Entrainment Prediction in Kettle Reboilers, Why the Element Is Not Free to Size, and What CFD Is Actually For

Entrainment Control in LNG Pre-Cooling Kettle Reboilers: K Values, Disengagement and Shell Diameter
11/15/2026
Why a Shorter Kettle Bundle Can Need a Larger Shell: Length, Diameter and Which Area Binds
11/15/2026
Entrainment Control in LNG Pre-Cooling Kettle Reboilers: K Values, Disengagement and Shell Diameter
11/15/2026
Why a Shorter Kettle Bundle Can Need a Larger Shell: Length, Diameter and Which Area Binds
11/15/2026

Two things are commonly assumed about entrainment in a kettle reboiler, and both are wrong in a way that costs money. The first is that a capacity factor is enough: pick a K value, divide the vapour flow by the allowable velocity, get an area, done. The second is that the mist eliminator inside the vessel is a free design parameter that can simply be made larger when the number is short. In a reboiler it cannot. The element lives inside a shell whose diameter, bundle, nozzles, level range and manways were set before anyone asked about droplet capture.

This page covers how entrainment is actually predicted, why a reboiler constrains the element rather than the other way round, and where CFD earns its place in that chain.

How entrainment prediction is structured

The published structure of kettle entrainment prediction is consistent across the main sources, and it is iterative rather than closed-form. It runs in four steps:

  1. Entrainment ratio. The design specifies a maximum liquid mass per unit vapour mass at the outlet, typically a fraction of a per cent for a machine that must not see liquid. That figure is the target, not a result.
  2. Vapour velocity in the disengagement space. The velocity is a combination of the produced vapour flow and the actual kettle geometry, meaning the free area above the pool. It is not the nozzle velocity and not the tube-side velocity.
  3. Critical entrainment velocity. A threshold velocity at which shear force balances gravity and droplets of a given size are no longer retained. It is obtained from semi-empirical and theoretical approaches, and calculating it requires the target entrainment ratio, the vapour flow rate and the physical properties of both phases.
  4. Iterate the diameter. The kettle diameter is found by iteration, because changing the diameter changes the free area, which changes the velocity, which changes the predicted entrainment ratio.

The authoritative correlations for this are held in HTRI’s proprietary reports on kettle entrainment, referenced in the LNG pre-cooling literature as BK1-2 (analysis and prediction of entrainment and its application to kettle reboiler shell sizing) and BK1-10 (prediction of kettle entrainment). They are not public, and any page quoting a closed-form formula as the HTRI correlation is inventing it. What is public, and what belongs in a process specification, is the structure above and the inputs it needs.

That structure has a consequence worth stating plainly: the prediction depends on the geometry it is applied to. A single capacity factor assumes a uniform approach velocity across the whole free area. In a kettle the vapour does not arrive uniformly, because the heat flux along the bundle is not uniform, so the correlation is being asked to describe a field, not a point. That is why the same element can meet its number on paper and still pass liquid in service.

Why the reboiler does not let you size the element freely

Three constraints bind, and none of them is about droplet capture.

The dome is round. A horizontal element sits as a disc inside a circular shell. Its diameter is limited by the chord available at the height it is installed, not by the shell diameter. On a 3.9 m shell, an element whose plane sits 0.9 m above the shell centre line can be at most 3.46 m across:

Quantity Value Note
Shell inside diameter 3.9 m example, high-pressure propane stage
Element plane height above centre 0.9 m set by maximum level plus drainage space
Maximum element diameter 3.46 m chord at that height
Element area available 9.40 m2 against 11.94 m2 of full shell section
Capacity lost to geometry about 21% at 0.57 m/s, 5.36 m3/s instead of 6.81 m3/s

So the shell section buys 6.81 m3/s of capacity and the element actually gets 5.36 m3/s of it. That 21 per cent is not an inefficiency to be engineered away; it is the geometry. The only ways to recover it are to change the element type to one with a higher capacity factor, or to lower the level, which brings the second constraint.

Level and area are the same variable. Raising the liquid level improves the flooded condition of the bundle but moves the element plane upward into a narrower chord, and reduces the disengagement space to the outlet nozzle. On a 3.9 m shell, 0.1 m of rise near that height costs about 0.11 m of element diameter, and the loss accelerates towards the crown: around 0.26 m per 0.1 m once the plane reaches 1.5 m above the centre. A level control band of half a metre is therefore also an area band, and the element must be sized for the worst end of it, not the normal level.

The element must coexist with everything else in the shell. It cannot sit so low that it touches the bundle or blocks circulation; it must drain into the pool on the correct side; it must clear the nozzle projection and the manway, since it has to be installed and removed as modules; and it must survive the two-phase splash zone that the boiling surface produces. Those constraints, not the droplet cut point, usually decide the final layout.

This is the point the LNG pre-cooling literature makes when it treats in-vessel elimination as a design discipline rather than a purchase: the element’s position relative to the vapour generation is what a CFD study is for, and the disengagement space above the mesh to the nozzle is a named constraint, not an afterthought.

What this means in practice

Four rules follow directly, and they are the ones to check before accepting a sizing sheet:

  • Do not size the element on the shell’s full section. Size it on the disc available at the installed height, at the highest level in the control band.
  • Treat the level band as a capacity parameter. If the level range is wide, the achievable velocity is set by the top of the band, and a tighter level control is worth more than a bigger element.
  • Where the geometry is tight, element type earns its money. Inside a fixed chord, a double-pocket vane at 0.305 carries nearly three times the vapour of a vertical mesh at 0.107 for the same area. That is why compact rebollers use combinations rather than bigger pads.
  • Verify the local velocity, not the average. The correlation and the capacity factor both act on an average. The vessel acts on the maximum, which sits where the vapour is produced fastest.

Where CFD earns its place, and where it does not

CFD is used in this application for four specific questions, and none of them is “what is the entrainment rate”.

Question What CFD answers What it cannot
Does vapour short-circuit from a nozzle to the outlet? Plume trajectory and the position where the element is bypassed Whether the bypassed liquid actually reaches the compressor
Is the element loaded uniformly? Local velocity distribution across the disc, hot spots at the shell wall Absolute separation efficiency of the element
What level is safe? Sensitivity of local velocity and freeboard to the level band The effect of foaming, which is not in most models
Does the deflector plate work? Whether drainage returns to the pool or re-enters the rising flow The stability of the drainage over time and fouling

The honest position is that CFD is a distribution tool, not an efficiency tool. It can show you that one third of the element is doing most of the work, or that a nozzle is shooting straight through the middle of it, and those findings change the layout. It cannot tell you the mass of liquid that leaves the vessel unless it is validated against measurement, because the droplet size distribution entering the element, the coalescence inside the media and the film behaviour at the drainage edges are all sub-grid processes.

The verification chain

The defensible way to size this equipment is four layers, each of which covers a gap the others leave:

Layer Establishes Fails when
Element test data from the supplier Capacity factor, cut point, pressure drop, drainage behaviour Tested with air-water, applied to a refrigerant
Entrainment correlation for the vessel Required free area and diameter for the target entrainment ratio Applied to a geometry with a strongly non-uniform vapour source
CFD of the installed arrangement Local velocity, bypass paths, effect of level and deflector Boundary conditions taken from the average instead of the load profile
Operating data from comparable units Whether the prediction survives real transients, fouling and start-up Nobody recorded the level and load when the carryover occurred

The last layer is the one that is usually missing, and it is the cheapest to fix: a unit that logs level, load and any compressor liquid indication against time can tell you more about the real entrainment behaviour than a third CFD case.

What to require from a CFD deliverable

If a simulation is part of the justification, these are the items that decide whether it is evidence or decoration:

  • Boundary conditions from the actual vapour load profile, not a uniform inlet over the bundle. A uniform source models a vessel that does not exist.
  • Mesh independence shown, with the local velocity at the element reported at two mesh densities.
  • Residual and convergence criteria stated, and the case run to those criteria rather than to a wall-clock limit.
  • Sensitivity runs: at least 70 per cent, 100 per cent and 110 per cent load, and the top and bottom of the level band.
  • Maximum local velocity reported alongside the average, with its location identified.
  • Validation case against a measured arrangement, even if it is a small air-water rig. Without one the output is a picture, not a prediction.
  • Review by the element supplier, since the media behaviour the model cannot resolve is exactly what the supplier has tested.

The length and diameter consequences of a shorter bundle, stage by stage, are set out in why a shorter kettle bundle can need a larger shell.

Reference

The process context, capacity factors and the treatment of in-vessel elimination follow Knoepfler, Buffet, El Hajal, Lang, Provost, Rambure and Redjem Saad, LNG2023, Vancouver, paper 3.2.1, whose literature cites HTRI reports BK1-2 and BK1-10 for the entrainment correlations and the model of critical entrainment velocity used here.

Frequently Asked Questions

Where do the kettle entrainment correlations come from?

The published structure is iterative: specify the allowable liquid mass per unit vapour mass, compute the vapour velocity in the free area above the pool from the produced flow and the actual kettle geometry, obtain a critical entrainment velocity above which shear overcomes gravity, then iterate the shell diameter because changing it changes both. The authoritative correlations are held in proprietary HTRI reports on kettle entrainment, so a specification should state the target ratio and the inputs rather than quote a formula.

Why can a mist eliminator not simply be made larger inside a reboiler?

Because the element is a disc inside a circular shell, and its diameter is limited by the chord available at its installed height. On a 3.9 m shell an element sitting 0.9 m above the centre line can be at most 3.46 m across, which is about 21 per cent less area than the full shell section. It also has to clear the bundle, the nozzle projection and the manway, and drain into the pool on the correct side.

How does the liquid level affect separation capacity?

Directly and in the wrong direction. Raising the level moves the element plane into a narrower chord of the shell, so the same element loses area exactly when the vapour load is hardest to handle, and the freeboard to the outlet nozzle shrinks at the same time. That is why the element should be sized at the top of the level control band rather than at normal level.

What can CFD tell you that an entrainment correlation cannot?

Where the flow actually goes. It shows whether vapour short-circuits from a nozzle to the outlet, which part of the element is overloaded, how sensitive the local velocity is to the level band, and whether a deflector plate returns drained liquid to the pool or back into the rising flow. What it cannot do without validation is predict the absolute mass of liquid leaving the vessel.

What should be required of a CFD study submitted with a chiller design?

Boundary conditions derived from the real vapour load profile rather than a uniform bundle source, mesh independence shown at the element, convergence criteria stated, sensitivity runs at 70, 100 and 110 per cent load and at both ends of the level band, the maximum local velocity reported with its location, a validation case against a measured rig, and review by the element supplier whose test data covers what the model cannot resolve.

Leave a Reply