Commissioning day. Two 500 kVA sets, paralleled onto a load bank dialed to 800 kW. The kind of morning where everything is supposed to just work.
Set A was carrying 620 kW. Set B was carrying 180.
Nobody said anything for a few seconds. Then the client’s engineer looked at me, and I looked at the gauges, and the gauges just sat there, perfectly calm, showing us a lie we’d all agreed to believe: that two generators in parallel share the load. They don’t. Not automatically. Not ever. They negotiate it — constantly, silently, through governors and voltage regulators that each think they’re in charge — and when the negotiation breaks down, one machine does the work of two while the other one loafs along, burning fuel and wearing out for nothing.
I’ve watched this scene play out in just about every flavor. And after twelve years, I’ve stopped being surprised. Uneven load sharing is the default state of paralleled generators. Even sharing is the achievement.
Here’s the thing most people get wrong on day one. “Load sharing” sounds like one job. It’s two. Completely separate mechanisms, completely separate hardware, and they can fail completely independently of each other.
Active power — kW, the stuff that actually does work — is divided by the governors. More fuel to one engine, it takes more of the real load. That’s it. Simple.
Reactive power — kVAR, the magnetizing current sloshing back and forth — is divided by the AVRs, the automatic voltage regulators. Whichever machine holds its voltage a hair higher pushes out more reactive current. Also simple. Separately simple.
And here’s where it gets fun: you can have beautiful kW sharing and catastrophic kVAR sharing at the same time. I once stared at a paralleling switchboard showing 400 kW on each of two sets — textbook — while one machine’s ammeter read nearly double the other’s. Same real power. Wildly different currents. The client was proud of the kW balance. The alternator on the overloaded machine was cooking in its own reactive current, and nobody was watching the right meter. kW balance tells you the engines are sharing. It tells you nothing about the alternators.
That extra current has a name: circulating current. It flows between the two generators, round and round, doing absolutely no useful work for the load. It just heats copper.
It happens when the two machines disagree about voltage — even by a little. Set A’s AVR thinks the bus should be 401 volts. Set B’s AVR is holding 398. That three-volt argument drives a current that circulates between the alternators, limited only by their internal impedance, which is small by design. The load never sees it. The fuel gauges never see it. The winding temperature sensors see it, eventually, and by then you’re reading the story backwards.
The fix, in theory, is cross-current compensation — CTs (current transformers) on each machine feeding a sample of the other’s load current into its AVR, so each regulator backs off when it sees itself pushing too hard. In theory it’s elegant. In practice, it’s a wiring diagram with about forty ways to be wrong, and I’ve met most of them. A reversed CT doesn’t just fail to help — it actively lies to the AVR, telling it to push harder when it should back off. The compensation becomes an accelerant. I’ve seen a set with a flipped cross-current CT run its field current to the ceiling while the operators congratulated themselves on how stable the voltage looked.
An old colleague — thirty years on engines, and he has a saying for everything — insists droop is all you ever need. “Set both governors to 3% droop and walk away,” he says. “They’ll share. Physics does it for free.”
He’s half right, which is the most dangerous kind of right.
Droop governing does share kW naturally: as load rises, frequency sags a little, and each engine takes load in proportion to its droop setting. No communication between sets needed. Beautiful. The catch is the sag — your 50 Hz bus is now running at 49, or 48.5 under heavy load, and everything downstream gets to live with it. For a construction site, fine. For a data center with UPS systems watching frequency like hawks, not fine.
Isochronous governing holds frequency dead flat — but two isochronous governors in parallel will fight like cats, because neither will yield. So you add load-share lines: the governors talk to each other over dedicated wiring and agree on the split. Works great until someone mixes modes — one set in droop, one in isochronous — and then the isochronous machine tries to carry every load change by itself while the droop machine watches. Worst case, one set ends up motoring the other: the generator becomes a motor, spinning its diesel engine from the bus, and the engine sits there being dragged along, burning nothing, producing nothing, while its protection wonders what’s happening. Reverse-power protection should catch it. “Should” is doing a lot of work in that sentence — I’ve found reverse-power relays with the wrong pickup setting more than once, set by someone who copied the numbers from a different job.
My take, for what it’s worth: droop for simple sites where frequency wobble is tolerable, isochronous with proper load-share lines where it isn’t, and never mix them unless you enjoy excitement. My colleague still disagrees. He’s allowed. His sites are construction camps. Mine have UPS rooms.
If I had to bet on one cause for every load-sharing complaint I’ve ever been called about, I’d bet on the current transformers. Every time.
CTs are humble little doughnuts of iron and copper, and the entire paralleling control scheme trusts them completely. Reversed polarity on a cross-current CT: the AVR gets told the opposite of the truth. Wrong ratio programmed into the controller — the CT says 1000:5 but the controller thinks 800:5: every reading is off by 25% and the sharing math quietly drifts. Loose burden resistor: the signal sags. And the one that keeps me up at night — an open-circuited CT secondary, which can develop thousands of volts across the terminals and has started fires. I check CT shorting links the way other people check their mirrors. Habit, now.
The commissioning-day story I opened with? Forty minutes of head-scratching, and then a junior tech — the quiet ones always find it — pointed at the cross-current CT on set B and said the white wire and the black wire looked swapped compared to set A. They were. Two wires, swapped back — loads snapped to 400 and 400 like they’d been waiting for permission. The whole morning’s mystery was six inches of copper in the wrong terminals, and I still think about that whenever someone tells me paralleling problems are complicated. Sometimes. Sometimes it’s two wires.
People ask me for a number — how close is close enough? Here’s my field rule, not a standard, just what I’ve learned to live with: kW within about 10% of each other, currents within about 10%, and power factor readings that roughly agree. If one machine’s power factor is 0.8 and the other’s is 0.4, your kVAR sharing is broken no matter what the kW gauges say. Watch the ammeters, not just the wattmeters. The ammeters never lie about heat.
And test it under real load, not just at the load bank’s polite 50%. Sharing errors scale with load — a 5% imbalance at quarter load becomes an ugly one at full load, right when you can least afford it. The load bank test is where you prove it, step by step, watching both sets at every step. If you only ever test each machine alone — and plenty of sites do — you’ll never see the argument at all. You’ll just wonder why one engine always seems tired.
One more thing, because it connects to something I keep writing about: the set that loafs. A machine chronically carrying 20% load while its partner sweats isn’t just inefficient — it’s a wet stacking candidate: light load, cool cylinders, unburned fuel glazing the bores. And bad sharing doesn’t just waste fuel today. It shortens the lazy machine’s life, quietly, over years. The paralleling gear gets the blame for the complexity; the maintenance budget pays for the consequences.
So no, paralleled generators don’t share the load. They negotiate it, through governors, AVRs, CTs, and settings that all have to agree with each other. When the negotiation works, it’s invisible — two machines humming along at 400 and 400, and nobody thinks about it at all. That’s the goal. Invisible.
If your paralleled sets aren’t splitting evenly — or if you’re planning a parallel installation and want the negotiation settled before commissioning day — talk to us. We’ve chased the two-wire mysteries before. Usually it doesn’t take forty minutes the second time.
Email：[email protected]
Work：Monday to Friday,9:30-18:30
