are infeasible.
  3. In the WHIMS logarithmic response, the dimensionless criterion relative to M₅₀ is one and physical capture is 0.08. Find total capture. Answer. Magnetic capture is 0.50, giving total capture 0.58.

Reading

Napier-Munn (1991), dense medium separation models, doi:10.1016/0892-6875(91)90139-M. Dunglison and Napier-Munn (1999), a general dense medium cyclone model, proceedings of the 6th Central Queensland Coal Preparation Symposium. Dobby and Finch (1977), high-gradient capture, doi:10.1016/0032-5910(77)85044-4. The active dry-basis, feed-volume and susceptibility conventions in this chapter follow the stated DPSIM implementations where they differ from a broad literature family.