Dry classification with an air separator
A high-efficiency air classifier separates by aerodynamic response in a rotor and air stream. Air flow appears in more than one model quantity. It affects the relation of the cut size to rotor speed and fine feed and also determines the overall dust loading. At fixed solids rate, increasing air flow can shift the model's cut size and change its fine bypass and sharpness at the same time. A survey should measure rotor tip speed, bulk product flows and loading under the same operating condition.
In the Altun–Benzer form, dust loading L = 1,000 S/Qₐ uses feed solids S in t/h and air Qₐ in m³/h, giving L in kg/m³. Calculate fines flow S₍₃–₃₆₎ = S[F(36 µm)−F(3 µm)]. When direct sampling is available it can supply that interval mass. For rotor tip speed v in m/s and chamber diameter D in m, the correlations used here are
Use the fitted unit scales for the dimensions entering the empirical powers. The factor of 1,000 converts the cut output from mm to µm, and division by 3,600 converts air to m³/s within Rₐ. The corrected separation uses the Whiten–Beta reduced overflow form with a scale β to locate the corrected 50% crossing. The actual recovery to the coarse product is p = 1−(1−b)G(βx/d₅₀꜀), bounded to [0,1]. A water split in an otherwise dry classifier model is separate from the dust-loading bypass term. Sparse fine sizing around 3 µm can have a direct influence on the estimated cut.
Worked air separator calculation
Take S = 120 t/h, Qₐ = 120,000 m³/h, S₍₃–₃₆₎ = 100 t/h, v = 20 m/s and D = 3.55 m. Use the reference coefficients k_d = 1/2.47, p = 0.7775, k_α = 0.905, n = 1.2679, k_b = 10.467, r = 1.4171, k_β = 0.4417, s = 1.4171 and c_β = −0.1293. Then L = 1.00 kg/m³, Rₐ = 33.333/(20 × 100) = 0.016667, d₅₀꜀ ≈ 16.78 µm, α ≈ 4.511, b ≈ 0.10467 and β = 0.3124. Solving the descending 50% reduced overflow crossing gives β* ≈ 1.1163. This preserves 0.50 corrected coarse recovery at d₅₀꜀. The actual coarse recovery there is 0.10467 + 0.89533 × 0.50 ≈ 0.5523.
For an illustrative population of 40 t/h in a class represented by 5 µm, 60 t/h at 15 µm, and 20 t/h at 40 µm, evaluate actual coarse partitions about 0.04846, 0.40994 and 0.99910. Coarse solids are 1.938 + 24.597 + 19.982 ≈ 46.517 t/h, leaving 73.483 t/h fines. The fine-class partition below the nominal bypass comes from the fitted hook. Compare that fine-end behavior against measured recoveries. A large disagreement may indicate undersampling in the feed fines interval or a fit extrapolation. When reusing these coefficients, evaluate the dimensioned air-to-fines index, dust loading and any raw partition bounds.
Gravity classification with settling velocity
For an ideal suspension classifier, a particle near the cut settles against the upward liquid/slurry motion. The Plitt–Flintoff gravity model uses an effective pool area A and observed overflow slurry flow Qₒ. Its target velocity is vₒ = Qₒ/A, which must be converted to m/s for the settling equation. With particle diameter d in m, solids and suspension densities ρₛ and ρₚ in kg/m³, viscosity μ in Pa·s and shape parameters a and b, the model solves
At small size the viscous term dominates and speed is proportional to d². At large size the inertial term dominates and speed grows as √d. The change in sensitivity means a single Stokes-law extrapolation over the entire size range can misstate the change in cut for a fixed change in pool flow. Effective slurry density must remain below feed solids density for this settling model.
With Qₒ = 30 m³/h and A = 6 m², target v is 5 m/h or 0.00138889 m/s. For ρₛ = 2,700 kg/m³, ρₚ = 1,100 kg/m³, μ = 0.00120 Pa·s and a = b = 1.10, the equation gives d₅₀꜀ ≈ 27.12 µm. Substituting this diameter in metres into the right-hand side reproduces the target velocity. A Reid–Plitt partition with selected sharpness and liquid bypass then predicts separated size-class products. With sharpness m = 2, water bypass bᵥ = 0.25 and d = 2d₅₀꜀, E(d) = 1−exp(−4 ln 2) = 0.9375. The actual underflow partition is 0.25 + 0.75 × 0.9375 = 0.953125.
Observed overflow volume in this gravity model fixes the settling target. Calculated overflow volume separately follows from predicted solids in overflow and the specified water split. Check the two volumes together. A difference can show that effective pool area, suspension properties, water recovery or curve calibration needs attention. For a feed containing components with evidence of strongly different settling speeds, a single bulk density limits the possible agreement by component.
Inverse design from a product P₈₀
A target P₈₀ can specify a process design objective without specifying the required separator cut size. Start with a conservative family of curves, choose selected outlet, curve sharpness α and lower plateau b. For selected fines, use
Selection of coarse instead replaces the exponential by its complement. For each trial midpoint d, apply p_selected to feed cells, sum selected product mass, normalize its passing curve, then interpolate P₈₀ on the sieve grid. Choose a midpoint minimizing the relative error against the target, then report both attained size and selected-product yield. The product selection defines which asymptote is governed by b. A nonzero plateau for selected fines can allow some coarse material to report there even as d becomes very small. P₈₀(d) can then be nonmonotonic. At widely different trial d values, predicted P₈₀ can be nearly the same with very different mass recovered. Bracket or sample across midpoint scales before local refinement and check feasible yields.
Worked inverse class balance
Use sieve boundaries 800, 400, 200, 100 and 0 µm and feed masses 10, 30, 40 and 20 t/h, respectively, in the four intervals. Let the representatives be 565.69, 282.84, 141.42 and 0 µm. Choose selected fines, α = 2, b = 0.05 and target P₈₀ ≈ 243.15 µm. The table gives selected product masses at d = 300 µm.
| Interval, µm | Selection to fines | Fines, t/h |
|---|---|---|
| 800–400 | 0.130797 | 1.30797 |
| 400–200 | 0.563028 | 16.89085 |
| 200–100 | 0.864382 | 34.57527 |
| 100–0 | 1.000000 | 20.00000 |
| Total | 72.77409 |
Passing 400 µm is 0.982027 and passing 200 µm is 0.749927. Linear interpolation in sieve size gives about 243.15 µm at 0.80 passing. At this candidate, coarse solids are 27.22591 t/h, so the product mass closes at 100 t/h. An inverse run targeting the same percentile can find another midpoint with similar error when a fines plateau is present. Compare the resulting yield and per-class partitions before choosing an operating interpretation. In a component-wise inversion, each component can have a target P₈₀. Aggregate percentile must still be recalculated from the summed mass distribution.
A classifier in a recycle loop
If a whole-circuit approximation models classifier coarse return as a constant fraction r of combined feed solids F+R, recycle satisfies R = r(F+R) and R/F = r/(1−r). For F = 80 t/h and r = 0.60, recycle is 120 t/h and the classifier sees 200 t/h. The fine product is 80 t/h. Raising r to 0.66 at fixed fresh feed raises recycle to 155.29 t/h. In a full size model, update the mill product distribution, return different sizes according to their partition and re-evaluate pressure, wet-screen loading or air loading from the current circulating feed. The scalar solution illustrates why a cut-size fit can be inadequate at circuit scale even if it matches the single-pass product percentile.
Exercises and solutions
1. Dust-flow scaling
Hold rotor speed and 3–36 µm feed flow fixed. If air flow rises by 20%, estimate the cut size multiplier for p = 0.7775 and the corresponding dust loading multiplier for fixed S.
Solution. The cut factor is 1.20⁰·⁷⁷⁷⁵ ≈ 1.152. Dust loading scales with the inverse of air flow and becomes 0.8333 times its old value. Sharpness and bypass must be recomputed using that new loading.
2. Pool area sensitivity
In the viscous limit, what is the cut diameter ratio if effective pool area doubles at fixed overflow and slurry properties?
Solution. The target velocity halves. Since velocity is proportional to d², the diameter ratio is √(1/2) ≈ 0.707. If the inertial term dominates the actual response is different, approaching a diameter ratio of 1/4 in the inertial limit.
3. Identifiability of circuit cut
Two inverse partitions hit the same fine product P₈₀, with fines yields 70% and 50%. Which matters for a closed circuit?
Solution. Both size and yield matter. In the scalar closed circuit approximation the coarse recycle fractions are 0.30 and 0.50, giving R/F = 0.4286 and 1.000 respectively. The recycle loads differ materially even before accounting for size-dependent breakage.
Reading
- Altun, O. and Benzer, H. (2014). Selection and mathematical modelling of high efficiency air classifiers. University publication record.
- Plitt and Flintoff (1985), settling velocity correlation as reproduced in Principles of Mineral Processing, equation 4.32.
