What Happens When Two Pumps Run in Parallel

Why this matters

Two identical pumps in parallel do not move twice the water. On a typical friction system they move about a tenth more than one pump does, and the customer who paid for the second pump is entitled to know that before it is installed rather than after. Worse, the second pump does not simply fail to help: starting it pushes both machines left along their curves toward the flow region where pumps damage themselves, so a badly staged pair can be quietly destroying two pumps to deliver what one was already delivering. This article is a worksheet you fill in before you commit, with every correction the arithmetic needs shown being applied.

Isolation on a parallel set is not just the disconnect

A pump sitting idle in a parallel set has live system pressure on its discharge. If its check valve leaks or is held open, the header will drive liquid backwards through it and spin the rotor in reverse with the motor disconnect locked open and proved dead. Locking out the electrical supply does not stop a parallel pump from turning. So before a guard comes off or a hand goes near a coupling: close and lock the branch isolation valves on both suction and discharge, relieve and drain the branch to a routed drain and confirm zero on a gauge, and confirm the rotor is stationary and blocked, all under 29 CFR 1910.147, alongside the electrical isolation and live-dead-live proving at the disconnect (29 CFR 1910.333(b)(2); NFPA 70E-2021, 120.5). A coupling guard goes back before power is restored (29 CFR 1910.219). Reverse rotation also has a mechanical consequence worth knowing: a threaded impeller nut or a threaded shaft component can unscrew when a rotor is driven backwards.

The combined curve is built by adding flow, not head

This is the whole construction and it takes one sentence. Pick a head. Read how much flow each pump passes at that head. Add the flows. That is one point on the combined curve. Repeat for a range of heads.

Because you add at constant head, the combined curve has the same shutoff head as a single pump and twice the flow at every head below it. It is a wider curve, not a taller one.

The operating point is then wherever that wider curve crosses the system curve, exactly as it is for a single pump, and the system curve has not changed at all. That is the source of every surprise in parallel operation: you doubled the pump and left the pipe alone.

Why the second pump adds less than a pump

The system curve rises with flow. Push more flow through the same pipe and the pipe demands more head, so the whole pair has to work against a higher head than one pump did, and at a higher head each individual pump passes less flow than it did alone. The steeper the system curve, the more of the second pump's contribution the pipe eats.

Which gives the rule worth carrying: parallel pumps pay off on a flat system curve and disappoint on a steep one. A system with a large static component - real elevation or a pressure held at the far end - has a flatter curve in the region of interest and gets more from a second pump. A system that is nearly all friction gets the least. This is the mirror of what static head does to speed control, where the static-dominated system is the one that disappoints, and the affinity-law article owns that half.

The shape of the piping, because it decides what you can isolate

 source
   |
   +--> pump one --> check --> branch valve --+
   |                                          |
   +--> pump two --> check --> branch valve --+
                                              |
                                     common header
                                              |
                                            system

Each branch needs its own check valve, to stop the header driving a stopped pump backwards, and its own isolation valve on both sides, so one machine can be worked on while the other runs. A set built with a single shared check valve at the header cannot be worked on at all without shutting the service down, and that is a decision made at installation that somebody lives with for twenty years.

The worksheet

Fill in these fields before a second pump is committed. Each one is a field because leaving it blank is a specific way the job goes wrong.

A workable field model of a centrifugal curve is head equals shutoff head minus a constant times flow squared, with the constant set so the model passes through the measured duty point. That model is what makes the rest of the sheet arithmetic rather than guesswork.

The filled-in worksheet: an all-friction system

Measured. One pump alone delivers 100 percent flow at 90 ft of head. Curve sheet gives shutoff head 120 ft and best efficiency flow at 100 percent. The manufacturer's continuous minimum flow is 40 percent of best efficiency flow. Static head is zero, so the system is all friction.

Curve model. 120 - 90 = 30 ft between shutoff and duty at a flow fraction of 1.0, so head = 120 - 30 x (flow fraction) squared for one pump. Combined for two pumps at the same head, the flow fraction of each is half the total, so head = 120 - 30 x (total / 2) squared = 120 - 7.5 x total squared.

System model. All friction, passing through 90 ft at a total of 1.0, so head = 90 x total squared.

The crossing. 90 x total squared = 120 - 7.5 x total squared, so 97.5 x total squared = 120, total squared = 1.2308, total = 1.109. Two pumps deliver 110.9 percent of what one pump delivered. The second machine bought 10.9 points of flow.

Correction one, branch losses, printed. The bare pump curve is not what the header sees. Each branch's check valve, isolation valve and pipe cost this installation 6 ft at a branch flow fraction of 1.0, and that loss rises with flow squared like any other friction. The 90 ft was measured on the running pump, so part of it is already being spent in that pump's own branch, and the shared header only takes 90 - 6 = 84 ft at a total of 1.0. Re-base the system to head = 84 x total squared BEFORE moving the branch loss onto the pump side, or the same 6 ft gets charged twice. The effective per-pump constant becomes 30 + 6 = 36, so the combined model is 120 - 36 x (total / 2) squared = 120 - 9 x total squared. Re-crossing: 93 x total squared = 120, total squared = 1.2903, total = 1.136. The gain RISES from 10.9 points to 13.6 points, because a branch loss is per-branch friction and each pump only meets it at half the total flow, so splitting the flow across two branches cuts that loss to a quarter of what the single pump was paying. Fold it onto the pump side without taking it back off the system side and you have counted it twice and got the sign of this correction backwards, which is the easiest mistake on the whole sheet. Small here, and not small at all on a set with a long branch run or a strainer in it, which is why the field is on the sheet.

Correction two, per-pump flow against minimum flow, printed. Each pump now passes half the total: 1.109 / 2 = 0.554, or 55 percent of best efficiency flow. Against the manufacturer's 40 percent continuous minimum, that clears - but note what happened. Running alone, each pump sat at 100 percent of best efficiency flow. Running as a pair, each sits at 55 percent. Starting the second pump moved both machines 45 points to the left, and it will do the same thing again if a third is added, which on this system would put each machine at roughly a third of best efficiency flow and under the minimum.

Correction three, the measurement, printed. The single-pump and two-pump flows were read on the same meter, at the same location, minutes apart. A meter's fixed calibration offset is a systematic error from one instrument, so it largely cancels in the ratio of the two readings, leaving only the repeatability spread, about 1 percent of reading on each. Those two spreads are independent, so they combine in quadrature: the square root of (1 squared plus 1 squared) is 1.4 percent. The measured gain is 10.5 plus or minus about 1.6 points, which agrees with the predicted 10.9 and is a real, if unexciting, improvement.

What the same worksheet says about a static-dominated system. Take the same two pumps into a system whose 90 ft at duty is 60 ft of static lift plus 30 ft of friction. The system model is head = 60 + 30 x total squared. Crossing against the bare combined curve: 60 + 30 x total squared = 120 - 7.5 x total squared, so 37.5 x total squared = 60, total squared = 1.6, total = 1.265. The second pump buys 26.5 points instead of 10.9, and each machine sits at 63 percent of best efficiency flow instead of 55. Same pumps, same head at duty, more than double the benefit, entirely because of what the system curve is doing.

What getting this wrong looks like. A shop sizes a duty-plus-standby pair, the customer asks whether both can run for peak demand, everyone says yes, and the peak arrives to a 10 percent improvement and a bill for a starter, a branch, and a control change. The number that would have prevented it is on line one of the sheet.

The mismatched pair, which is the other common failure

The two pumps do not have to be identical, but their shutoff heads have to be close, and here is why in arithmetic. Suppose the second pump has a shutoff head of 112 ft against the first pump's 120 ft, with the same curve constant.

The first pump running alone on the all-friction system above sits at 90 ft, so the second pump can contribute. But once both are running the head rises, and a mismatched pair has to be crossed on its own terms, because the combined curve is no longer symmetric: at a head H each pump passes the square root of its own shutoff minus H, over 30, while the system needs 90 times the total squared. That solves at about 106.7 ft and a total of 1.089, with the first pump at 0.67 of design flow and the second at the square root of (112 - 106.7) over 30, about 0.42 of its design flow. It is passing well under half of what it should and sitting barely over its own 40 percent continuous minimum with nothing in hand, so any further wear, any steepening of the system or any deeper mismatch puts it under, and it is being damaged by internal recirculation the entire time it runs.

From the outside this pump looks like it is working. It is spinning, its check valve is open, its discharge gauge reads full header pressure, and its motor is drawing current. The only reading that exposes it is flow in its own branch, which almost nobody has. When a parallel set has one pump that keeps needing repairs and one that never does, mismatched shutoff heads are the first thing to check, and the check is reading two curve sheets, not opening anything.

How to verify the pair is behaving

Measure total flow with one pump running and with both, at the same system condition, on the same meter, and compare the ratio against the prediction. If the measured gain is far below the prediction, the system curve is steeper than the model, which usually means a restriction that was not in the model, or one pump is contributing far less than its share.

Then check each machine individually: branch flow if there is a way to measure it, motor current on each, and discharge pressure at each pump before the header. Two pumps at the same header pressure with noticeably different currents are not sharing the load, and the one drawing less is the one running left of where it belongs.

References

• Pump manufacturer's curve sheet for each machine in the set, which owns shutoff head, best efficiency flow and continuous minimum flow
• ANSI/HI 9.6.3, the Hydraulic Institute guideline on rotodynamic pump operating regions, in the edition your engineering specification adopts
• 29 CFR 1910.147 for isolating the branch including the discharge, since electrical lockout alone does not prevent reverse rotation, with 29 CFR 1910.333(b)(2) and NFPA 70E-2021, 120.5 for proving the driver dead and 29 CFR 1910.219 for the coupling guard
• See related: What a Pump Curve and a System Curve Do Together; Why Minimum Flow Exists and What Happens Below It; What Series Operation Buys and What It Costs