Choose a mechanism before choosing a curve
A gravity separator is useful when particle motion depends strongly enough on density to overcome the effects of size, shape and the surrounding suspension. In an unbounded Newtonian liquid, balance gravity corrected for buoyancy against the drag on a sphere. At sufficiently low relative Reynolds number, the signed Stokes terminal velocity is v = g(ρₚ − ρₗ)d²/(18μ). It scales with size squared as well as density contrast. A light coarse particle can therefore move like a finer heavy one. A wide size distribution entering a gravity separator can degrade density selectivity even if the densities of the pure minerals differ substantially.
The Stokes relation is a limiting model. When the relative Reynolds number Reₚ = ρₗ|v|d/μ is appreciable, a drag correlation depending on Reₚ is needed. Crowding can change both the velocity scale and the effective medium density. Particle shape changes the drag and contact geometry. A mineral label does not by itself establish the apparent density of a composite particle. Give a gravity model the size–density population it needs or explicitly use an empirical partition from the same feed and duty.
The useful comparison among model families is the route by which they convert material properties into selectivity.
| Active DPSIM formulation | Selectivity variable and calculation | Useful calibration evidence |
|---|---|---|
| Gravity Partition Separator | Increasing size partition, shared between components at equal size | Surveyed dry partition by size |
| King–Rao equilibrium jig | Density stratification within size intervals and a lower bed cut | Stratified bed samples or product partitions by density and size |
| Tucker shaking table | Sorted, crowded size–density bands cut at an edge position | Band locations and recoveries at varied cutter positions |
| Tucker spiral concentrator | Inner radial response normalized by outer fine and coarse terms | Splitter surveys at the selected take-off |
| Smooth-wall centrifugal concentrator (Kroll–Rabotin) | Migration toward the bowl wall in flowing slurry | Bowl settings, slurry flow and feed and product size–density data |
| Fluidized-bowl concentrator (Chen) | Calibrated velocity ratio from centrifugal settling and fluidization | Measured capture by component across sizes and fluidization settings |
| Weibull force-ratio centrifugal concentrator | Calibrated force ratio and component recovery asymptote | Component recovery versus a consistent pressure, speed and characteristic diameter |
A bed stratifies before it is split
The King–Rao equilibrium jig partitions each size interval independently. It represents the vertical concentration of each density population through a normalized profile. Let z run from zero at the bottom to one at the top. Let ρᶜ be the particle density of component c in t/m³ and wᶜ its fitted basal weight for that size interval. With α in m³/t, the dimensionless concentration within the layer is
The denominator matters. Without it, profiles of different components would change the amount of bed volume assigned to a layer. At higher z the heavier components decay faster relative to lower-density ones. Basal weights are fitted so the integral of each normalized profile equals the inlet fraction by solids volume. The feed basis is qᶜ/ρᶜ divided by the sum of those terms. Scale α by a power of size relative to 1 mm if size-dependent stratification is calibrated. Recovery to the lower cut is the partial profile integral divided by the full profile integral for that component. Density is constant within a class, so the same fraction applies to its inlet dry mass.
DPSIM uses a stratification discretization with height intervals and an iterative adjustment of the weights. An interval that contains the cut contributes in proportion to the fraction below the cut. Convergence and mesh resolution should be reviewed together. For strong stratification, too few intervals can miss a thin transition zone even where a numerical balance closes. In an equilibrium model, changing throughput does not directly add a settling-time constraint. A feed-rate limit needs independent evidence or an appropriately transient model.
Worked example. Grade and lower-bed cut
At a single size, feed 50 t/h of a particle class at 2.50 t/m³ and 80 t/h at 4.00 t/m³. Each contributes 20 m³/h solids. Enter α = 1.5 m³/t with no size scaling and consider a cut at z = 0.40. For equal solids volumes, the heavy class profile is a symmetric logistic with exponent a(z − 0.5), where a = α(4.00 − 2.50) = 2.25.
The heavy fraction over the full bed averages 0.5. Use the antiderivative z − ln[1 + exp(2.25(z − 0.5))]/2.25 to integrate from zero to 0.40. Dividing by 0.5 gives heavy recovery 0.52817. The corresponding light recovery is 0.27183 because 0.5 times the sum of these recoveries equals the extracted bed volume of 0.40. The product contains 42.253 and 13.592 t/h of heavy and light solids, giving 75.66% heavy grade versus 61.54% in the feed. With α reset to zero, recoveries are both 40% and product grade equals feed grade. This is a useful control calculation to separate the action of the cut from the action of stratification.
Gravity bands on decks and in troughs
The Tucker shaking table represents occupied size–density classes as bands ordered along the discharge edge. Rank first establishes which bands precede others. Higher throughput of early bands can then crowd later bands away from the concentrate cutter. This dependence on feed mass by class explains why a change in gangue feed can change grade even if the valuable component's density is unchanged. It also explains why fixed per-component recoveries can fail when applied to a substantially changed table feed.
The DPSIM implementation uses fitted 7.5% and 50% recovery anchors for each band. Let x₇.₅ be the shifted 7.5% anchor, x the normalized cutter location and w the band width from the 7.5% to the 50% anchor. With x ≥ x₇.₅ and x > 0:
Below a positive anchor the response follows 0.075(x/x₇.₅)². For x₇.₅ = 0.12, w = 0.18 and x = 0.21, recovery is 0.30008. If crowding shifts x₇.₅ outward to the cutter at 0.21, recovery falls to 7.5% with the same width. A one-variable cutter trial without feed-flow measurements cannot tell this shift from changed wash-water conditions. The dry solids and water partition need different evidence.
The Tucker spiral formulation divides the trough into an inner region and outer paths. For a given density contrast relative to the liquid, a fitted inner density term receives a radial cutter factor based on the fraction of inner-zone width. Outer fine and coarse terms enter the denominator of recovery. At the inner-zone boundary the numerator can collect all of the inner term while leaving the outer terms in the denominator. Therefore recovering all inner-zone material does not imply recovery of the whole size–density class. This calibrated variant represents one take-off within the inner zone. Choose the appropriate surveyed port and use measured partitions at that port to evaluate the response to feed changes.
Smooth bowls and fluidized bowls
In the Kroll–Rabotin smooth-wall centrifugal concentrator, the slurry carries particles along a smooth rotating wall. The model bounds the competing radial migration and through-flow response. In its dilute variant the partition is proportional to particle radius squared, density contrast, speed squared and a bowl geometry factor before clipping to one. Let rₚ be particle radius, ω angular speed, Rᵢ and Rₒ inlet and outlet bowl radii, H bowl height, μ liquid viscosity, Q slurry volume flow in m³/s, C the calibration factor and ρₗ the liquid density.
The concentrated variant multiplies K by a fitted hindrance factor that falls with feed solids volume fraction. It also replaces liquid density in the driving contrast with an apparent density calculated from material remaining suspended. A fixed-point solution is needed because the density depends on partition and partition depends on density. Excess solids concentration can invalidate the multiplier or convergence. This is a model limit that requires attention to the feed and calibration, not an instruction to increase bowl speed until a desired recovery appears.
For original illustrative inputs C = 0.70, N = 1000 rpm, Rᵢ = 0.040 m, Rₒ = 0.055 m, H = 0.065 m, μ = 0.0012 Pa·s and Q = 0.0015 m³/s, the dilute K is 8.51503 × 10⁵ m/kg. A 20 µm diameter gives rₚ = 10⁻⁵ m. A class 3000 kg/m³ denser than water has partition 25.55%, while a class with contrast 1600 kg/m³ has 13.62%. At unchanged flow and speed, increasing diameter to 40 µm quadruples the unclipped response. The heavy class then reaches its bound. This apparent selectivity depends strongly on size, which is why feed preparation can be a more effective variable to learn about than speed alone.
Chen's fluidized-bowl model forms a ratio λ of nominal Stokes centrifugal settling speed to superficial fluidization velocity. Fit a clipped Weibull-like response a − b exp(−c λᵈ) per component. Fit the class response using tests across the intended fluidization range. Feed water split is an independent mass balance input. A fluidization rate entered only for partition does not create water in the flowsheet. The Weibull force-ratio model uses a different characteristic force ratio. In the implementation, nominal drag is pressure times particle cross-sectional area, centrifugal buoyancy is subtracted from the centrifugal force, and response is an increasing fitted Weibull recovery in drag divided by net centrifugal force. Force ratio and velocity ratio have different bases and may exhibit different trends. Establish the sign and usable range empirically. Neither model's labels make its curve a universal physical efficiency law.
Exercises and answers
- Two particles settle in a valid Stokes regime at the same speed. One is twice as large in diameter. What ratio of density contrasts is required? Answer. The larger particle has one quarter the density contrast of the smaller particle at equal viscosity.
- In the illustrative dilute smooth-wall bowl case, halve the slurry flow while keeping solids sufficiently dilute and everything else fixed. Answer. The nominal diffuse migration scale doubles. At 20 µm the heavy recovery is about 51.09% and the light recovery about 27.25%. A concentrated feed would not necessarily follow this proportional response.
- An equal-density bed has two resolved size intervals but no density contrast within either interval. Under the King–Rao within-size normalization, can changes in stratification strength alone create density selectivity within a class? Answer. No. For equal density the exponential height factor cancels between numerator and denominator, leaving the volume fractions uniform with height.
Reading
Woollacott, Bwalya and Mabokela (2015), validation of King stratification, doi:10.17159/2411-9717/2015/v115n2a2. Tucker and coworkers (1991), shaking table model, doi:10.1016/0892-6875(91)90141-H, and Tucker, Lewis, Hobba and Wells (1985), spiral concentration proceedings. Kroll–Rabotin and coworkers (2013), physical analysis and modeling of a smooth-wall centrifugal concentrator, International Journal of Mineral Processing 121. Chen and coworkers (2020), fluidized-bowl particle-to-fluid velocity ratio, doi:10.1016/j.mineng.2020.106236. The equations above distinguish the calibrated DPSIM variants from a full equipment model.
