When the Load Dropped Off a Cliff, Our Generator Forgot How to Behave

The load bank was humming at full load. Four hundred kilowatts, steady as a metronome, the 500 kVA set sitting at exactly the point it was built for.

Then somebody hit the main breaker. Not on purpose. A protection relay test on the switchgear side, wired to the wrong auxiliary contact. I heard the clack before I understood what it meant.

And the generator room changed its tune in about half a second.

The engine note jumped — not a scream, more like a yelp — the exhaust note climbing fast, then hunting, then climbing again. The frequency meter on the panel swung past 55 Hz. I remember thinking, with unusual clarity for a Tuesday morning: nobody ever tested this part.

We test generators for load acceptance all the time. Slam the load on, watch the voltage dip and recover, log the numbers, sign the paper. I’ve written that exact test into more commissioning plans than I can count. But a load falling off — the whole thing, instantly, at full output — that’s the test nobody runs. Until it runs itself, uninvited.

The engine doesn’t find out for a quarter of a second

Here’s the thing people miss. When the load vanishes, the engine is still making full-load power. Fuel racks wide open, turbo spooled, the whole rotating mass carrying all that energy. The load is gone, but the power isn’t. Not yet.

For a fraction of a second, there’s nowhere for that energy to go except into speed. The crankshaft accelerates. Frequency climbs. On a 50 Hz set, you’ll typically see it touch 52, 53, sometimes higher before the governor even wakes up. That’s not a malfunction. That’s physics outrunning electronics.

The governor does respond — a good electronic governor starts closing the fuel rack in well under a second. But it’s reacting to something that’s already happened. The overshoot is the price of admission. On older mechanical governors, the price is steeper and the recovery takes longer, and I’ve watched panels where the needle bounced like it owed the gauge money.

Voltage spikes first, and your electronics notice

While the engine is sorting itself out, the electrical side has its own bad moment. The AVR was holding voltage against a full-load field current. Suddenly the load’s gone, the stator current collapses, and that field is still there for a few electrical cycles. Voltage jumps. Typically you’ll see a spike of 10, 15, sometimes 20 percent above nominal before the AVR pulls the field back down.

Most rotating equipment shrugs this off. Motors don’t care about a 200-millisecond voltage party. But the control gear downstream — the PLC power supplies, the ATS controller, the building management panels — they’re the ones that file complaints. I once chased a “generator problem” through three site visits before somebody admitted the UPS kept logging overvoltage events every time the load bank test ended. The generator was fine. The load rejection transient at the end of every test was hammering the UPS inputs. We changed the test procedure, not the generator.

The test nobody writes into the spec

So here’s my honest recommendation, and I put it in writing now whenever a client lets me near the spec: test load rejection alongside load acceptance. Not as a footnote. As a line item.

It doesn’t need to be complicated:

• Run the set to full rated load and hold it until temperatures stabilize.
• Trip the breaker (a real breaker, at real load — a contactor on a resistive bank counts).
• Log frequency and voltage for the first ten seconds. Peak overshoot, time back inside the band.
• Do it at 50 percent too, because partial-load rejection behaves differently — less drama, sometimes a worse hunt afterward.

That last point surprises people. A full-load rejection is violent but honest; the governor knows exactly what happened. A half-load rejection can leave the engine hunting around its new setpoint like it can’t decide what it wants to be. I’d rather find that out in the factory yard than in a hospital basement at 2 a.m.

The old-timer disagrees with me, and he’s half right

My old mentor — thirty years on generators, hands like sandpaper — heard me ranting about load rejection testing once and laughed. “Engines like it,” he said. “Blows the carbon out. You worry too much.”

He’s not entirely wrong. A brief unload doesn’t hurt the diesel. No wet stacking from one event, no glazing from ten seconds of light load. The engine is the toughest thing in the room.

But he’s solving the wrong problem. The engine survives load rejection just fine. It’s the governor hunting, the voltage spike, the downstream electronics, and — this is the one that keeps me up — the protection settings that were never validated for the transient. An overfrequency trip set too tight will shut down a perfectly healthy generator the first time a real load rejection happens. And then your backup power “failed” during the event it was bought for, because of a setting nobody thought to check against the thing nobody tested.

That said, I still buy him tea when I see him.

Ask for both numbers on the nameplate’s paperwork

ISO 8528-5 doesn’t just define how a generator should take load — it defines performance classes for the whole transient, acceptance and rejection both. G2, G3, those ratings on serious quotes? They cover voltage and frequency behavior in both directions. If your supplier can only show you load-acceptance curves, you’re buying half the story.

Next time you’re writing a commissioning plan — or just buying a generator for something that matters — add the line. “Verify load rejection response at 50% and 100% of rated load. Record peak frequency and voltage deviation and recovery time.” It costs one breaker trip and ten seconds of logging. It catches the overshoot, the hunting, the mis-set protection, the downstream equipment that can’t take the spike.

Every generator I’ve ever commissioned passed the load test. The interesting ones were the ones that had to survive the load leaving.

If you want the loading-side version of this story — what happens in the first 200 milliseconds when load slams ON instead of falling off — start with our walkthrough of block load acceptance. The governor piece explains why that quarter-second delay exists in the first place. And if you’re writing the commissioning plan from scratch, the acceptance checklist covers what to log while you’re at it. Our load bank testing guide covers how to run the resistive side safely. Talk to us if you want the rejection test written into your spec before the factory ships the set — it’s a lot cheaper than discovering it at 2 a.m.

When the Load Dropped Off a Cliff, Our Generator Forgot How to Behave. It remembered, about four seconds later. I’d rather the spec remembered first.