Five Paralleling Mistakes That Kill Good Diesel Generators

My phone buzzed at 2:47 a.m. Two 500 kVA sets at a logistics warehouse, paralleled for redundancy. One had tripped offline. The other was screaming along at a load it was never meant to carry alone, and the facility manager wanted to know — in colorful language — why his “redundant” system had just become very non-redundant.

I’ve seen this movie before. And here’s the thing: paralleling diesel generators isn’t complicated physics. Matching voltage, matching frequency, matching phase angle, then closing a breaker — technicians have done it for a century. What kills parallel systems isn’t the physics. It’s the five mistakes people keep making around it.

Mistake 1: Paralleling Sets That Were Never Meant to Be Twins

The warehouse had two 500 kVA gensets. Same rating on the nameplate. Different brands. Different engines, different alternators, different controllers — one with a decade on the clock, one fresh off the truck.

Look, you can parallel mismatched sets. The brochures swear you can. But the two machines need to agree on everything that matters: how fast their governors respond, how their AVRs behave under load steps, how their windings handle circulating currents. When they disagree, the stronger set ends up doing most of the work and the weaker one just rides along — or worse, starts motoring and trips on reverse power.

On projects I’ve been part of, the rule of thumb is simple: parallel identical sets if you can. Same engine, same alternator, same controller. It costs more upfront. It costs far less at 2:47 a.m.

Mistake 2: Leaving Load Sharing on Whatever the Default Was

Here’s where even identical sets get killed. Two generators in parallel have to split two kinds of load: real power (kW) and reactive power (kVAR). Real power sharing comes from the governors. Reactive sharing comes from the AVRs. Both need to be set for the same sharing strategy — and somebody has to actually set them.

I once watched a commissioning crew run a pair of 400 kVA sets for three hours without ever opening the load-sharing menu. One set was sitting at 70% load. The other was loafing at 30%. The overloaded one was drifting toward a high-winding-temperature alarm while its partner barely warmed up. All because the sharing gains were still at factory default and the droop curve wasn’t.

That said — defaults exist for a reason, and they’re fine for a single set. For a parallel pair, “fine” is a lie. Check the droop settings. Check the kW and kVAR sharing. Then load-step both machines together and watch the meters. If one set hogs the load, you’re not paralleling. You’re towing.

Mistake 3: Protecting Each Generator — and Forgetting the Bus

This one’s my favorite, because it shows up in drawings that look perfect.

Every set in a parallel system gets its own protection: overcurrent, reverse power, loss of excitation, the works. Then the two feeders land on a common busbar. And the bus — the single most expensive shared component in the system — gets protected by whatever breaker someone happened to have in stock.

A fault on the bus takes out everything downstream of it. Both sets trip, or the fault doesn’t clear fast enough and you lose the whole plant. I’ve stood in front of switchgear where the bus protection was literally an afterthought, scribbled in the margin of the drawing. Bus differential protection isn’t sexy. It also isn’t optional if you care about the word “redundancy.”

While you’re at it: think about what happens when one set fails while the other carries the load. That’s the whole point of N+1. But if set A’s failure knocks out the control power or the shared fuel supply feeding set B, you don’t have N+1. You have N. Draw the failure paths. All of them.

Mistake 4: Testing Every Set, Never the System

Every paralleling project I’ve audited had the same paperwork: each set load-banked individually, passed, stamped. And almost none of them had ever run the sets together under a real stepped load.

That’s insane when you say it out loud. A solo load bank test tells you nothing about how two sets share load, nothing about how the breakers sequence, nothing about what the controllers do when you slam 200 kW onto the bus. Load bank testing the combined system is the only test that matters for a parallel installation — and it’s the one most often skipped, because it needs the whole system wired up and somebody competent to run it.

Honestly? If your contractor tells you individual tests are enough, that’s the contractor you walk away from. A parallel system that hasn’t been tested in parallel is a parallel system in name only.

Mistake 5: The Cables Nobody Drew on the Drawing

The least glamorous mistake. And the one that causes the weirdest symptoms.

Parallel sets should connect to the bus with equal cable lengths and impedances. Sounds fussy. But unequal feeder runs create circulating currents between the sets — current that flows from one generator to the other and back, doing no useful work, heating windings, confusing the load-sharing controllers. I’ve measured circulating current on a 2×800 kVA installation that was big enough to push one alternator against its kVAR limit at half rated load.

Twelve meters of extra cable on one side did that. Twelve meters that somebody saved by running the shortest path instead of the equal path. Get the installation basics right and keep the electrical runs symmetrical — your future self will thank you at 2:47 a.m.

So When Is Paralleling Actually Worth It?

After all that, you’d think I’d tell you not to parallel. I won’t. Paralleling is genuinely the right answer when your load outgrows a single frame, when you can’t afford a single point of failure, or when your plant spends most of its time at 30–40% load and you’d rather shut one set down than wet-stack one giant machine. (Sizing it right is still half the battle — paralleling doesn’t fix an undersized system.)

It just demands more discipline than one big genset. Identical sets. Set the sharing properly. Protect the bus. Test the system, not the components. Keep the runs equal.

That warehouse at 2:47 a.m.? The fix took a day: reprogrammed the droop and sharing gains, adjusted the reverse-power protection, and ran the pair through a stepped load test together. The sets hadn’t been broken. The paralleling had just never been finished. Most of the time, that’s exactly what it is.