On-Site Oxygen Generators: PSA Technology and Industrial Applications

PSA broke the lock-in. A factory buys a machine, connects power, and makes its own oxygen from ambient air. On-site generation costs a quarter to a third of delivered LOX for any operation above about 500 Nm³/day. The major gas companies responded by developing their own PSA and VPSA product lines and, in some cases, offering build-own-operate arrangements that preserved the subscription revenue structure even with on-site equipment.

LiLSX, lithium-exchanged low-silica X-type zeolite, sits inside every modern PSA oxygen generator. Nitrogen selectivity runs about six to one, triple the older 13X it replaced. The performance gap opened through the 1990s, driven by work at Air Products and CECA (now Arkema), and it is what turned oxygen PSA from a specialty product into a standard supply method.

The manufacturing process matters and the people buying PSA equipment generally do not think about it. You start with a low-silica NaX base zeolite, then run lithium ion exchange in a heated solution. Temperature, concentration, and contact time all affect how completely the lithium replaces the sodium on the cation sites. At 96% exchange the material performs to spec. At 92% it underperforms in ways that the equipment buyer cannot detect without running an adsorption isotherm on a pulled sample, which equipment buyers do not do. The OEM receives a certificate of analysis with each shipment. The good OEMs verify it. The others do not.

The reason lithium exchange percentage matters is that lithium cations have a stronger electric field gradient at the adsorption site than sodium cations. Nitrogen molecules, with their quadrupole moment, feel that gradient and are more strongly attracted to lithium sites. Every sodium cation that was not exchanged during manufacturing is a site that captures nitrogen less effectively. The effect is nonlinear. The last few percent of exchange, going from 92% to 96%, matter more than the improvement from 80% to 92% because those remaining sodium sites tend to be in locations where they are harder to reach during the exchange process, and those locations also happen to be the sites that contribute most to nitrogen uptake in the high-pressure region of the isotherm where PSA operates.

Crush strength is a different manufacturing variable and its consequences show up differently. Weak beads fracture under cycling. The fragments drop to the support screen at the bottom of the vessel and accumulate over years. Eventually the fines layer gets thick enough to create channeling. You can see it at zeolite changeout. Pop the manhole, look at the bottom of the bed, and the depth of fines sitting on the screen says everything about the crush strength of the original bead charge.

Particle size distribution, lithium exchange ratio, crush strength, residual moisture at the time of loading. These vary between production batches. Two systems from the same manufacturer, same model number, commissioned six months apart, can produce different oxygen output because the zeolite differs. Arkema’s CECA division, Zeochem, Shanghai Hengye, Luoyang Jianlong supply most of the world volume. Pricing is bilateral, unpublished, and varies by a factor of two depending on purchase volume. Replacement zeolite after eight to fifteen years puts the OEM in the driver’s seat. The system is installed. The customer needs the OEM for vessel access and re-commissioning. Aftermarket pricing at 1.5 to 3 times the original cost is standard.

Nitrogen sticks to the zeolite. Oxygen does not. That is the entire separation. The physics underneath is electrostatic. Nitrogen’s quadrupole moment is about four times oxygen’s. Lithium cations on the zeolite surface attract nitrogen preferentially. The gas that exits the bed is enriched in oxygen because the nitrogen has been detained. Argon caps the purity. Atmospheric argon is 0.934% of air, and zeolite cannot distinguish it from oxygen. Peak PSA purity is about 95.5%, with the product containing roughly 93% O₂, 4% Ar, 2% N₂.

At 90% purity, recovery is around 52%. At 95%, high 30s. The curve steepens near the top. Procurement specs routinely call for 93 to 95% without any process engineering analysis of what the application needs. Wastewater runs fine at 88 to 90%. Aquaculture runs fine at 88 to 90%. The WHO endorsed 93 ± 3% as medical-grade during COVID-19.

Some operators modulate purity in response to electricity pricing. Drop a point during peak tariff, save 10 to 15% on compressor load. The capability exists in the PLC firmware of several commercial systems. It sits unused on most installed units.

Gas above the mass transfer zone at the end of the adsorption stroke is compressed and oxygen-rich. Venting it wastes energy and product. Equalization connects both beds briefly. One stage recovers about 45% of reclaimable energy. A second gets another 15 to 20%. Each stage adds two automated valves and needs PLC timing under 100 milliseconds. The tuning is empirical. A process engineer sits at the panel during commissioning, tweaks valve timing in small increments, watches the analyzer respond. The optimal timing depends on bed geometry, packing density, pipe lengths, manifold dead volume. Two identical systems at different sites with different piping may need different equalization timing. Some commissioning teams do this properly. Others punch in defaults and leave.

The default works. The system produces oxygen. It just uses more energy per cubic meter than it should, and that penalty recurs for every hour the system operates for the next fifteen years. Budget manufacturers skip equalization. The annual energy penalty on a 200 Nm³/hr system is on the order of 60,000 to 100,000 kWh. The valve cost for a single equalization stage is two pneumatically actuated butterfly valves in the 50 to 150 mm size range, plus solenoid pilots, plus PLC I/O. A few thousand dollars in hardware. The energy savings over fifteen years on a medium system can be fifty thousand dollars or more.

A Roots blower pulls the regenerating bed to minus 0.5 to minus 0.7 bar gauge. Feed pressure drops to 0.3 to 0.5 bar gauge. Specific energy falls to around 0.32 kWh/Nm³. The crossover against standard PSA sits near 20 TPD, shifting with electricity cost and altitude.

The Roots blower itself is a positive-displacement machine with two counter-rotating lobes. It has no internal compression. It moves gas by displacement, and the pressure difference is created by the resistance of the bed and piping downstream. Roots blowers are mechanically simple, tolerant of dust, and have long service lives if the timing gears and seals are maintained. They are also loud. Acoustic enclosures or silencers are standard. In practice, the vacuum blower is the second-largest power consumer in a VPSA system after the feed blower, drawing 30 to 40% of total system power.

Vacuum flips the leak problem. On a pressurized system, leaks go outward. On VPSA, they go inward. Raw ambient air gets sucked into the regenerating bed. Valve stem packing, flange gaskets on large-diameter ductwork, compression ferrules on instrument fittings. Each admits a small flow. Together they can total the equivalent of a 2 to 3 mm orifice, which at minus 0.6 bar gauge admits around 20 liters per minute of untreated air. In tropical humidity, roughly 18 grams of water per hour reaches zeolite that cannot tolerate it. Degradation shows up within a few months.

Helium mass spectrometer testing finds these leaks. The technician pressurizes the vacuum circuit slightly above atmospheric, sprays helium around each joint, reads the spectrometer. A day’s work for the entire circuit. Soap bubble testing, which is what field crews usually reach for, misses sub-millimeter leaks. Startup is a separate issue. Full vacuum depth takes several cycles. During ramp-up, purity is below spec. Well-designed systems divert off-spec gas through a vent valve. Systems without that feature push contaminated oxygen into the downstream process on every restart.

The feed compressor draws 70 to 85% of system power. PSA needs the aftercooler-treated feed air. When a coalescing element ages past its replacement interval, the differential pressure rises. Operators delay replacement because the element is expensive and the compressor still runs. Oil breakthrough during that delay can increase by an order of magnitude.

The residual, whatever it actually is on a given day, deposits hydrocarbon on lithium cation sites over years. Those molecules carbonize in place. The nitrogen isotherm flattens progressively. The zeolite looks fine. It weighs the same. Output declines by a percent or two per year and operators call it aging. The only way to see oil fouling on zeolite is an adsorption isotherm on a pulled sample sent to a lab. This requires opening the vessel, pulling a representative sample from the correct depth, resealing, and finding a lab with the right equipment. Nobody does this as part of routine maintenance.

Oil-free screw machines cost 35 to 50% more and make the entire issue disappear. Zeolite lasts past fifteen years. The avoided replacement, somewhere in the 80,000 to 200,000 USD range, pays back the compressor premium. Centrifugal machines take over above 300 kW with better efficiency. Why oil-lubricated machines keep getting specified is a budgeting structure issue, not a technical one. The compressor buyer and the zeolite buyer sit in different departments.

Water adsorbs onto lithium cation sites more strongly than nitrogen and does not desorb during the pressure swing. The guard bed, activated alumina or silica gel at the air inlet end, intercepts moisture and regenerates each cycle. At steady state a saturation front oscillates within the alumina layer, never reaching the LiLSX.

The guard bed gets undersized because the manufacturer sizes it for a stated ambient temperature. Say 25°C. The aftercooler cools compressed air to within 8 to 15°C of ambient. At 25°C ambient the remaining moisture load is manageable. At 38°C ambient during a heat wave, aftercooler outlet temperature jumps and the moisture load exceeds what the alumina can handle in one cycle. The saturation front passes through and enters the LiLSX.

One bad week per year might not be fatal. Three bad weeks per year will be. After two or three summers the purity slips a quarter percent per month. Output drifts. The operator compensates. By the time someone compares current data to commissioning data, a substantial fraction of zeolite capacity is gone and the damage is permanent without thermal regeneration, which means taking the system offline.

A dewpoint transmitter between the alumina layer and the LiLSX would catch breakthrough before it reaches the main bed. Hardware and wiring cost a few thousand dollars. The reason this transmitter does not appear in most procurement specifications is that the specifications come from manufacturer templates, and a manufacturer has no commercial motivation to include an instrument that could reveal the limits of their own pre-treatment design.

PSA oxygen at 93% contains about 4% argon and 2% nitrogen. Argon is inert. It does not combust, does not react biologically, does not corrode. Oxy-fuel combustion sees a flame temperature difference of about 20°C between 93% and 99.5% oxygen, out of 2,800°C total. Fuel savings from removing nitrogen ballast run 40 to 60% regardless. Wastewater aeration, aquaculture, and gold cyanidation all work at 90 to 93% purity without meaningful performance penalty. Glass melting tightens the purity requirement. Nitrogen in the oxygen supply dissolves in the melt and nucleates as seed defects during fining. Container glass and commodity float glass tolerate the 1 to 2% nitrogen in PSA oxygen. Specialty glass with tight seed specifications does not.

Coarse bubble diffusers transfer maybe 20% of supplied oxygen into the water. The Speece cone, a pressurized downflow dissolution device, gets above 85%. The spread between those two numbers is so wide that it dominates the entire system economics.

Treatment plants that already have fine bubble diffuser grids from air-based aeration sometimes try to repurpose them for pure oxygen. Fine bubble diffusers produce 1 to 3 mm bubbles that rise through a 4 to 6 meter water column in 10 to 15 seconds. The higher oxygen partial pressure from 93% PSA oxygen does increase mass transfer rate per bubble, but the contact time is still too short for high transfer efficiency. Results end up around 35 to 50%, better than coarse bubble but not in the same category as pressurized dissolution. A feedback loop connecting a dissolved oxygen analyzer in the basin to the PSA control system through a 4-20 mA signal lets oxygen production track biological demand. It requires a signal wire and some PLC configuration. On most projects the PSA scope and the wastewater scope are written by different teams and the wire does not get run.

Oxy-fuel conversion raises flame temperature 500 to 700°C and halves flame length. Existing air-fuel burners produce the wrong flame. It goes short and concentrated. Crown overheats. Far end of the furnace gets less radiation. The inner surface temperature of a silica crown above an oxy-fuel burner can run 100 to 200°C above the same spot under air-fuel. Silica refractory life is exponentially sensitive to temperature. On a furnace expected to run eight to twelve years between rebuilds, crown overheating from burner mismatch can cut the remaining campaign in half.

Purpose-designed oxy-fuel burners exist from several manufacturers. The burner and crown modification costs 30 to 50% of the VPSA system cost and in most projects is not budgeted because the oxygen procurement and the furnace operations run through separate teams. VPSA delivers oxygen at 0.5 to 2.0 bar gauge. LOX vaporizers deliver at 4 to 8 bar. Burner nozzle geometry depends on supply pressure. Swapping supply source without recalculating the nozzle changes the flame, even at identical purity and flow rate. In some cases burners that worked well on LOX will not even ignite reliably on VPSA-pressure oxygen without nozzle modification.

Dissolved oxygen in a recirculating aquaculture system at commercial stocking density can drop to lethal levels within twenty minutes of supply interruption. Backup LOX with automatic switchover is a standard element in well-designed RAS packages. It costs 5 to 15% of the PSA system. Demand swings two to three times over a 24-hour cycle. A system without variable speed drive on the compressor runs at full load continuously and vents during low-demand periods. The energy waste during low-demand hours can be 30 to 40% of the daily electricity bill. Variable speed drives add 10 to 15% to the compressor cost and pay for themselves within a year or two on most aquaculture duty cycles.

The oxygen dissolution method in aquaculture matters as much as it does in wastewater, and the same hierarchy applies. Bubble diffusion into fish tanks or raceways wastes oxygen. Low-head oxygenators, oxygen cones, and pressurized sidestream injection achieve transfer efficiencies above 80%. The fish farm that buys a premium PSA generator and dissolves the oxygen through open-pipe injection into a raceway is wasting most of what it produces.

Air density drops 12% per 1,000 meters. A PSA system rated at 100 Nm³/hr at sea level delivers about 72 at 3,000 meters. At 4,000 meters, around 60. Zeolite capacity drops further with temperature. The factors multiply. The bidding dynamic at mine sites produces the same outcome repeatedly. Three manufacturers bid. The one using sea-level ratings quotes the smallest unit. Lowest price wins. The system arrives at altitude and produces 60 to 65% of expected output. The shortfall does not present as an alarm. It manifests as a gradual decline in gold recovery or slower leach kinetics, mixed in with all the other variables in a hydrometallurgical circuit. Months pass before someone instruments the oxygen supply line and finds the flow rate is below design.

Specifying output at actual elevation and temperature, with a commissioning test and financial penalties, prevents this. Standard in large VPSA procurement. Not standard for smaller systems, and most mine-site PSA units are small. What makes the mine-site situation different from other industrial applications is the remoteness. A glass plant or wastewater treatment facility can get emergency LOX delivery within hours if the PSA underperforms. A mine at 4,200 meters in the Andes cannot. The PSA system has to work as specified because there is no backup supply chain.

Oxygen delignification requires 90 to 95% purity at 8 to 12 bar. PSA output pressure is 0.5 to 4.0 bar. An oxygen booster compressor bridges the gap. Oxygen compression is its own discipline. High-purity oxygen at pressure is an aggressive oxidizer. Standard air compressor lubricants and seal materials can ignite. There have been fires. Oxygen-service machines need fluorinated lubricants, cleaned internals per CGA G-4.1 and EIGA Doc 13, and materials verification on every wetted surface. Pulp mills run around the clock at steady consumption, which keeps the PSA or VPSA at its design point where efficiency is highest.

Each main switching valve sees about a million actuations per year. Fifteen years, fifteen million. Seat life under clean dry gas runs five to eight million cycles. A worn seat lets product leak backward. The analyzer dips slightly. Output volume drops gradually. Operators check the compressor, check the zeolite. Both look fine. The diagnostic difficulty with valve leakage is that it mimics zeolite degradation. Both produce the same symptoms: gradually declining output and a slight purity drop. The two failure modes are indistinguishable from the control room unless the system has valve-specific diagnostics. Position feedback sensors confirm whether the valve is fully opening and fully closing. An automated leak test sequence, which isolates one valve at a time and monitors pressure decay across it, can quantify internal leakage. These diagnostics cost a few thousand dollars per valve. Budget control packages leave them out.

Conventional cycle times run 20 to 60 seconds. Rapid-cycle PSA puts thin zeolite layers on metallic or ceramic substrates in parallel-channel monoliths. Pressure drop falls 80 to 90%. Diffusion paths shorten tenfold. Cycles drop to 1 to 5 seconds. Five times the cycling rate, five times the throughput per kilogram of zeolite. The valve engineering is the constraint. At a 2-second cycle each valve actuates 1,800 times per hour. Pneumatic butterfly valves cannot keep up. Rotary and poppet designs are the alternatives. Gas distribution into a structured monolith also has to be uniform to within a few percent across the face, because unlike packed beds where maldistribution self-corrects through lateral redistribution, monolith channels are isolated from each other.

The coating process for structured adsorbent involves binder chemistry, dip or spray application, drying, and calcination to activate the zeolite without cracking or delaminating the layer. Consistent thickness at industrial volumes has not been solved. QuestAir Technologies (later Xebec, later Schlumberger) was an early developer. Air Products published lab results below 0.30 kWh/Nm³ at 93% purity. The boundary between PSA/VPSA and cryogenic ASU economics sits near 200 to 300 TPD. Adsorbent performance has improved 5 to 8% per decade for thirty years. Small cryogenic plants below 150 TPD are exposed.

Specifications get copied from a previous project or taken from a manufacturer’s template. The template favors that manufacturer’s standard package. A specification that does its job covers output at actual site conditions with financial penalties for shortfall, zeolite grade and replacement pricing, compressor type with lifecycle justification, number of equalization stages, valve specifications, guard bed dewpoint monitoring, commissioning test protocol, and turndown requirements. Most procurement documents include three or four of those.