Cyclone separator calculator: efficiency, cut size and pressure drop

Enter a cyclone's dimensions, the gas and the dust, and get the cut size d50 and the grade efficiency of every particle size by three classic models (Lapple, Leith & Licht, Iozia & Leith), the pressure drop by three correlations, and a check of the particle counts from an Ansys Fluent DPM run. Free, no sign-up.

Lapple · Leith & Licht · Iozia & Leith · Shepherd-Lapple · Casal-Martinez · Dirgo · 26 verifier checks · Updated 20 Nov 2026

Cyclone (mm)

Gas, dust and flow

Defaults: the cyclone used in Skill-Lync's Fluent DPM challenge as students measured it, Fluent's default air, anthracite at 1,550 kg/m³ (check your Fluent database). Standard designs scale with D. Temperature only enters Leith & Licht's vortex exponent.

Lapple d506.77 µm35.3% at 5 µm
Leith & Licht d502.34 µm67.9% at 5 µm
Iozia & Leith d505.72 µm42.6% at 5 µm
0%25%50%75%100%0.5125102050particle diameter (µm, log scale)LappleLeith & LichtIozia & Leith
SizeLappleLeith & LichtIozia & Leith
1 µm2.1%32.8%2.1%
2 µm8.0%46.5%9.0%
3 µm16.4%55.7%19.4%
4 µm25.9%62.6%31.2%
5 µm35.3%67.9%42.6%
10 µm68.6%83.2%77.4%
  • Lapple: Ne = 6.0 turns. Leith & Licht: n = 0.534, G = 543, natural vortex length 466 mm. Iozia & Leith: vt,max = 5.4 m/s, core 60 mm wide and 516 mm long, β = 2.20.
  • The three cut sizes span 2.34 to 6.77 µm (2.9×).

How to calculate cyclone separator efficiency

A cyclone spins the gas. Each particle is flung outward by its own centrifugal force and dragged inward by the gas flowing towards the outlet. Big particles win and reach the wall, where they slide down to the dust outlet; small ones follow the gas out. Every model on this page boils that contest down to a cut size d50, the diameter collected half the time, and a curve for the sizes either side of it.

Lapple (1951) assumes the gas makes Ne turns at the inlet velocity and asks which particle can drift across half the inlet width in that time:

For the challenge cyclone (inlet 100 × 50 mm, cylinder 402 mm, total height 802 mm) Ne = 6.02 turns. At 3 m/s with anthracite in air, d50 = 6.77 µm, so a 5 µm particle is collected 35.3% of the time and a 1 µm particle 2.1%. The curve 1/(1 + (d50/d)²) is an algebraic fit to Lapple's original chart; well below the cut size it reads higher than the chart (13.8% against the chart's 9% at 0.4 × d50 in EPA's example). Textbooks also disagree on Ne: EPA's course manual uses 0.5 to 3 turns for common cyclones, while this geometric formula gives about 6 for Lapple's own design, so treat the cut size as an estimate.

Leith & Licht (1972), the model the challenge calls Koch & Licht, treat the gas as fully mixed by turbulence in each cross-section, with a vortex whose tangential speed rises towards the centre as v·rⁿ = constant. The geometry enters through one factor G built from the inlet size and the gas volume down to the end of the "natural" vortex:

For the challenge cyclone n = 0.534 and G = 543; the vortex would naturally run 466 mm below the vortex finder, inside the 552 mm available. d50 = 2.34 µm.

Iozia & Leith (1989-1990) apply Barth's force balance at the surface of the central vortex core, where the gas spins fastest, using correlations for that speed and the core size fitted to measured cyclones, then a logistic curve whose slope β also depends on the geometry:

Here the core is 60 mm across, wider than the 50 mm dust outlet, so it ends inside the cone and is 516 mm long; vt,max = 5.4 m/s, d50 = 5.72 µm and β = 2.20, a steeper curve than Lapple's slope of 2. The β correlation gives ln β, with d50 in cm; one widely copied paper prints it without the logarithm, which gives a much flatter curve. The model was fitted on cyclones about 0.25 m across, and is reported to predict poorly below 0.1 m.

For a core wider than the dust outlet, we use the core length as Leith's own 1990 report prints it, (H − S) − (H − S)(dc/B − 1)/(D/B − 1). A 2025 thesis uses (H − h) in the numerator instead, which is where the cone actually narrows to the core width; for the challenge cyclone that gives 526 mm instead of 516 mm.

Li & Wang (1989), the fourth model in the challenge, adds turbulent diffusion and particle bounce at the wall. It is left out here: the readable descriptions of it never give a value for the bounce (re-entrainment) coefficient, and the result depends on it.

The three cut sizes span 2.34 to 6.77 µm for the same cyclone. That is the honest uncertainty of a hand calculation; it is why designers test, and why a CFD result that lands anywhere in that band is not "validated" by one of them.

Cyclone pressure drop

Pressure drop is counted in inlet velocity heads: ΔP = ΔH × ½ρv², so it grows with the square of velocity while the cut size only falls with its square root. In Lapple's model, doubling the velocity cuts d50 by about 30% and costs four times the pressure drop and fan power.

For the challenge cyclone ab/De² = 0.50, so Shepherd & Lapple give ΔH = 8.0 heads: 4.9 Pa at 1 m/s, 44.1 Pa at 3 m/s and 122.5 Pa at 5 m/s. Casal & Martinez-Benet give 33.9 Pa and Dirgo 47.6 Pa at 3 m/s. Industrial cyclones run much faster: a Stairmand high-efficiency cyclone sized for 10,000 m³/h has a 15.3 m/s inlet. At 15 m/s the challenge cyclone would lose 1,103 Pa by Shepherd & Lapple.

On 98 cyclones with measured pressure drop, Leith's 1990 report found Dirgo's equation closest (measured ÷ predicted, geometric mean 0.98), with Shepherd & Lapple under-predicting (1.17).

One unit trap: the equation printed with a constant 0.0027, ΔP = 0.0027 Q² / (kc De Bc Hc (Lc/Dc)^⅓ (Zc/Dc)^⅓), is EPA's correlation for cyclones with inlet vanes (kc = 1 or 2) and uses mixed units: Q in ft³/min, lengths in cm and ΔP in cm of water. With metres and m³/s typed in, it gives 9.63e-3 Pa for the challenge cyclone at 3 m/s.

Separation efficiency from a Fluent DPM run

In the Discrete Phase Model the dust outlet is set to trap and the gas outlet to escape. Fluent then reports how many particles it tracked (N), and how many were trapped (T), escaped (E), aborted or still incomplete when the tracking stopped (I). Two efficiencies follow:

They agree only when nothing is incomplete. Incomplete particles are the ones still circling when Fluent hit its step limit, typically fine particles at low velocity. Counting them as escaped (T ÷ N) understates efficiency; leaving them out (T ÷ (T + E)) assumes they behave like the rest. The fix is in the solver, not the formula: raise Max. Number of Steps (and check the step length factor) until incomplete is close to zero, then both numbers agree.

The second limit is sample size. A surface injection releases one particle per inlet face, often only 100-400. With 120 particles and a true efficiency of 50%, the 95% range is 41%-59%, so a difference of a few points between two runs is noise. The checker shows that range for every run.

What 551 student reports show

Skill-Lync's Fluent course has a DPM challenge: simulate a cyclone separator with anthracite particles, vary particle size from 1 to 5 µm at 3 m/s and inlet velocity from 1 to 5 m/s at 5 µm, and report separation efficiency and pressure drop, after describing four empirical models. 551 reports with "cyclone" in the title are public; we read all of them (counts as of 20 Nov 2026, from pattern searches with samples checked by hand):

  • The models are described, not used. 413 reports describe Lapple's model, 393 Iozia & Leith and 415 Koch & Licht, but we found fewer than 10 that put this cyclone's numbers into any of them and compare with their CFD. This calculator does that step.
  • Incomplete particles counted as escaped. 30 reports print Fluent's particle summary as text, for 172 runs. 75 of those runs had incomplete particles and 45 had more than 10%, up to 86%. Where the printed efficiency of such a run can be matched to a formula (39 runs), 32 are trapped ÷ tracked and the rest are ambiguous. In one report, 254 of 435 particles were trapped, 27 escaped and 154 were incomplete: the printed 58.4% is 90.4% of the particles whose fate is known.
  • Small samples. The median run tracked 136 particles; the most common counts were 378, 120, 91, 136.
  • Pressure drop that should follow v². Where a velocity sweep is printed as text, most rise about 25-fold from 1 to 5 m/s, as ½ρv² says: 1.0 to 28.4 Pa, 2.8 to 86.8 Pa, 1.1 to 37.4 Pa. One rose only from 21.0 to 30.4 Pa, a sign that the value read was not the cyclone's loss. For the same cyclone at 5 m/s those sweeps span 28-87 Pa, against 94-132 Pa from the correlations: the student runs disagree with each other by 3.1×, more than the correlations do (1.4×).
  • A mixed-units formula typed in SI. One report types metres and m³/s into EPA's 0.0027 inlet-vane equation and prints 0.001 to 0.03 Pa for velocities that give 5 to 120 Pa; another quotes the same formula without its units.

Where the inputs come from

  • Challenge cyclone: D 204, inlet 100 × 50, gas outlet 100, vortex finder 250, cylinder 402, total 802, dust outlet 50 mm, from the dimension table in one public report and the MATLAB lines in another (which measured the inlet as 104 × 54 mm). It is close to Lapple's general-purpose proportions with a longer vortex finder. Measure your own CAD before relying on it.
  • Standard designs: Stairmand, Lapple, Swift and Peterson & Whitby ratios from the Politecnico di Torino thesis table 2.2, cross-checked against Gimbun et al. and Altmeyer et al.
  • Air: Ansys Fluent's default (1.225 kg/m³, 1.7894 × 10⁻⁵ Pa·s, from the Fluent User's Guide). Anthracite: Fluent's default particle material; 1,550 kg/m³ is the density 17 reports quote (4 quote 1,506). We could not confirm it in a public Ansys document, so check Materials › Fluent Database in your version and type it in.

What this calculator does not cover

  • Overall efficiency. The models give the efficiency of one particle size. For a real dust, weight each size's efficiency by its mass fraction.
  • Dust loading. Heavy loading raises efficiency and lowers pressure drop; these correlations are for dilute gas.
  • Other designs. Axial-inlet, multi-cyclone, scroll-inlet and uniflow cyclones need their own correlations.
  • Re-entrainment and bounce at the wall and the dust outlet, which lower real efficiency for large particles and high velocities.
Learn CFD with Ansys Fluent

The cyclone separator is a Week 7 project in Skill-Lync's Ansys Fluent course: discrete phase modelling, trap and escape boundaries, particle tracking and vortex cores.

How we checked the calculator

The independent Python script scripts/verify/cyclone.py runs 26 checks, and this page recomputes its numbers with the browser code at build time and refuses to build if they differ.

  • Lapple: a Monte Carlo of the model's own picture (particles spread across the inlet, drifting outward at their Stokes speed for Ne turns) collects exactly half of the d50 particles at 1, 3 and 5 m/s.
  • Leith & Licht: the gas volumes integrated numerically from the cyclone's shape match the closed-form volumes, and the closed-form d50 matches bisection on the curve.
  • Iozia & Leith: d50 solved directly from Barth's force balance at the core surface matches the formula; both core-length cases are exercised.
  • Published examples: EPA's Lapple problems (20.0 and 25.8 µm), EPA's Leith & Licht problem 19 (46.7% against the printed 48%, the book's gas viscosity is not given), EPA's inlet-vane pressure example (1.53 cm of water) and Dirgo's ΔH and ΔP for six standard cyclones in the Torino thesis table; all but problem 19 to the printed rounding.
  • Pressure drop and the Wilson interval (checked by inverting the score test numerically), and the particle-summary parser on lines copied from reports.

Frequently asked questions

How do you calculate the efficiency of a cyclone separator?

Work out the cut size d50, the particle diameter collected with 50% efficiency, then read the efficiency of each size off a grade-efficiency curve. Lapple's model gives d50 = √(9μb / (2πNe vi (ρp − ρg))) with Ne = (h + (H − h)/2)/a turns, and η = 1/(1 + (d50/d)²). For the Skill-Lync DPM challenge cyclone (D = 204 mm) at 3 m/s with anthracite (1,550 kg/m³) in air, Lapple gives d50 = 6.77 µm and 35% at 5 µm. Overall efficiency is the grade efficiency weighted by the dust's mass size distribution.

What is the cut size (d50) of a cyclone?

The particle diameter the cyclone collects half of. Smaller particles mostly escape with the gas, larger ones mostly reach the wall. It falls with the square root of inlet velocity in Lapple's model: for the challenge cyclone 11.72 µm at 1 m/s, 6.77 µm at 3 m/s and 5.24 µm at 5 m/s.

Why do the cyclone efficiency models disagree?

They make different assumptions. Lapple is a simple force balance with an empirical curve; Leith & Licht adds turbulent mixing in the vortex and a gas-volume geometry factor; Iozia & Leith uses Barth's balance on the central vortex core with correlations fitted to measured cyclones. For the challenge cyclone at 3 m/s, d50 is 6.77 µm (Lapple), 2.34 µm (Leith & Licht) and 5.72 µm (Iozia & Leith). A spread of 2× or more between models is normal, so treat any one of them as an estimate.

How do you calculate the pressure drop of a cyclone?

Multiply a number of inlet velocity heads ΔH by ½ρv². Shepherd & Lapple's ΔH = 16ab/De² for a tangential inlet gives 44.1 Pa for the challenge cyclone at 3 m/s (ΔH = 8.0); Casal & Martinez-Benet give 33.9 Pa. Pressure drop grows with the square of velocity: 122.5 Pa at 5 m/s.

How do you calculate separation efficiency in Ansys Fluent DPM?

Set the dust outlet to trap and the gas outlet to escape, inject particles of one size, and read the summary Fluent prints: number tracked, escaped, trapped and incomplete. Trapped ÷ (trapped + escaped) is the efficiency of the particles whose fate is known. If many are "incomplete", the tracking ran out of steps: raise Max. Number of Steps until the incomplete count is near zero instead of counting them as escaped.

What are incomplete particles in Fluent DPM?

Particles still inside the domain when the tracking stopped, because they hit the maximum number of steps. They were neither trapped nor escaped. In a cyclone they are usually the ones spinning in the vortex for a long time, so counting them as escaped understates efficiency, and counting them as trapped overstates it.

Is this cyclone calculator free?

Yes. It runs in your browser, needs no sign-up and sends nothing anywhere.

Related on Skill-Lync

Sources

  • US EPA Air Pollution Training Institute, Course 413 Control of Particulate Matter Emissions, Student Manual (3rd ed., 2000), chapter 7 "Mechanical Collectors": Lapple and Leith-Licht equations 7-4 to 7-21, problems 7-1, 7-2, 19 and 20.
  • D. Leith, Cyclone Performance and Optimization, 12th Quarterly Report, US DOE grant DE-FG22-87PC79922 (1990): Dirgo's pressure drop equation, its test on 98 cyclones, and the Iozia & Leith (1989) equations in the appendix.
  • J. Gimbun, T. S. Y. Choong, A. Fakhru'l-Razi and T. G. Chuah, "Prediction of the effect of dimension, particle density, temperature, and inlet velocity on cyclone collection efficiency", Jurnal Teknologi 40(F) (2004) 37-50.
  • R. Venancio Durand, Performance evaluation of high-efficiency cyclones, MSc thesis, Politecnico di Torino (2025): standard geometry table 2.2 and Dirgo results table 2.3.
  • Universidad de Zaragoza, final project TAZ-PFC-2010-039, section 2.8 (Shepherd & Lapple and Casal & Martinez-Benet pressure drop).
  • Original papers, cited as given in the sources above: Shepherd & Lapple, Ind. Eng. Chem. 31 (1939); Lapple (1951); Leith & Licht, AIChE Symp. Ser. 68 (1972); Iozia & Leith, Aerosol Sci. Technol. 10 (1989) and 12 (1990); Casal & Martinez-Benet, Chem. Eng. (1983); Dirgo, thesis (1988).
  • Ansys Fluent 12.0 User's Guide, sections 8.3.2 and 8.4.1 (default air properties).
Free counselling

From one Fluent project to a CFD job.

Particle tracking, swirl and turbulence models, and knowing when a result can be trusted, are what CFD teams test freshers on. A Skill-Lync counsellor can map the course that fits the role you want.

Talk to a CFD course counsellor

Free. We call you back, usually the same day.

By submitting you agree to our Terms and Privacy Policy.

Talk to a career counsellor

Talk to a career counsellor

Free 20-minute call. We'll map the right course to where you want to work.

By submitting you agree to our Terms and Privacy Policy.