

Commissioning a direct air capture module produces a failure that has no equivalent anywhere in oil and gas. The contactor is evacuated for the leak test, it will not hold, and the leak is never where the team looks first. It is not on the seat. It is on the stem of a DN50 valve that has never carried more than two bar in its life, and it is not a manufacturing defect: the packing was a chevron set that depends on internal pressure to spread its lips against the stem, and the test pulled the pressure the other way.
The second leak on the same module was at a body joint, on a spiral wound gasket that had passed a hydrostatic proof at 150 percent of its rating a fortnight earlier. At 0.9 bar of external pressure the gasket was drawn inward rather than pushed outward, and the inner windings lifted off their seating face.
Two leaks, one cause. Every sealing element on that module had been selected, tested and certified for pressure acting from inside to outside. Half of the plant operates with the pressure acting in the opposite direction, and nobody had written that sentence into a specification. This article is about what a solid sorbent direct air capture plant actually asks of its valves: why the fleet is small, cheap and numerous, why the vacuum side breaks the rules the rest of the industry works to, and what has to be written in a purchase order so that the commissioning test is not the first time anyone thinks about direction.
1. A plant built around four hundred and twenty parts per million
Carbon dioxide is 420 parts per million by volume in air, which is 0.064 percent by mass once the molecular weights are taken into account. Divide one by the other and a kilogram of product requires 1,570 kilograms of air, or about 1,300 cubic metres at ambient density. A tonne of CO2 therefore needs 1.3 million cubic metres of atmosphere to pass through the contactor, and For a module capturing 4,000 tonnes a year over 8,000 operating hours that is 650,000 cubic metres an hour, and at a face velocity of 1.5 metres per second it is 121 square metres of sorbent bed running continuously.
That arithmetic has a direct consequence for the valve list, and it is the opposite of what most engineers expect. Because the volumetric flow is enormous and the available pressure drop is almost nothing, the main air path is built from dampers, louvres and large rectangular ducts. No ball valve appears there. What the ball valves sit on is the auxiliary circuits, and those are small: steam, condensate, purge gas, product CO2, vacuum pump isolation, sorbent make up, instrument air. The largest isolation valve on a four thousand tonne a year module is typically a DN150 on the steam header. Everything else is DN15 to DN80, and the whole fleet costs less than a single cryogenic body in an LNG terminal.
So the selection problem is not one of pressure or size. It is one of quantity, cycle count and direction, and all three of those are invisible on a datasheet.
Circuit | Bore | Absolute pressure, bar | Metal temperature, C | Strokes per day | What ends it |
|---|---|---|---|---|---|
Contactor air inlet | DN800 to DN1600 damper | 1.01 | 5 to 45 | 30 | Pressure drop, not wear |
Desorption steam isolation | DN80 to DN150 | 2 to 4 | 120 to 145 | 60 | Seat thermal cycling |
Product CO2 outlet | DN50 to DN100 | 0.1 to 0.3 | 60 to 100 | 60 | Inward leakage of air |
Vacuum pump isolation | DN80 to DN150 | 0.05 to 0.2 | 40 to 80 | 30 | Stem seal direction |
Condensate drain | DN15 to DN40 | 0.1 to 1.0 | 60 to 95 | 60 | Carbonic acid at pH 4.5 |
Sorbent make up and dump | DN50 | 1.01 | 20 to 110 | 1 per year | A bead crushed on the seat |
Purge and seal gas | DN25 to DN50 | 1.0 to 3.0 | 20 to 110 | 120 | Stem O-ring hardening |
Instrument air to pilots | DN15 | 6 to 8 | 20 | 240 | Water and oil carryover |
2. The seat that vents into your vacuum
Most of the valves in that table are DN50 floating ball valves, because that is what a DN50 line at two bar needs. A floating ball valve with a single piston seat has a designed relief path from the ball cavity to the upstream port. Its purpose is to stop the cavity from being pressurised by thermal expansion of trapped liquid, and in nine applications out of ten it is the correct and necessary arrangement.
In a vacuum desorption line it is a permanent leak. Close the valve with air in the cavity and a line at 0.1 bar absolute behind it, and the cavity now sits 0.9 bar above the line. The relief path opens, exactly as designed, and holds it open for as long as the module remains evacuated. The open area is not a scratch, it is the seat's own relief clearance, a few square millimetres. Three square millimetres at 0.9 bar with a discharge coefficient of 0.62 passes 0.62 times 3e-6 times 387, which is 7.2e-4 cubic metres per second, or 2.6 cubic metres an hour. That is not a weeping seat. That is a valve which is functionally open while the limit switch reports closed.
The fix is a double piston seat, and then the cavity has to be dealt with separately. Once both seats block in both directions the cavity is isolated at both ends, so it needs its own relief to a safe location and its own pressure transmitter, because a cavity that cannot vent is a pressure vessel that nobody is monitoring. The same volume of air trapped at 1 bar and heated from 20 to 110 degrees C during desorption reaches 1.31 bar at constant volume, which is harmless, but only if somebody has done that sum.
Seat arrangement | What happens at 0.9 bar reverse differential | Fits a vacuum line |
|---|---|---|
Single piston, self relieving | Relief path from cavity to upstream port opens and stays open | No |
Double piston, unidirectional | Upstream seat blocks, cavity has no path to either line | Only with external cavity relief |
Double piston, bidirectional | Both seats seal both ways, cavity fully isolated | Yes, with relief and a gauge |
Elastomer O-ring seat, pressure energised | Seal is bidirectional by geometry, but the ring is drawn into the gap | Yes, if the gap stays under 0.15 mm |
Metal seat with graphite insert | Leaks by design at low differential | No |
3. Three hundred and thirty thousand strokes
A temperature vacuum swing cycle on a solid sorbent runs between thirty and sixty minutes. Take forty five minutes, which is thirty cycles a day, eleven thousand a year, and three hundred and thirty thousand over a thirty year plant life. Two of those strokes are the valve opening and closing, which means the actuator and the stem seal see six hundred and sixty thousand seal passes, and the seat insert sees three hundred and thirty thousand load and unload events.
A soft seated ball valve is normally quoted at a hundred thousand cycles. In this duty that is nine years, which lands inside the first major overhaul rather than beyond it, and on a module with a hundred and forty valves it turns into a scheduled consumable of a hundred and forty seat kits. That is not a defect in the valve; it is a fact about the duty that belongs in the maintenance plan and in the spare parts budget on day one.
The second consequence of the cycle is thermal, and it is the one that gets missed. The bed swings ninety kelvin on every cycle. A seat ring at a mean radius of eighty millimetres grows radially by 12e-5 times 90 times 0.08, which is 0.86 mm, while the stainless housing around it grows 0.12 mm. The seat therefore moves about three quarters of a millimetre relative to the body, once per cycle, three hundred and thirty thousand times. A wave spring with one millimetre of travel absorbs that comfortably, but a wave spring is rated by stroke rather than by load: the same spring quoted at a million cycles at 0.75 mm may be quoted at a hundred thousand at 2 mm. Specify the stroke and the cycle count together, or the spring becomes the item that fails while the seat is still serviceable.
4. A bed that is slowly turning into dust
Solid sorbents are beads between 0.3 and 1.0 mm, and they attrit. A contactor holding forty tonnes of sorbent at one percent loss a year generates four hundred kilograms of fines below fifty microns, and those fines go wherever the air goes, which includes the seat pocket of every valve on the purge and product lines.
A single 0.5 mm fragment caught on the three to five millimetre contact band of a PTFE seat is enough. The ball closes on it, the polymer yields around it, and the indentation of 0.05 to 0.1 mm survives the fragment being blown away. That is not a leak anyone finds at one bar. It is a leak that appears at 0.1 bar absolute, on a test that most suppliers have never run, and it gets attributed to the seat material rather than to the particle that made the mark.
There are three practical answers and only one of them is a valve feature. Mount the valve so the cavity drain points down, so debris falls away from the seat rather than settling on it. Purge the seat pocket with clean gas before the final closing movement on the product line. And where the duty justifies it, fit a scraper ring ahead of the seat so the ball wipes itself clean on the way in, which is standard on slurry valves and almost never offered on a DN50 process valve unless it is asked for.
5. Steam, condensate and the acid nobody sizes
Desorption heat for an amine functionalised sorbent is around 3 gigajoules per tonne of CO2. At 2 bar the latent heat of steam is 2.2 megajoules per kilogram, so a tonne of product consumes roughly 1.4 tonnes of steam, and every kilogram of that steam becomes a kilogram of condensate. A four thousand tonne a year module therefore produces about 5,600 tonnes of condensate a year, and that condensate has been in contact with CO2 at a partial pressure high enough to take a fraction of a gram per kilogram into solution.
The result is water at pH 4.5 to 5.5 at 60 to 95 degrees C. Type 316L handles that at rates of a few hundredths of a millimetre a year, so the body material is not the problem. What fails is everything that is not the body: zinc plated stems and bonnet bolts, because zinc is amphoteric and dissolves in acid as readily as in alkali; carbon steel drain plugs; and the threaded joints, where a crevice holds the acid against a surface that never sees the bulk flow. The specification line that prevents most of it is short: no zinc on any wetted or condensate exposed part, and the drain valve in 316L rather than in the carbon steel that the line is made of.
6. The flange loaded from the wrong side
At 0.1 bar absolute the net external pressure on every joint is 0.9 bar. On a DN100 flange that is only a few hundred newtons across the gasket, which sounds negligible, and the force is not the issue. The direction is. A spiral wound gasket is located by friction and by the compression of its windings, and under external pressure the inner edge is drawn toward the bore. Without an inner ring the windings can lift and unwind at the inner diameter, and the joint that passed a hydrostatic proof at 15 bar will pass a vacuum test at 0.9 bar only by luck.
This is the single most useful sentence in this article for anyone writing a specification: a proof test at 150 percent of the rating demonstrates that a joint holds pressure in, and demonstrates nothing at all about whether it holds air out. The two are different load cases, they load different faces of the same gasket, and only one of them is normally tested. Ask for an inner ring on every spiral wound gasket, ask for a tongue and groove or an O-ring groove on soft gaskets, and ask for the body joint to be included in the vacuum leak test rather than only in the hydrostatic one.
7. Sizing a line that sits at a tenth of an atmosphere
Valve sizing in this plant is not a flow coefficient exercise. On the main air path the velocity is 1.5 metres per second and the valve contributes twenty or thirty pascals, which is inside the noise of the calculation. On the vacuum side the situation inverts: because the gas is at a tenth of atmospheric density, the volumetric flow at the pump inlet is ten times what the same mass flow would occupy at ambient, and the line has to be sized on that expanded volume.
A vacuum pump moving 200 cubic metres an hour at its inlet, connected through a DN100 line at 0.1 bar absolute, runs that line at 7 metres per second. The same 200 cubic metres an hour at atmospheric pressure in the same pipe is 0.7 metres per second. Nothing about the pipe has changed; the density has, by a factor of ten, and a line sized on the atmospheric figure will be running at ten times the velocity its designer intended. The velocity limit here is not erosion, since there is almost no kinetic energy to erode with, but pressure drop, and the available driving force is the whole 0.9 bar that the vacuum pump is fighting for.
8. Where the energy actually goes
Two energy numbers decide whether a valve is cheap or expensive in this plant, and neither of them appears in a valve quotation.
The first is the fan work on the main air path. Moving 1.3 million cubic metres of air through a tonne of CO2 at a total system pressure drop of 250 pascals and a fan efficiency of 0.75 consumes 121 kilowatt hours. Every additional 100 pascals on that path is another 48 kilowatt hours per tonne, which at eight cents a unit is about four dollars per tonne of product, on a process whose entire value proposition is measured in dollars per tonne. A valve or damper that adds 30 pascals to that path costs more in twenty years than it costs to buy, and the reverse is also true: a full port design that removes 20 pascals pays for itself without any maintenance argument at all.
The second is the vacuum work. A tonne of CO2 at 0.1 bar absolute and 60 degrees C occupies 6,293 cubic metres. Compressing that isothermally back to atmosphere is 1e4 times 6,293 times the natural log of ten, which is 40 kilowatt hours. Drop the desorption pressure to 0.05 bar and the same tonne occupies twice the volume and costs about 52 kilowatt hours, so the choice of desorption pressure is a twelve kilowatt hour per tonne decision that is usually made on sorbent chemistry and rarely checked against the vacuum system.
Air leaking into that volume, by contrast, costs almost nothing to pump. One cubic metre an hour of leakage is a rounding error in the pump duty. Leaks in a direct air capture plant are a purity problem rather than an energy problem: two percent air ingress puts 4,200 parts per million of oxygen into a product stream that a transport network may cap at a hundred, and two percent air is a 3.5 mm hole. Nothing about a seat leak looks like a 3.5 mm hole, which is exactly why vacuum tightness on these lines should be specified as a leak rate or a pressure rise and not as a bubble count.
9. What to write in the order
Seven lines convert all of the above into something a supplier can be held to, and none of them adds cost to a competent bid.
First, state the absolute pressure and the direction of the differential for the seat test. A seat proved at 110 percent from the pressure side has not been proved at all for a line that will be evacuated, so add a reverse test at 0.9 bar with the bore evacuated and the cavity at atmosphere.
Second, name the seat arrangement: double piston, bidirectional, with an external cavity relief and a cavity pressure transmitter. Do not leave the phrase self relieving anywhere near a vacuum line.
Third, make the stem seal explicitly bidirectional, and give it a number. Either a helium leak rate of 1e-4 millibar litres per second, or a pressure rise test on an isolated section: fifty litres that gains ten millibar in an hour is leaking at 0.14 millibar litres per second, which is a figure a site can measure with a gauge and a stopwatch.
Fourth, gaskets with inner rings, and the body joint included in the vacuum test.
Fifth, a cycle life stated at the actual stroke and the actual temperature, three hundred and thirty thousand operations, with the seat kit part number and its price fixed at the time of order rather than quoted at year nine.
Sixth, a scraper or wiper on any seat exposed to sorbent fines, or a written statement that the supplier does not offer one so that the alternative can be designed in.
Seventh, 316L and no zinc for anything on the condensate side.
J-VALVES builds ball valves from DN15 to DN1400 in ASME classes 150 to 2500, covering a band that spans minus 196 C to plus 650 C. Every shell is hydro-tested at 150 percent of its rating and every seat at 110 percent, and the test report names the side the pressure came from. None of that is what a direct air capture plant needs from us. What it needs is a seat that holds 0.9 bar in the reverse direction and a stem seal that does not care which side the pressure is on, and both of those are quotable options rather than standard features, which means they have to be asked for. Send the absolute pressure, the strokes per year and the line size, and a priced selection comes back the same working day.
10. Frequently Asked Questions
Why does a ball valve that passed a hydrostatic test still fail a vacuum test?
Because the two tests load the sealing elements in opposite directions. A hydrostatic proof at 150 percent of the rating pushes the gasket outward, energises a chevron packing from inside and drives the seat toward the ball. At 0.1 bar absolute the same joint sees 0.9 bar from outside: the gasket is drawn toward the bore, the packing lips are pulled away from the stem and any self relieving seat opens its cavity vent. Every one of those elements was selected for the direction that was tested.
Which seat arrangement should be used on the vacuum side of a DAC module?
A double piston bidirectional seat, with the cavity relieved externally through its own line and monitored with a pressure transmitter. A single piston self relieving seat has a designed vent from the cavity to the upstream port, and in a vacuum line the cavity is always above line pressure, so that vent stays open and passes about 2.6 cubic metres an hour at 0.9 bar. The relief is not a fault, it is the design working, which is why it cannot be tested for.
How many cycles does a direct air capture valve see in its life?
A forty five minute temperature vacuum swing cycle gives thirty a day, about eleven thousand a year, and three hundred and thirty thousand over a thirty year plant. A soft seated ball valve quoted at a hundred thousand cycles is therefore consumed in nine years, which falls inside the first major overhaul. On a module with a hundred and forty valves that is a hundred and forty seat kits, and the kit part number and price should be fixed at the time of the valve order.
Does the low CO2 concentration change the material selection?
Not for the air side, which is simply air at ambient temperature and needs no exotic alloy. It changes the material selection on the condensate side, where the desorbed CO2 dissolves to give water at pH 4.5 to 5.5 at 60 to 95 degrees C. Type 316L is adequate there at a few hundredths of a millimetre a year. What fails is zinc plated hardware, which is amphoteric and dissolves in acid, and carbon steel drain plugs left in an otherwise stainless line.
Can a standard soft seated ball valve hold 0.1 bar absolute?
The seat usually can, because a 0.9 bar differential against a floating ball gives a few kilonewtons of seating force and the polymer deforms enough to seal. The joints are what fail. A stem packing energised by internal pressure leaks inward, a spiral wound gasket without an inner ring lifts at its inner edge, and a self relieving seat vents the cavity straight into the line. Specify the stem seal as bidirectional and the gaskets with inner rings, and the seat becomes the easy part of the problem.
