The load bank operator counted down — three, two, one — and closed the contactor. Three hundred kilowatts hit the first 500 kVA set in a single step. The lights in the test bay blinked, the engine note dropped half a tone, and four seconds later everything sat rock steady. Voltage dip: twelve percent. Recovery: clean. Textbook.
Then we wheeled the cables over to the second set. Same rating — 500 kVA, same engine family, same shiny canopy. Same 300 kW step. This time the lights didn’t blink. They sagged, browned out, and stayed down for what felt like a very long six seconds. The contactor feeding the test load started chattering like a machine gun, and the frequency meter wobbled down toward 46 hertz before clawing its way back.
Same nameplate. Same load. Two completely different machines in the first ten seconds.
That was a commissioning job I ran a couple of years back — a factory expansion where the client had bought one set from us and one from a competitor, and, being a careful man, wanted both proven before he signed off. Smartest purchasing decision he ever made, because the nameplate kVA told him nothing about what he was about to see. The first ten seconds after a block load hits — that’s the part of the generator business nobody puts in the brochure.
Here’s the thing most people get backwards. When a big load slams on, the voltage doesn’t wait for the controller to “decide” anything. In the first few cycles — we’re talking a tenth of a second — the alternator is on its own. The magnetic field in the rotor can’t change instantly, so the machine delivers whatever its windings and iron allow. Engineers call the limiting factor sub-transient reactance. Plain English: it’s how stiff the alternator is.
A stiff alternator shrugs and gives up the current, and the voltage dips maybe ten percent. A soft one folds. The voltage can fall twenty-five, thirty percent before anything has time to react. And here’s the kicker: two alternators bolted to similar engines, both stamped 500 kVA, can have very different stiffness. The rating on the plate is a thermal rating — how much heat the windings can shed forever. It says nothing about the first 200 milliseconds.
I’ve watched a UPS transfer to battery because a generator’s voltage sagged for two seconds during a step load. The generator was “fine.” The nameplate was honest. The load still saw a brownout, and the UPS did exactly what it was built to do — protect itself from the generator. If that sounds familiar, I wrote about why UPS units keep dropping the generators they were bought to protect.
After the initial dip, the race starts. The AVR — automatic voltage regulator — sees the sagging terminal voltage and slams excitation to the ceiling, forcing current into the rotor field to drag the voltage back up. Meanwhile the engine governor feels the sudden load through the shaft, watches the speed fall, and pours on fuel. The turbo, which was idling along half asleep, takes a breath and spools up.
This is where sets separate. A good AVR with a strong excitation source — a permanent-magnet pilot exciter, a PMG, or an auxiliary winding — can force the field hard and fast. Voltage climbs back inside a couple of seconds. Frequency dips, the governor catches it, and by second eight or ten you’re back at 50 hertz and 400 volts like nothing happened.
A weak excitation system just… tries. A plain shunt-excited machine feeds its AVR from its own output terminals — which, remember, just sagged twenty-five percent. So the AVR is trying to recover using a power supply that’s collapsed. It gets there eventually. “Eventually” is doing a lot of work in that sentence.
And the engine side matters as much. A turbocharged diesel at no load is barely awake. Slam sixty percent load on it in one step and the turbo needs seconds to deliver boost. During those seconds the engine runs short of air, down on torque, and the speed sags. A fast electronic governor with good fuel mapping rides through it. A tired mechanical governor hunts — and step loads are exactly where governor hunting starts.
None of this is a secret. It’s just written in a code most buyers never learn to read: the ISO 8528 performance classes. G1, G2, G3 — you’ve seen them on data sheets, probably right next to the kVA number you actually looked at.
G1 is the loose class. Basic lighting, simple loads, the kind of site where a flicker becomes an anecdote. G2 is commercial — offices, workshops, the loads most of us actually run. G3 is the tight class: telecom, data, hospitals, anything where the load notices a wobble. Each step up demands smaller transient deviations and faster recovery times. The standard literally grades how gracefully a set picks up a block load.
Here’s what I’d tell a friend: when you’re comparing two quotes and the kVA matches, look at the G-class line. If one vendor’s data sheet claims G3 transient performance and the other doesn’t mention a class at all, you’re not comparing two 500 kVA sets. You’re comparing a set that’s been tested against a standard with a set that hasn’t. That second set in my story? No G-class on the data sheet. I should have read that as a warning instead of a blank.
An old-timer I worked with for years used to say “kVA is kVA, son.” He was a fine engineer and wrong about this one thing. I told him so exactly once, after a step-load test, and he never said it again. The rating tells you what the machine can carry. The class tells you how gracefully it picks the load up.
This is the part I push hardest with clients, because it’s cheap and it settles arguments. A step-load test with a load bank — apply a big block of load in one step, typically half to two-thirds of rated kVA, and record what happens. Voltage dip percentage. Frequency dip. Recovery time back inside the steady-state band. Three numbers. That’s the whole test, and I walked through the method in the load bank testing guide.
You don’t even need fancy gear for a rough version. Most modern controllers log minimum voltage and frequency during events — pull the event log after a monthly test where a real load transfers. But a proper witnessed test with a power analyzer, numbers written into the acceptance protocol, is better. Write the acceptance limits into the purchase contract: max voltage dip, max frequency dip, max recovery time, at a defined step size. If the vendor balks at that clause, you’ve learned something valuable before spending a cent.
I insist on this for any site with motor loads, and doubly for sites with a UPS downstream. The UPS doesn’t care about your nameplate. It cares about the waveform it sees, and it will abandon a generator that sags too far, exactly when you need it most. Test the pair together if you can — generator, transfer switch, UPS, real load. The system is the product, not the box. And if your plant starts big motors across the line, read the motor starting guide too — starting current is what really sizes the set, and transient response is what decides whether the start succeeds.
If you’re buying and you remember nothing else from this piece, put these four lines in your specification:
Back at that factory, the client kept set number one and sent set number two back with a polite letter and a copy of the test report. The vendor’s salesman called it an unfair test. Sixty percent in one step, he said, nobody does that in real life. The client’s maintenance chief just pointed at the motor control center across the yard — three 90 kW motors on across-the-line starters, all on one transfer switch. “That’s Tuesday,” he said.
He was right. Real life doesn’t ramp loads gently. It slams contactors. Spec for the slam, test for the slam, and the nameplate kVA can take care of itself.
Dealing with motor loads, a UPS that keeps dropping your generator, or a data sheet full of numbers nobody explained? Talk to our engineers — we’d rather argue about step loads before the purchase order than after the blackout.
Email：[email protected]
Work：Monday to Friday,9:30-18:30
