How to Safely Parallel Multiple Air Compressors – Screw Air Compressors Manufacturer

Two compressors into one pipe. Simple to sketch, slow to get right.

Strap a data logger to the header for two weeks before buying anything. Most shops run their existing machine at half load with spikes a few times per shift. That usually means a VFD retrofit beats a second machine. Redundancy is arithmetic: cost of an hour without air versus cost of a standby unit.

The check valve gets most of the space in this article because it causes most of the damage when it is wrong, and because the specific ways it fails are poorly understood even among people who install compressors for a living. Everything else in a parallel system, the piping, the receivers, the electrical, the controls, matters. The check valve matters more.

Each branch line gets one. Spring-loaded poppet. After the aftercooler and moisture separator. Cracking pressure 2 to 5 psi.

Swing checks do not work in compressor discharge piping. The reason has nothing to do with build quality or brand. It is a fundamental mismatch between the valve's operating principle and the flow profile it encounters. A reciprocating compressor does not produce steady flow. Each compression stroke shoves a slug of air into the pipe. Between strokes, the instantaneous pressure at the discharge port drops below the line pressure for a fraction of a cycle as the piston reverses and the cylinder refills from the intake. This is not minor ripple on top of a steady signal. The pressure at the discharge flange swings above and below header pressure 20 times per second on a 1,200 RPM single-cylinder machine. A swing check flapper has low mass and no spring. It follows the oscillations. Not all the way to the seat, not all the way to full lift. It hovers at partial opening, vibrating, tapping the seat with each reverse pulse. Each tap is light. Over a million cycles per day the cumulative effect is a wire-drawn groove at the contact line, maybe two or three thousandths deep, invisible without pulling the valve apart. A static bench test with steady regulated air reads fine. Under pulsating service the valve leaks. Nobody discovers it until one compressor goes down for maintenance and pressure slowly bleeds backward through the valve that was supposed to isolate the idle machine.

Spring-loaded poppets hold against individual reverse pulses because the spring's closing bias exceeds the pulse energy. Parker Hannifin's application notes for pulsating gas service specify poppets. Swing checks belong in water lines.

Location in the piping is the second half of the check valve problem, and this one takes months to show damage instead of weeks, which makes it worse because by the time anyone notices, the aftercooler is already ruined.

A check valve upstream of the aftercooler traps hot discharge air between the compressor and the valve face during every shutdown. That trapped air is at discharge temperature. Around 300 degrees Fahrenheit on a single-stage machine compressing to 125 psi, give or take depending on valve efficiency and ambient conditions and how recently the intake filter was changed. At that temperature the air holds a large mass of water in vapor form. Machine stops. Trapped air cools overnight. Water condenses on the inside of the aftercooler tubes. Every shutdown, every night, 250 times a year in a one-shift shop. Carbon steel tubes pit from the inside at condensate pooling points in the lower rows of the tube bundle. Copper-finned cores last longer in the tube material but the braze joints between tubes and headers are the weak link. Dissimilar metals, condensate carrying dissolved CO2 from the compressed air making a weak carbonic acid solution, galvanic cell at the joint. Pinhole leaks at the tube-to-header braze joints in eighteen months to two years. Aftercooler core replacement runs 1,200 to 2,000 dollars depending on size and who makes it.

Put the check valve downstream of the aftercooler and separator. Now the aftercooler is on the header side, drains freely through the automatic drain. The trapped air between the compressor and the valve is on a short section of pipe that costs forty dollars to replace if it ever corrodes through.

Without a check valve at all, header pressure feeds backward into the stopped machine and the consequences depend on compressor type.

Reciprocating machines: 125 psi of reverse pressure reaches the piston through the discharge valves. Reed valves are spring steel strips around 20 thousandths thick. They flex open under forward differential millions of times. Sustained static reverse pressure bends them past yield. They crack at the clamping root or take a permanent set and stop sealing. Ring valves in larger machines survive the reverse pressure structurally but the return springs are undersized for 125 psi of static reverse load. Springs yield, rings unseat, air leaks past them into the cylinder, through the piston rings into the crankcase. On a two-stage machine the HP cylinder catches the full header pressure and the crankcase pressurizes fast enough to blow the oil fill cap across the room.

Screw machines lose an airend. Reverse airflow spins the rotors backward. The male rotor thrust bearing, a matched pair of angular contact balls shimmed during assembly for a specific directional preload, takes axial load from the wrong side. Balls skid instead of rolling. Race discolors from heat in seconds, spalls in minutes. A Quincy QGS-15 airend replacement runs around 4,500 dollars. The check valve that prevents it costs less than lunch.

This is the second most important topic and the one that produces the most confusing failures because the system shows no fault code when it goes wrong.

Lead machine gets the higher pressure band. Lag gets the lower one. The gap between them must exceed the lag pressure switch's hysteresis. That word, hysteresis, is doing a lot of work in that sentence and most installers have not looked up the hysteresis spec for whatever switch they bolted to the machine.

Hysteresis is the spread between a switch's actuation point on falling pressure and its reset point on rising pressure. On a Barksdale 96201, about 7 psi when new. Wider after years of vibration.

The failure: commissioning day, everything starts from zero, all switches in reset state, both machines cycle correctly. Thursday in production, the lag sits idle during a demand spike with a green light and no fault code. What happened is the lag ran earlier, unloaded, stopped. The lead took over, cycling between its own load and unload points. The lag switch needs pressure to rise above its reset point before it can fire again. If the system pressure during lead-only cycling never stays above that reset point long enough for the mechanical linkage to complete its travel, the switch is locked out. No alarm. No indication. The machine looks healthy from every diagnostic angle. This is a race condition between switch response time and system pressure dynamics, and it changes with demand patterns, which is why it works some days and fails others.

Configuration that works for 125 psi: lead loads 115, unloads 125. Lag loads 105, unloads 118. Lag reset point with 10 psi hysteresis is 115. Pressure reaches 125 every lead cycle, well above 115, switch resets.

Configuration that fails: lag unload at 122, cut-in at 112. Reset at 122. Pressure peaks at 125 for a fraction of a second each lead cycle before demand pulls it back down. If the switch cannot reset during that brief peak, the lag is locked out until someone depressurizes the entire system. Miserable to troubleshoot without knowing to look for it.

A sequencing controller with an electronic transducer has no mechanical hysteresis. No reset condition. No race. For tight pressure bands, a controller is the path with fewer surprises. Kaeser's Sigma Air Manager is the strongest product in this space for North American installations, largely because Kaeser's distributor network tends to attract people who commission systems rather than just selling boxes. The commissioning matters more than the hardware.

Startup timing adds another constraint. The lag takes 8 to 15 seconds (screw) or 30 seconds or more (large recip with heavy flywheel and wye-delta starter) from start signal to full delivery. During that window, pressure is falling and the lag contributes nothing. Lag cut-in must be high enough that pressure stays above minimum tool pressure through the entire startup. If tools need 90 psi, pressure decays at 1 psi per second, and the lag needs 20 seconds to come online, lag cut-in cannot go below 110.

Header sized for combined flow at under 20 fps. Pressure drop with the square of velocity. Two machines with 1-inch discharge connections need a 1.5 or 2 inch header. Ring headers over dead-end in any facility large enough to notice the pressure gradient, which in practice means anything over about 5,000 square feet. Compressors at opposite sides of the ring give the best distribution. Most shops put them side by side for maintenance convenience.

Slope horizontal runs at 1 inch per 10 feet toward drains. Drain legs 18 to 24 inches long at low points. Zero-loss electronic drains. Timer drains across multiple collection points bleed 3 to 5 CFM of air that cost full compression energy to produce.

Receiver downstream of the merge point. One gallon per CFM minimum. Wet receiver on each branch plus dry receiver downstream if the budget allows.

Each machine on its own circuit sized for locked-rotor amperage per NEC Article 440. Ten to fifteen second delay between starts. VFD on the lead if the budget allows, sized for the swing range so it runs at 70 to 90 percent speed.

PAO and PAG compressor oils are immiscible. Mixed through aerosol carryover in a shared header they form a gel. Kaeser's TB-109 covers this. Standardize on one oil. Having to scrape PAO/PAG gel out of a receiver tank is a half-day job with rags and patience because it does not dissolve in anything safe to pour into a pressure vessel. It has the consistency of cold honey and it gets everywhere.

Oil-water separator on all condensate from lubricated machines before sewer discharge.

A screw machine paralleled with a recip encounters a pulsation interference problem that does not appear when both machines are the same type and only manifests when both run simultaneously.

The recip's piston-frequency discharge pulses travel through the header and reach the screw machine's pressure sensing port. If the screw uses inlet-valve modulation, it reads these pressure oscillations as demand changes and oscillates its inlet valve trying to track them. The signal is piston noise, not demand information. The screw hunts for a stable operating point that does not exist. Output fluctuates. Energy consumption rises.

Shops have spent weeks chasing this. Each machine tests fine in isolation. The problem only appears with both running. It looks like a control system fault or a sensor issue. Replacing the transducer does nothing. Recalibrating the controller does nothing. The problem is upstream of both.

Tap the screw machine's sensing line from the system receiver instead of the header. The receiver's volume damps the pulsations below the control system's response threshold. Takes an afternoon. Alternatively a snubber orifice on the sensing line.

Screw on lead for modulation. Recip on lag, full-load-or-off only. Recips modulate through step unloaders that hold cylinders open in groups. The transitions are abrupt and the capacity jumps are coarse. A recip hunting between load stages on a gradually changing demand signal wastes energy and beats up the unloader linkage. Screws modulate smoothly through inlet valve or VFD speed control. Play to each machine's strength.

Each compressor needs one on the trapped volume between discharge and check valve. The receiver needs a separate one. ASME Section VIII.

Two motors at slightly different speeds produce a beat frequency. Isolation pads under each machine. Flexible connectors on pipe connections. Closer pipe support spacing with damped clamps.

Start the lead alone, lag isolation valve closed. Verify load and unload pressures with a calibrated test gauge. Panel gauges drift under vibration and the drift is always in the direction that makes the system look like it is working when it is not. Open the lag isolation valve with the lag motor off. Read the gauge between the lag and its check valve. Zero means the valve seals. Any reading means it leaks and gets replaced before the lag motor is ever energized. This test takes thirty seconds and it is skipped on the majority of parallel installations.

Start the lag alone. Same checks. Then both running, demand exceeding lead capacity. Time the lag from start signal to full delivery. Watch the header pressure gauge during that interval and note the lowest pressure the system reaches. If that number is below what the tools need, the lag cut-in is too low. Recalculate per the startup timing formula.

Drop demand. Verify the lag unloads before the lead. If they unload simultaneously or the lead unloads first, the bands overlap and need adjustment.

Test every automatic drain by manually triggering a cycle and watching for condensate flow. Pull the test ring on every relief valve at reduced system pressure. Leak-check every new pipe joint with soapy water or an ultrasonic detector. Write down every measured pressure and every switch setting. Tape the paper to the wall next to the machines.

— end —