By pore size and diffusion speed rather than by chemical preference.

Carbon molecular sieve is manufactured with pores of a very specific size - close to the diameter of the gas molecules themselves. An oxygen molecule is slightly smaller than a nitrogen molecule, and at that scale the difference matters enormously: oxygen diffuses into the pores several times faster than nitrogen does.

So when compressed air is pushed through the bed, oxygen is drawn into the pore structure while nitrogen, moving much more slowly into the pores, continues through the bed and out as product.

This is a kinetic separation - it depends on the difference in rate, not on an equilibrium preference. Given enough time, nitrogen would also enter the pores and the separation would be lost.

That is precisely why the process is cyclic and the timing is critical: the bed must be switched while the oxygen is in and the nitrogen is still on its way in.

Because the sieve has to be emptied of the oxygen it captured, and because the separation only works within a limited time window.

During the production phase, oxygen accumulates in the pores. That capacity is finite, and as it fills the separation degrades. The bed is therefore taken offline and depressurised, which releases the adsorbed oxygen to atmosphere and regenerates the sieve, while the second vessel produces.

Cycle time is set by two competing pressures. It must be short enough that nitrogen has not begun diffusing into the pores in quantity, since that would contaminate the separation. It must be long enough to make efficient use of each pressurisation, because every switch vents a vessel's worth of compressed gas.

The manufacturer optimises this, and it is not a user setting. What the user sees is the consequence: switching valves cycling continuously, and a characteristic pulsing air demand that an inlet receiver is fitted to smooth.

Steeply, and the relationship is worth understanding before specifying.

Higher purity is achieved by taking a smaller fraction of the gas passing through the bed - drawing off only the leading portion, which is the most thoroughly separated, and venting more. The same machine therefore delivers considerably less product at high purity than at moderate purity.

The air consumed per unit of nitrogen produced rises at the same time, so the running cost per unit climbs on both counts: less product for the same machine, and more compressed air for each unit of that product.

The practical implication is that specifying purity generously 'to be safe' is expensive in three places at once - a larger generator, a larger compressor, and higher electricity for the installation's whole life.

Manufacturers publish flow at several purity levels. Compare quotations at your actual required purity, not at the headline figure, or the comparison is meaningless.

A stabilisation period rather than instantly - which matters for intermittent duty and for the startup procedure.

When a PSA machine starts from cold, the beds are at atmospheric conditions and the cycle has not established its steady-state profile. It takes a number of cycles for the adsorption and regeneration to settle into the pattern that produces the rated purity.

During that period the product is off specification.

Standard practice is therefore an automatic vent that dumps product to atmosphere until the purity analyser confirms specification, and only then opens the supply to the process. Without that, out-of-specification gas reaches the application at every startup, which on a food packaging or laser cutting line means scrapped product.

The consequence for duty pattern is that PSA suits continuous running. Where the machine will start and stop frequently, a membrane system - which is on specification almost at once - is often the better technology.

Valve service and upstream filtration, with the sieve itself usually lasting many years.

The carbon molecular sieve is not consumed in normal operation. Properly protected from oil and water, it lasts a long time and is not a routine replacement item.

What it is not protected against is contamination. Oil aerosol coats the sieve and blocks the pores permanently; liquid water floods it. Both destroy capacity irreversibly, and both come from inadequate air treatment upstream. Filter element changes on schedule are therefore the most important maintenance task on the installation, even though they are not on the generator itself.

The switching valves are the wearing part. They operate continuously and accumulate very large cycle counts, and they are overhauled or replaced at the manufacturer's stated interval.

Beyond that: calibrate the purity analyser, monitor the trend, and investigate a gradual purity decline promptly - it usually indicates a leaking valve or sieve degradation, and finding it early distinguishes a valve service from a sieve replacement.

Within the machine's range, yes - and it is one of the more useful properties of PSA.

Purity is governed by how much product is drawn from each bed before switching, so it is adjusted through the control settings rather than by changing hardware. A machine commissioned at one purity can generally be re-set to another within its design envelope.

The trade is immediate and unavoidable: raising purity reduces available flow and increases air consumption per unit produced. Lowering it does the reverse.

That flexibility is worth exploiting. Sites frequently commission at a conservatively high purity and never revisit it, paying the flow and energy penalty for years. If the process turns out to tolerate less, backing the setting off increases capacity and cuts running cost with no capital spend.

Confirm the achievable range with the manufacturer before assuming a large change is possible - the envelope is finite, and a machine sized for high purity is not necessarily efficient at low purity.

Contaminated sieve, leaking switching valves, or a change in the duty it is being asked to do - and they are distinguishable.

Sieve degradation from oil or water shows as a slow, steady decline over months that does not recover, accompanied by no change in the machine's mechanical behaviour. It traces back to filtration that was not maintained.

A leaking switching valve usually appears more abruptly and often with a change in the cycle sound or in air consumption, because gas is escaping or crossing between vessels at the wrong point in the cycle.

A duty change is the one most often missed: demand has grown, or a new user has been connected, and the machine is simply being asked for more flow than it can deliver at that purity. Nothing is faulty.

Check consumption against the original design before investigating the machine. It is the cheapest diagnostic and it is right more often than people expect.