foratvise
Get Started
Milling, extraction and separation

Separation (centrifugation, filtration, membranes)

Centrifugal separation (disc-stack skimming of whole milk into skim milk and cream)

Validated modelOEE bound 0.913 [0.847–0.945]20,000 Monte Carlo samples · seed 20261115 · computed November 16, 2026

In short

Separation (centrifugation, filtration, membranes) can use at most 91 of every 100 planned hours. The rest goes to limits no plant can avoid.

  1. 91usable at most
  2. 9are a physical limit

The three limits that set it

Running time: up to 98 %

The rest are stops imposed by physics or by regulation: cleaning, changeovers, defrosting.

Speed: up to 93 %

How close to its maximum speed it can run without damaging the product or the equipment.

Good product: up to 100 %

Some product falls outside specification even with the process well under control.

See the technical detail

Technical detail

Uncertainty intervals, physical models per station and sources, for anyone who wants to check each figure.

1Scenario

Warm whole milk (55–65 °C) is fed to a self-cleaning disc-stack separator that splits it continuously into skim milk and 40 % cream. Fat globules rise through the separation channels under the centrifugal field (Stokes settling); globules too small to reach the disc above before they leave the channel stay in the skim milk. The skim milk must meet a residual-fat target; sediment (cells, dirt) collects at the bowl periphery and is ejected at intervals. The same ceiling structure (Stokes settling, equivalent clarification area, daily cleaning) is applied as an approximation to the clarification of fruit juice and beer and, for the OEE structure only, to the sifting of wheat flour (see the derivation, section 7).

Provenance of inputsComputed1Declared assumption7Verified quote22

2Results

Point value from the nominal parameters and 95% interval from 20,000 Monte Carlo draws of the uncertain parameters.
  • Log standard deviation of the native fat-globule size distribution0.453 [0.335–0.539]
  • Serum viscosity at the separation temperature0.920 [0.858–0.990] mPa·s
  • Critical globule diameter that just meets the skim-fat target1.55 [1.26–2.06] um
  • Critical globule diameter of the capable process (mean skim fat z sigma below the target)1.50 [1.19–2.00] um
  • Share of the feed fat recovered in the cream at the skim-fat target0.9887 [0.9831–0.9913]
  • Cream flow as a share of the feed0.0989 [0.0869–0.1107]
  • TheoreticalMaximum feed per square metre of equivalent clarification area at the skim-fat target (Stokes settling, sigma concept)0.246 [0.160–0.438] L/h/m²
  • Feed per square metre of equivalent clarification area of the capable process0.229 [0.144–0.411] L/h/m²
  • Equivalent clarification area needed for the nominal flow with the capable process109,289 [60,858–173,822] m²
  • Centrifugal acceleration at the outlet radius relative to gravity6,522
  • TheoreticalThermodynamic minimum: osmotic work of concentrating the fat globules into the cream5.55 × 10⁻¹¹ [3.20 × 10⁻¹¹–9.54 × 10⁻¹¹] kJ/kg
  • Angular-momentum work given to the feed at the outlet radius, without paring-disc recovery12.8 kJ/kg
  • TheoreticalMaximum yield (only the sediment leaves the product streams)0.99990 [0.99981–0.99995]
  • TypicalYield with a full sediment-space discharge at each ejection0.9995 [0.9980–0.9998]
  • Share of skim milk above the fat target (re-separated)0.00135 [0.00025–0.00579]
  • TheoreticalMaximum availability (daily cleaning as the only stop)0.9844 [0.9791–0.9886]
  • TheoreticalMaximum performance factor (capable flow / flow at the skim-fat target)0.929 [0.861–0.962]
  • TheoreticalMaximum OEE quality factor (skim fat within the target)0.9987 [0.9942–0.9997]
  • TheoreticalPhysics-bounded OEE0.913 [0.847–0.945]

3Ceilings per dimension

Each dimension shows the ceilings the model computes and, where units are comparable, documented benchmarks. Colour and marker shape identify the ceiling type throughout the Oracle.
  • Theoretical
  • Best available technique
  • Best documented
  • Typical
ThroughputL/h/m² · higher is better
  • TheoreticalMaximum feed per square metre of equivalent clarification area at the skim-fat target (Stokes settling, sigma concept)0.246 [0.160–0.438] L/h/m²
Energy per unit of productkJ/kg · lower is better
  • TheoreticalDOE bandwidth study, thermodynamic minimum energy intensity for fluid milk · Benchmark0 kJ/kg
  • TheoreticalThermodynamic minimum: osmotic work of concentrating the fat globules into the cream5.55 × 10⁻¹¹ [3.20 × 10⁻¹¹–9.54 × 10⁻¹¹] kJ/kg
  • Best available techniqueDOE bandwidth study, dairy separation step, state-of-the-art on-site energy intensity (same boundary) · Benchmark32.6 kJ/kg
  • TypicalDOE bandwidth study, dairy separation step, current typical on-site energy intensity (fuel and electricity at the plant; a wider boundary than the separator drive) · Benchmark41.9 kJ/kg
Yieldhigher is better
  • TheoreticalMaximum yield (only the sediment leaves the product streams)0.99990 [0.99981–0.99995]
  • TypicalYield with a full sediment-space discharge at each ejection0.9995 [0.9980–0.9998]
Availabilityhigher is better
  • TheoreticalMaximum availability (daily cleaning as the only stop)0.9844 [0.9791–0.9886]
Performancehigher is better
  • TheoreticalMaximum performance factor (capable flow / flow at the skim-fat target)0.929 [0.861–0.962]
Qualityhigher is better
  • TheoreticalMaximum OEE quality factor (skim fat within the target)0.9987 [0.9942–0.9997]
OEEhigher is better
  • TheoreticalPhysics-bounded OEE0.913 [0.847–0.945]

4Loss cascade

What separates each ceiling from the next one, from the physical limit to typical operation.
  1. TheoreticalBest available techniqueEnergy per unit of product

    The reversible work of concentrating the fat globules into the cream is the osmotic work of a dilute colloid, of the order of 10⁻¹¹ kJ/kg: practically zero, as the DOE thermodynamic minimum states. A real separator spends its energy accelerating the whole feed to the bowl speed (about 13 kJ/kg at 0.2 m and 5 400 rpm before any recovery), on the friction of bowl, bearings and air, and on the paring discs that convert part of the rotational energy back into outlet pressure. The DOE state-of-the-art step intensity adds the plant boundary (feed pumps, standardisation, heat losses).

  2. Best available techniqueTypicalEnergy per unit of product

    Between the DOE state-of-the-art and current typical intensities (14 and 18 Btu/lb) lie drive efficiency, bowl design (hermetic inlets accelerate the milk gently, open inlets cause turbulence and losses) and part-load operation.

  3. TheoreticalTypicalYield
    0.99990 [0.99981–0.99995] → 0.9995 [0.9980–0.9998] (gap 4.36 × 10⁻⁴)

    The physical floor of the mass loss is the sediment itself (about 0.1 kg per m³). Each ejection also discharges the milk that fills the sediment space, a few to thirty litres every 30–60 min; this is a property of the bowl and of the ejection programme, not of the physics, and partial-discharge systems reduce it.

  4. TheoreticalTypicalThroughput

    The capacity per unit of equivalent clarification area follows from Stokes' law at the critical globule diameter that just meets the skim-fat target. Running at that capacity leaves no margin for disturbances in feed fat, temperature or flow; keeping the mean skim fat z standard deviations below the target forces a smaller critical diameter and a lower flow, which is the performance factor of the OEE.

5Physics-bounded OEE

The highest OEE the operation can reach: availability, performance and quality limited only by unavoidable stops and physical capability.
Fraction of loading time that remains after each unavoidable loss. Axis starts at 0.80.
  • Loading time (ideal)
    1.000
  • Availability
    factor 0.9844 [0.9791–0.9886]
    0.984 (−0.016)
  • Performance
    factor 0.929 [0.861–0.962]
    0.914 (−0.070)
  • Quality
    factor 0.9987 [0.9942–0.9997]
    0.913 (−0.001)
  • Physics-bounded OEE
    0.913 [0.847–0.945]

6Full derivation

Mass and energy balances, rate-limiting step, process capability and unavoidable stops, with every term explained.

Scope. Warm whole milk (55–65 °C) is split continuously in a self-cleaning disc-stack separator into skim milk and 40 % cream. The skim milk must meet a residual-fat target; sediment collects at the bowl periphery and is ejected at intervals. Section 7 states where the same ceiling is applied, as an approximation, to juice and beer clarification and to flour sifting. Symbols and their values, units, uncertainty distributions and sources are listed in the parameter table generated from config/physics/separation.yaml.

1. Mass balance — maximum yield

A separator neither creates nor destroys product; it splits one stream into two. With feed fat xFx_F, cream fat xCx_C and skim fat xSx_S, the total and fat balances per unit of feed are

1=c+(1−c),xF=c xC+(1−c) xS,1 = c + (1 - c), \qquad x_F = c\,x_C + (1-c)\,x_S ,

where cc is the cream share. If a fraction ff of the feed fat escapes into the skim milk (Section 3), the cream carries xF(1−f)x_F(1-f) and

c=xF (1−f)xC,xS=xF f1−c.c = \frac{x_F\,(1-f)}{x_C}, \qquad x_S = \frac{x_F\,f}{1-c}.
  • xFx_F — fat of the whole milk (declared 3.5–4.5 %).
  • xCx_C — fat of the cream (40 %, the handbook's example and a plant set point).
  • ff — residual-fat fraction, set by Stokes settling (Section 3).

Skim milk and cream are both products, so the split itself is not a loss. The only mass that physically must leave the product streams is the sediment (somatic cells, bacteria, dirt), about csedc_{sed} = 1 kg per 10 000 L of milk:

Ymax=1−csedρm.Y_{max} = 1 - \frac{c_{sed}}{\rho_m} .

A self-cleaning bowl ejects the content of its sediment space (VsedV_{sed}, 3.5–30 L depending on size) every tejt_{ej} = 30–60 min. Discharging the whole sediment space each time also throws away the milk in it; at the nominal flow V˙\dot V this is the typical yield

Ytyp=1−max⁡ ⁣(VsedV˙ tej, csedρm).Y_{typ} = 1 - \max\!\left(\frac{V_{sed}}{\dot V\,t_{ej}},\ \frac{c_{sed}}{\rho_m}\right).

The BAT reference document attributes most of the milk lost during dairy processing to separator discharges; that statement is listed as a benchmark but is not comparable with YmaxY_{max} (it is a loss share with an ambiguous basis).

2. Energy balance — minimum energy per unit of product

Thermodynamic minimum. Separation produces no chemical change. The only free-energy change is the demixing of the fat globules: NN globules per kilogram of milk are confined from the volume of the feed into the volume of the cream. Globules are colloidal particles, so the reversible (osmotic) work of this compression is that of an ideal dilute solute,

wmin=N kB T ln⁡1c,N=xFρf 6π dv50−3 e4.5 σ2,w_{min} = N\,k_B\,T\,\ln\frac{1}{c}, \qquad N = \frac{x_F}{\rho_f}\,\frac{6}{\pi}\,d_{v50}^{-3}\,e^{4.5\,\sigma^2},

where kBk_B is the Boltzmann constant, TT the separation temperature, ρf\rho_f the fat density and NN the number of globules of a log-normal (by volume) population with median dv50d_{v50} and log standard deviation σ\sigma (Section 3). Because globules are a million times heavier than molecules, NN is small (about 5×10125\times10^{12} per kg) and wminw_{min} is of the order of 10−1110^{-11} kJ/kg: zero for every practical purpose, in agreement with the zero thermodynamic minimum the DOE bandwidth study assigns to fluid milk.

Where real separators spend energy. The feed must be given the angular momentum of the outlet radius ror_o. Without recovery the work per kilogram is

wrot=(ω ro)2,w_{rot} = (\omega\, r_o)^2 ,

about 12.8 kJ/kg for the handbook's example (0.2 m, 5 400 rpm). Paring discs convert part of this rotational energy back into outlet pressure, and bowl friction, bearings and air drag add losses. This value has no ceiling label: it is a property of the reference geometry, not a documented technique. The documented levels are the DOE intensities of the dairy separation step (state of the art 14 Btu/lb, current typical 18 Btu/lb), which include the plant boundary.

3. Rate-limiting step — Stokes settling and the sigma concept

Flotation velocity. A fat globule of diameter dd, lighter than the serum by Δρ\Delta\rho, rises in a gravity field at the Stokes velocity

vg=Δρ g d218 η,η=ηref μw(Tsep)μw(Tref),v_g = \frac{\Delta\rho\, g\, d^2}{18\,\eta}, \qquad \eta = \eta_{ref}\,\frac{\mu_w(T_{sep})}{\mu_w(T_{ref})} ,

where η\eta is the serum viscosity at the separation temperature. The handbook gives Δρ\Delta\rho = 48 kg/m³ and ηref\eta_{ref} = 1.42 mPa·s at about 35 °C; the viscosity is moved to 55–65 °C with the ratio of water viscosities μw\mu_w (IAPWS through CoolProp), i.e. at constant relative viscosity, which gives about 0.92 mPa·s at 60 °C. In the centrifugal field the acceleration is ω2r\omega^2 r instead of gg; at 0.2 m and 5 400 rpm it is about 6 500 times gravity, the figure the handbook states.

Sigma concept. For a disc stack the centrifugal geometry (number of discs, radii, cone angle, speed) collapses into one equivalent clarification area Σ\Sigma: a feed QQ is clarified of every particle whose gravity velocity exceeds vc=Q/Σv_c = Q/\Sigma (Ambler, 1959). The critical diameter is therefore

dc=18 η QΔρ g Σ,d_c = \sqrt{\frac{18\,\eta\,Q}{\Delta\rho\, g\,\Sigma}} ,

and, in laminar channel flow, a smaller globule of diameter dd is captured with probability (d/dc)2(d/d_c)^2.

Residual fat of a globule population. Native milk fat globules are log-normal by volume with median dv50d_{v50} = 3.41 µm (laser diffraction of raw milk). The width follows from the volume-weighted and surface-weighted mean diameters of native globules: d43d_{43} = 4.4 µm and a specific surface of 1.8 m²/g fat, i.e. d32=6/(ρfSfat)d_{32} = 6/(\rho_f S_{fat}) = 3.58 µm, and for a log-normal distribution d43/d32=eσ2d_{43}/d_{32} = e^{\sigma^2}, so σ≈0.45\sigma \approx 0.45. Integrating the escape probability 1−(d/dc)21 - (d/d_c)^2 over the volume distribution gives the closed form

f(dc)=Φ(u)−(dv50dc)2e2σ2 Φ(u−2σ),u=ln⁡(dc/dv50)σ,f(d_c) = \Phi(u) - \left(\frac{d_{v50}}{d_c}\right)^2 e^{2\sigma^2}\,\Phi(u - 2\sigma), \qquad u = \frac{\ln(d_c/d_{v50})}{\sigma},

with Φ\Phi the standard normal distribution function. Combined with Section 1 it yields the skim fat xS(dc)x_S(d_c), which rises monotonically with dcd_c.

Maximum throughput. Solving xS(dc)=xS,tx_S(d_c) = x_{S,t} for the declared target (0.05 %, inside the documented skimming ability of 0.04–0.07 %) gives the critical diameter, about 1.55 µm, consistent with the handbook's remark that globules below about 1 µm leave with the skim milk. The physics-bounded capacity per unit of equivalent area is

QmaxΣ=Δρ g dc218 η,\frac{Q_{max}}{\Sigma} = \frac{\Delta\rho\, g\, d_c^2}{18\,\eta} ,

about 0.25 L/h per m²: a 25 m³/h line needs an equivalent area of the order of 10510^5 m². The model reports capacities per square metre of Σ\Sigma because Σ\Sigma is a property of the machine, not of physics. The minimum residence time is not limiting; the limit is the settling distance.

4. Process capability — quality ceiling and the performance it costs

The skim fat fluctuates with feed fat, temperature and the solids ejections; at constant flow it is modelled as normal with standard deviation σS\sigma_S. To keep the share above the target at Φ(−z)\Phi(-z), the mean must sit z σSz\,\sigma_S below it:

xS,cap=xS,t−z σS,dc,cap=xS−1(xS,cap),Qmax=1−Φ(−z).x_{S,cap} = x_{S,t} - z\,\sigma_S, \qquad d_{c,cap} = x_S^{-1}(x_{S,cap}), \qquad Q_{max} = 1 - \Phi(-z).

Skim milk above the target is re-separated (a quality loss). The smaller critical diameter of the capable process needs a lower flow through the same machine; with Q∝dc2Q \propto d_c^2 the performance factor is

Pmax=(dc,capdc)2.P_{max} = \left(\frac{d_{c,cap}}{d_c}\right)^2 .

This is the only place where the capability margin enters the rate: a separator run at the flow that just meets the target on average would meet it only half of the time.

5. Unavoidable stops — maximum availability

Hygiene regulation requires milk equipment to be cleaned at least once each day used. The separator is cleaned in place with the pasteurizer circuit; the duration tCIPt_{CIP} is taken through a parameter link from the cleaning model (output cip_time_min, with its Monte Carlo interval). That model is valid for recirculated pipe and plate circuits; the bowl and disc stack are outside its validity, so the linked duration is a lower bound. Solids ejections last tenths of a second and do not interrupt the feed. With the daily cycle tcyct_{cyc} = 24 h,

Amax=1−tCIPtcyc.A_{max} = 1 - \frac{t_{CIP}}{t_{cyc}} .

6. Physics-bounded OEE

Scope of the bound. The theoretical label of OEEmaxOEE_{max} means physics-bounded under the declared scenario: the ceiling follows from the physics of the operation for the declared product, equipment and schedule, and it is a bound only conditional on those declarations. It follows the common definition of all models: AmaxA_{max} counts only stops required by regulation or forced by a physical mechanism, at their physical minimum duration, plus the start-up they force; PmaxP_{max} is the ratio of the physics-limited ideal cycle to the cycle of the capable process; QmaxQ_{max} is the share of conforming units, and mass losses of conforming product are reported only as yield.

OEEmax=Amax×Pmax×Qmax.OEE_{max} = A_{max} \times P_{max} \times Q_{max} .

The sediment and the milk discharged with it are yield losses, not OEE losses. The result, with the Monte Carlo interval over the declared uncertainty of every input, is reported in the output table; the margin of Section 4 (through σS\sigma_S and zz) dominates it.

7. Assumptions, validity and application to other processes

See the assumption and validity lists generated from the configuration file. The density difference is held at its 35 °C reference value; liquid fat expands more than the serum, so at 55–65 °C the true difference is somewhat larger and the capacity ceiling is conservative by the same factor.

The codebook assigns the operation separation to four processes. The model's numbers belong to milk skimming; for the others the same structure is used as an approximation:

  • Fruit-juice clarification and beer clarification (disc-stack clarifiers): the solids (pulp cells, yeast, protein–polyphenol haze) settle outward instead of rising, but the law is the same — Stokes velocity, critical diameter Q/ΣQ/\Sigma, capture probability (d/dc)2(d/d_c)^2, intermittent solids ejections and daily cleaning. Particle size distribution, density difference and viscosity of the juice or beer replace those of milk; filtration of beer through kieselguhr or membranes is governed by cake and membrane resistance and is not covered.
  • Wheat-flour sifting: separation by size on sieves, not by density; the rate law (sieve loading and passage probability) and the quality law (overs, cross-contamination of streams) differ. Only the OEE structure carries over: availability bounded by dry sanitation (the milling model's weekly stop), quality by the share of off-specification flour and performance by the margin that keeps it there.

References

  • Tetra Pak Processing Systems. Dairy Processing Handbook, chapter 7.2 Centrifugal separators and milk standardization, and chapter 2 The chemistry of milk (online edition).
  • Kim, S.-C., Yun, S.-Y., Ahn, N.-H., Kim, S.-M., Imm, J.-Y. (2020). Effect of homogenization pressure on plasmin activity and mechanical stress-induced fat aggregation of commercially sterilized ultra high temperature milk during storage. Food Science of Animal Resources, 40(5), 734–745. doi:10.5851/kosfa.2020.e48
  • Lopez, C., Cauty, C., Guyomarc'h, F. (2015). Organization of lipids in milks, infant milk formulas and various dairy products: role of technological processes and potential impacts. Dairy Science & Technology, 95, 863–893. doi:10.1007/s13594-015-0263-0
  • Ambler, C. M. (1959). The theory of scaling up laboratory data for the sedimentation type centrifuge. Journal of Biochemical and Microbiological Technology and Engineering, 1(2), 185–205. doi:10.1002/jbmte.390010206
  • U.S. Food and Drug Administration. Grade "A" Pasteurized Milk Ordinance, 2023 Revision.
  • European Commission, Joint Research Centre. Best Available Techniques (BAT) Reference Document for the Food, Drink and Milk Industries, 2019.
  • U.S. Department of Energy. Bandwidth Study on Energy Use and Potential Energy Savings Opportunities in U.S. Food and Beverage Manufacturing, 2017.
  • Bell, I. H., Wronski, J., Quoilin, S., Lemort, V. (2014). Pure and pseudo-pure fluid thermophysical property evaluation and the open-source thermophysical property library CoolProp. Industrial & Engineering Chemistry Research, 53, 2498–2508. doi:10.1021/ie4033999

7Parameters and provenance

Every input with its value, unit, role, uncertainty distribution and provenance. Open a quote to read the passage of the source that states the value.
  • xFx_F4 %

    Fat content of the whole milk fed to the separator

    Scenario · Uniform 3.5 to 4.5 %

    Declared assumption

    Declared raw-milk fat content; replace with the plant's intake average.

  • xCx_C40 %

    Fat content of the cream leaving the separator

    Scenario · Fixed

    Verified quote
    Show verbatim quote
    Assume that cream with a fat content of 40% is required.Dairy Processing Handbook, chapter 7.2: Centrifugal separators and milk standardization (opens in a new tab). Tetra Pak Processing Systems AB (online edition, openly readable), 2026. Tetra Pak Dairy Processing Handbook, online edition, chapter 7.2 (read 2026-11-15)
  • xS,tx_{S,t}0.05 %

    Residual-fat target of the skim milk

    Scenario · Uniform 0.04 to 0.07 %

    Declared assumption

    Declared target inside the documented skimming ability of separators (tetrapak_dph_separators: 'The remaining fat content in the skim milk normally lies between 0.04 and 0.07%'); 0.05 % is a design choice of the plant, not a legal limit.

  • σS\sigma_S0.003 %

    Standard deviation of the skim-milk fat content at constant flow (feed fat, temperature and discharge disturbances)

    Scenario · Uniform 0.002 to 0.005 %

    Declared assumption

    Process-capability assumption; replace with the measured value of the in-line fat analyser.

  • zz3

    Margin of the mean skim fat below the target, in standard deviations

    Scenario · Uniform 2.5 to 3.5

    Declared assumption

    Operating choice; the point value 3 is the centre of the declared range 2.5-3.5.

  • dv50d_{v50}3.41 um

    Volume-weighted median diameter of the native fat globules of raw milk

    Property · Normal, sd 0.2 um

    Verified quote
    Show verbatim quote
    The median particle diameter (D 50 ) of raw milk was about 3.41 μm and it was reduced by 0.358 to 0.648 μm by two-stage homogenization.Effect of Homogenization Pressure on Plasmin Activity and Mechanical Stress-Induced Fat Aggregation of Commercially Sterilized Ultra High Temperature Milk during Storage (Kim, Yun, Ahn, Kim, Imm; Food Science of Animal Resources 40(5): 734-745) (opens in a new tab). Korean Society for Food Science of Animal Resources (open access, CC BY-NC 3.0); full text served by Europe PMC, 2020. doi:10.5851/kosfa.2020.e48; PMC7492180
  • d43d_{43}4.4 um

    Volume-weighted mean diameter of native milk fat globules (cream from a plate separator)

    Property · Normal, sd 0.2 um

    Verified quote
    Show verbatim quote
    The size distribution of milk fat globules ranges from 1 to 12 μm with a mean volume-weighted diameter of 4.4 ± 0.2 μm (surface area = 1.8 m 2 .g -1 fat).Organization of lipids in milks, infant milk formulas and various dairy products: role of technological processes and potential impacts (Lopez, Cauty, Guyomarc'h; Dairy Science & Technology 95: 863-893) (opens in a new tab). Springer / INRA (open access); full text served by Europe PMC, 2015. doi:10.1007/s13594-015-0263-0; PMC4641158
  • SfatS_{fat}1.8 m²/g

    Specific surface of the native fat globules per gram of fat (same measurement)

    Property · Fixed

    Verified quote
    Show verbatim quote
    The size distribution of milk fat globules ranges from 1 to 12 μm with a mean volume-weighted diameter of 4.4 ± 0.2 μm (surface area = 1.8 m 2 .g -1 fat).Organization of lipids in milks, infant milk formulas and various dairy products: role of technological processes and potential impacts (Lopez, Cauty, Guyomarc'h; Dairy Science & Technology 95: 863-893) (opens in a new tab). Springer / INRA (open access); full text served by Europe PMC, 2015. doi:10.1007/s13594-015-0263-0; PMC4641158
  • ρf\rho_f930 kg/m³

    Density of milk fat (15.5 °C), used for the specific-surface conversion and the fat volume

    Property · Fixed

    Verified quote
    Show verbatim quote
    the lightest (density at 15.5 °C = 0.93 g/cm3)Dairy Processing Handbook, chapter 2: The chemistry of milk (opens in a new tab). Tetra Pak Processing Systems AB (online edition, openly readable), 2026. Tetra Pak Dairy Processing Handbook, online edition, chapter 2 (read 2026-11-15)
  • ρm\rho_m1,028 kg/m³

    Density of whole milk (converts the sediment load to a mass share)

    Property · Fixed

    Verified quote
    Show verbatim quote
    The density of cows’ milk normally varies between 1.028 and 1.038 g/cm3, depending on its composition.Dairy Processing Handbook, chapter 2: The chemistry of milk (opens in a new tab). Tetra Pak Processing Systems AB (online edition, openly readable), 2026. Tetra Pak Dairy Processing Handbook, online edition, chapter 2 (read 2026-11-15)
  • Δρ\Delta\rho48 kg/m³

    Density difference between skim milk and fat globules in the handbook's warm-milk reference state

    Property · Fixed

    Verified quote
    Show verbatim quote
    (ρp – ρl) = (980 – 1 028) = – 48 kg/m3Dairy Processing Handbook, chapter 7.2: Centrifugal separators and milk standardization (opens in a new tab). Tetra Pak Processing Systems AB (online edition, openly readable), 2026. Tetra Pak Dairy Processing Handbook, online edition, chapter 7.2 (read 2026-11-15)
  • ηref\eta_{ref}1.42 mPa·s

    Viscosity of the milk serum in the handbook's reference state

    Property · Fixed

    Verified quote
    Show verbatim quote
    h = 1.42 cP (centipoise) = 1.42 x 10–3 kg/m, sDairy Processing Handbook, chapter 7.2: Centrifugal separators and milk standardization (opens in a new tab). Tetra Pak Processing Systems AB (online edition, openly readable), 2026. Tetra Pak Dairy Processing Handbook, online edition, chapter 7.2 (read 2026-11-15)
  • TrefT_{ref}35 °C

    Temperature of the handbook's reference state

    Property · Fixed

    Verified quote
    Show verbatim quote
    The following average values apply at an ambient temperature of about 35 °CDairy Processing Handbook, chapter 7.2: Centrifugal separators and milk standardization (opens in a new tab). Tetra Pak Processing Systems AB (online edition, openly readable), 2026. Tetra Pak Dairy Processing Handbook, online edition, chapter 7.2 (read 2026-11-15)
  • TsepT_{sep}60 °C

    Separation temperature (warm separation after the regenerative preheater)

    Scenario · Uniform 55 to 65 °C

    Declared assumption

    Centre of the documented range (tetrapak_dph_separators: 'Usually, whole milk is heated to 55 – 65 °C in the pasteurizer before being separated.').

  • nn5,400 rpm

    Bowl speed of the reference separator

    Technology · Fixed

    Verified quote
    Show verbatim quote
    a radial position of 0.2 m in a centrifuge rotating at a speed of n = 5,400 rpmDairy Processing Handbook, chapter 7.2: Centrifugal separators and milk standardization (opens in a new tab). Tetra Pak Processing Systems AB (online edition, openly readable), 2026. Tetra Pak Dairy Processing Handbook, online edition, chapter 7.2 (read 2026-11-15)
  • ror_o0.2 m

    Radius at which the liquid leaves the rotating bowl (reference separator)

    Technology · Fixed

    Verified quote
    Show verbatim quote
    a radial position of 0.2 m in a centrifuge rotating at a speed of n = 5,400 rpmDairy Processing Handbook, chapter 7.2: Centrifugal separators and milk standardization (opens in a new tab). Tetra Pak Processing Systems AB (online edition, openly readable), 2026. Tetra Pak Dairy Processing Handbook, online edition, chapter 7.2 (read 2026-11-15)
  • csedc_{sed}0.1 kg/m³

    Mass of separable sediment per volume of milk

    Property · Uniform 0.05 to 0.2 kg/m³

    Verified quote
    Show verbatim quote
    The total amount of sediment in milk varies but can be about 1 kg/10,000 litres.Dairy Processing Handbook, chapter 7.2: Centrifugal separators and milk standardization (opens in a new tab). Tetra Pak Processing Systems AB (online edition, openly readable), 2026. Tetra Pak Dairy Processing Handbook, online edition, chapter 7.2 (read 2026-11-15)
  • VsedV_{sed}10 L

    Volume discharged at each solids ejection (sediment space of the bowl)

    Technology · Uniform 3.5 to 30 L

    Declared assumption

    Declared for a 25 m3/h separator inside the documented range (tetrapak_dph_separators: 'it can be from 3.5 litres up to 30 litres'); a full discharge of the sediment space is assumed at each ejection.

  • tejt_{ej}45 min

    Interval between solids ejections

    Technology · Uniform 30 to 60 min

    Declared assumption

    Centre of the documented range (tetrapak_dph_separators: 'Solids ejection is normally carried out at 30-to-60- minute intervals during milk separation.').

  • V˙\dot V25 m³/h

    Nominal feed flow of the separator (line capacity)

    Scenario · Fixed

    Verified quote
    Show verbatim quote
    the introduction of partial homogenisation into a pasteurisation line with a nominal capacity of 25 000 l/hBest Available Techniques (BAT) Reference Document for the Food, Drink and Milk Industries (opens in a new tab). European Commission, Joint Research Centre; Publications Office of the European Union, 2019. p. 404. EUR 29978 EN; JRC118627; doi:10.2760/243911
  • tcyct_{cyc}24 h

    Maximum interval between two cleanings of milk equipment

    Regulatory · Fixed

    Verified quote
    Show verbatim quote
    all equipment is thoroughly cleaned at least once each day usedGrade "A" Pasteurized Milk Ordinance, 2023 Revision (opens in a new tab). U.S. Department of Health and Human Services, Public Health Service, Food and Drug Administration, 2023. p. 98. PMO 2023 Revision
  • tCIPt_{CIP}0.374294 h

    Minimum duration of a cleaning-in-place programme (output cip_time_min of the cleaning model, with its Monte Carlo interval)

    Property · Triangular 0.242184, 0.374294, 0.536161 h

    Computed

    Output `cip_time_min` of the cleaning model: 22.46 min [14.53, 32.17].

8Benchmarks

Documented values from open sources used to check the ceilings; no documented value may beat the theoretical limit.
BenchmarkCeilingValueProvenance
Documented skimming ability of milk separators, lower value (residual fat in skim milk, % ; a spread of practice, not a ceiling of this model)
Show verbatim quote
The remaining fat content in the skim milk normally lies between 0.04 and 0.07%Dairy Processing Handbook, chapter 7.2: Centrifugal separators and milk standardization (opens in a new tab). Tetra Pak Processing Systems AB (online edition, openly readable), 2026. Tetra Pak Dairy Processing Handbook, online edition, chapter 7.2 (read 2026-11-15)
Best documented0.04 %Verified quote
Documented skimming ability of milk separators, upper value (residual fat in skim milk, %)
Show verbatim quote
The remaining fat content in the skim milk normally lies between 0.04 and 0.07%Dairy Processing Handbook, chapter 7.2: Centrifugal separators and milk standardization (opens in a new tab). Tetra Pak Processing Systems AB (online edition, openly readable), 2026. Tetra Pak Dairy Processing Handbook, online edition, chapter 7.2 (read 2026-11-15)
Typical0.07 %Verified quote
Ratio of centrifugal to gravity flotation velocity in the handbook example (0.2 m, 5 400 rpm); reproduced by the output separation_factor
Show verbatim quote
The sedimentation velocity in the centrifuge is 3,896.0/0.6 ≈ 6,500 times faster.Dairy Processing Handbook, chapter 7.2: Centrifugal separators and milk standardization (opens in a new tab). Tetra Pak Processing Systems AB (online edition, openly readable), 2026. Tetra Pak Dairy Processing Handbook, online edition, chapter 7.2 (read 2026-11-15)
—6,500 Verified quote
DOE bandwidth study, dairy separation step, current typical on-site energy intensity (fuel and electricity at the plant; a wider boundary than the separator drive)
Show verbatim quote
Final Storage Separation 18 14 14 0 18 14 14 0Bandwidth Study on Energy Use and Potential Energy Savings Opportunities in U.S. Food and Beverage Manufacturing (opens in a new tab). U.S. Department of Energy, Office of Energy Efficiency and Renewable Energy, Advanced Manufacturing Office, 2017. p. 56. DOE/EE-1571; OSTI ID 1512366
Typical18 Btu/lbVerified quote
DOE bandwidth study, dairy separation step, state-of-the-art on-site energy intensity (same boundary)
Show verbatim quote
Final Storage Separation 18 14 14 0 18 14 14 0Bandwidth Study on Energy Use and Potential Energy Savings Opportunities in U.S. Food and Beverage Manufacturing (opens in a new tab). U.S. Department of Energy, Office of Energy Efficiency and Renewable Energy, Advanced Manufacturing Office, 2017. p. 56. DOE/EE-1571; OSTI ID 1512366
Best available technique14 Btu/lbVerified quote
DOE bandwidth study, thermodynamic minimum energy intensity for fluid milk
Show verbatim quote
Fluid Milk 0 192,877 0Bandwidth Study on Energy Use and Potential Energy Savings Opportunities in U.S. Food and Beverage Manufacturing (opens in a new tab). U.S. Department of Energy, Office of Energy Efficiency and Renewable Energy, Advanced Manufacturing Office, 2017. p. 26. DOE/EE-1571; OSTI ID 1512366
Theoretical0 Btu/lbVerified quote
Share of raw milk lost during processing, most of it through separator discharges (BAT reference document; a loss, not a yield fraction, and the sentence leaves its basis ambiguous: not comparable with yield_max)
Show verbatim quote
It is reported that in dairies 1 % of raw milk supplied is wasted and that 0.1 % of this is during the processing of the milk and most of that is lost via discharges during centrifugal separation.Best Available Techniques (BAT) Reference Document for the Food, Drink and Milk Industries (opens in a new tab). European Commission, Joint Research Centre; Publications Office of the European Union, 2019. p. 423. EUR 29978 EN; JRC118627; doi:10.2760/243911
Best documented0.1 %Verified quote

9Unavoidable stops

Stops that physics or regulation impose; they set the maximum availability.
  • Cleaning at least once per day of useVerified quote

    Hygiene regulation: milk equipment must be cleaned at least once each day used. The separator is cleaned in place with the pasteurizer circuit; its duration is the minimum programme of the cleaning model (a lower bound for the bowl, assumption a5). Solids ejections take tenths of a second and do not stop the feed.

    Show verbatim quote
    all equipment is thoroughly cleaned at least once each day usedGrade "A" Pasteurized Milk Ordinance, 2023 Revision (opens in a new tab). U.S. Department of Health and Human Services, Public Health Service, Food and Drug Administration, 2023. p. 98. PMO 2023 Revision

10Assumptions

  1. Fat globules are rigid spheres that rise independently in laminar flow between the discs (Stokes' law); the size distribution of native globules is log-normal by volume, with the median of raw milk and the width obtained from the volume-weighted and surface-weighted mean diameters of native globules (sigma² = ln(d43/d32)).
  2. The serum viscosity at the separation temperature is the handbook reference value scaled with the viscosity of water (IAPWS through CoolProp), i.e. a constant relative viscosity; the density difference is held at its reference value. Liquid fat expands more than the serum, so at 55–65 °C the true difference is somewhat larger and the capacity ceiling correspondingly conservative.
  3. Sigma theory: in a channel loaded at Q = v_g Sigma, globules larger than the critical diameter are all captured and a smaller globule of diameter d is captured with probability (d/d_c)²; the equivalent clarification area Sigma of the machine is not specified, so capacities are reported per square metre of Sigma.
  4. Skim fat under control is normally distributed; skim milk above the target is re-separated (a quality loss). Standardisation of the cream is downstream and outside this model.
  5. Availability: the only physically or legally required stop is the daily cleaning of milk equipment. Its duration is the minimum programme of the cleaning model, which is valid for recirculated pipe and plate circuits; the bowl and disc stack are outside that validity, so the linked duration is a lower bound.

11Range of validity

  • Warm (55–65 °C) centrifugal skimming of bovine whole milk in disc-stack separators with native, non-homogenized fat globules.
  • Juice and beer clarifiers and flour sifters are covered only as stated approximations (derivation, section 7).

12Sensitivity

Elasticity: relative change of the output for a relative change of each parameter. Spearman rank correlation from the Monte Carlo draws is shown alongside.
  • d43d_{43}Volume-weighted mean diameter of native milk fat globules (cream from a plate separator)
    elasticity -0.168Spearman ρ -0.27
  • SfatS_{fat}Specific surface of the native fat globules per gram of fat (same measurement)
    elasticity -0.168Spearman ρ —
  • ρf\rho_fDensity of milk fat (15.5 °C), used for the specific-surface conversion and the fat volume
    elasticity -0.168Spearman ρ —
  • σS\sigma_SStandard deviation of the skim-milk fat content at constant flow (feed fat, temperature and discharge disturbances)
    elasticity -0.081Spearman ρ -0.81
  • xS,tx_{S,t}Residual-fat target of the skim milk
    elasticity +0.070Spearman ρ +0.42
  • zzMargin of the mean skim fat below the target, in standard deviations
    elasticity -0.067Spearman ρ -0.23
  • tcyct_{cyc}Maximum interval between two cleanings of milk equipment
    elasticity +0.016Spearman ρ —
  • tCIPt_{CIP}Minimum duration of a cleaning-in-place programme (output cip_time_min of the cleaning model, with its Monte Carlo interval)
    elasticity -0.016Spearman ρ -0.08
  • xFx_FFat content of the whole milk fed to the separator
    elasticity +0.012Spearman ρ +0.02
  • xCx_CFat content of the cream leaving the separator
    elasticity -0.001Spearman ρ —
  • Output rises with the parameter
  • Output falls with the parameter

13Validation checks

A model is published only when every check passes.
  • Dimensional analysis19 outputs converted to their declared unitsPassed
  • Every cited value verified against its source22 verified, 0 failedPassed
  • No documented value beats the theoretical limitPassed
  • Ceilings in physical orderPassed
  • Fractions within 0 and 1Passed

14Causes of loss in this operation

Probability that each cause is documented as a relevant loss of a line, pooled over all segments.

15References

  1. Dairy Processing Handbook, chapter 7.2: Centrifugal separators and milk standardization (opens in a new tab). Tetra Pak Processing Systems AB (online edition, openly readable), 2026. Tetra Pak Dairy Processing Handbook, online edition, chapter 7.2 (read 2026-11-15)
  2. Dairy Processing Handbook, chapter 2: The chemistry of milk (opens in a new tab). Tetra Pak Processing Systems AB (online edition, openly readable), 2026. Tetra Pak Dairy Processing Handbook, online edition, chapter 2 (read 2026-11-15)
  3. Effect of Homogenization Pressure on Plasmin Activity and Mechanical Stress-Induced Fat Aggregation of Commercially Sterilized Ultra High Temperature Milk during Storage (Kim, Yun, Ahn, Kim, Imm; Food Science of Animal Resources 40(5): 734-745) (opens in a new tab). Korean Society for Food Science of Animal Resources (open access, CC BY-NC 3.0); full text served by Europe PMC, 2020. doi:10.5851/kosfa.2020.e48; PMC7492180
  4. Organization of lipids in milks, infant milk formulas and various dairy products: role of technological processes and potential impacts (Lopez, Cauty, Guyomarc'h; Dairy Science & Technology 95: 863-893) (opens in a new tab). Springer / INRA (open access); full text served by Europe PMC, 2015. doi:10.1007/s13594-015-0263-0; PMC4641158
  5. Grade "A" Pasteurized Milk Ordinance, 2023 Revision (opens in a new tab). U.S. Department of Health and Human Services, Public Health Service, Food and Drug Administration, 2023. p. 98. PMO 2023 Revision
  6. Best Available Techniques (BAT) Reference Document for the Food, Drink and Milk Industries (opens in a new tab). European Commission, Joint Research Centre; Publications Office of the European Union, 2019. p. 404. EUR 29978 EN; JRC118627; doi:10.2760/243911
  7. Bandwidth Study on Energy Use and Potential Energy Savings Opportunities in U.S. Food and Beverage Manufacturing (opens in a new tab). U.S. Department of Energy, Office of Energy Efficiency and Renewable Energy, Advanced Manufacturing Office, 2017. p. 56. DOE/EE-1571; OSTI ID 1512366
  8. doi:10.1002/jbmte.390010206 — Sigma concept (equivalent clarification area) for scaling sedimentation centrifuges.
  9. doi:10.5851/kosfa.2020.e48 — Fat globule size distributions of raw and homogenized milk.
  10. doi:10.1007/s13594-015-0263-0 — Size and specific surface of native milk fat globules.
  11. doi:10.1021/ie4033999 — Thermophysical property library (viscosity of water).