Three years ago I got called to a packaging plant on the edge of town — the kind of place with a humming production line, a server cabinet in the corner of the office, and a brand-new 150 kVA standby diesel set out back that the owner was very proud of. The complaint was strange: every time the grid dropped and the generator took over, the packaging line’s variable-frequency drives tripped, the office servers rebooted, and the owner’s brand-new LED high-bay lights flickered like a nightclub. The generator itself ran beautifully. Smooth. Quiet. Started in nine seconds flat. The power coming out of it, though, was filthy.
That day I learned a lesson I now repeat on almost every site visit: a generator can be mechanically perfect and still produce electricity your equipment hates. The measure of it is THD — total harmonic distortion — and it’s the spec nobody reads until something expensive starts acting weird.
Grid power arrives at your building as something close to a perfect sine wave — smooth, symmetrical, boring. A generator is supposed to reproduce that wave. THD measures how badly it fails. It adds up all the little extra ripples — the harmonics — riding on top of the 50 or 60 Hz fundamental, and expresses them as a percentage of it.
Think of it like water. Utility power is a calm canal. A cheap generator is a canal with speedboats zigzagging through it. The water still flows. But try floating something delicate on it and you’ll see the difference fast.
What’s a good number? Under 5% is the figure you’ll see quoted on most genset spec sheets — and honestly, for motors, heaters, and lighting, under 5% is fine. Equipment built for the grid lives around 3% or less. But here’s the catch nobody prints in bold: that “under 5%” figure is measured at full linear load. Your load is not linear. And under real, messy, nonlinear loads, the same machine can throw off far worse numbers. The brochure tests the generator against an easy exam. Your site gives it the hard one.
I’ve commissioned maybe two hundred generator sets, and I can count on one hand the number of quotes that volunteered a THD figure for nonlinear loads. Zero is closer to the truth. Sales sheets quote it at full linear load because that’s the flattering number. It’s a bit like quoting a truck’s fuel economy downhill.
Last spring I was at a rural clinic — one of those jobs where the generator backs up vaccine fridges, a small lab, and the office computers. The 45 kVA set passed its factory test with flying colors. On site, though, the lab’s centrifuge controller kept throwing errors and the billing PC’s power supply died twice in three months. Both died on generator power, both lived happily on the grid. The owner’s first theory was bad luck. My meter said otherwise: distortion well into double digits whenever the lab equipment was running. The generator was fine. The waveform wasn’t.
Look, I’m not saying every site needs laboratory-grade power. Most don’t. But if your load list includes anything with a chip in it — and whose doesn’t, in 2026 — you should know what your generator’s waveform looks like before you find out the hard way. Ask for the number at nonlinear load, in writing. The good manufacturers have it. The ones who go quiet when you ask? That silence is your answer.
The damage is rarely dramatic. Nobody’s generator explodes from bad THD. Instead you get the kind of problems that eat a maintenance budget one small invoice at a time.
Motors run hotter and hum louder than they should. Neutrals carry current they were never sized for — I’ve measured neutral current at nearly double the phase current in an office building on generator power, and the electrician on site went pale. Transformers buzz. Circuit breakers trip for no reason anyone can find. UPS units — the things bought to protect the servers — sometimes refuse the generator outright and stay on battery until the batteries die, which is the cruelest joke in backup power and exactly what our UPS compatibility piece digs into.
The sneaky ones are the electronics. PLCs that reset mid-cycle. Medical imaging that throws artifact errors the radiologist can’t explain. LED drivers that fail a year early. None of it points at the generator at first, because the generator is running fine. It’s producing voltage. It’s producing frequency. It’s just producing a waveform your 1980s motor wouldn’t have minded and your 2026 electronics can’t tolerate.
Three things decide a generator’s waveform, and only one of them is the engine. First, the alternator design — its subtransient reactance, to use the proper term. A low-reactance alternator shrugs off nonlinear loads; a high-reactance one distorts under them. This is the spec that separates a budget set from a good one, and it’s the one buyers never ask about because they’ve never heard of it.
Second, the automatic voltage regulator. The AVR’s job is to hold voltage steady as loads slam on and off, and a slow or crude regulator lets the waveform wobble every time a big motor starts. We’ve written about what the controller and its electronics do for you — the AVR is the part of that brain doing the fastest, most invisible work.
Third, sizing. And this is the part that surprises people: an oversized alternator on a modest engine produces cleaner power than a perfectly matched one. The old field rule is to size the alternator at two and a half to three times the nonlinear load. It sounds wasteful. It’s not — it’s the cheapest harmonic filter you’ll ever buy. A generator that’s too small doesn’t just struggle to start motors; it distorts everything it feeds while trying.
There’s a fourth option for small sensitive loads: inverter generators, the kind with electronic output stages that hold distortion under 3% almost regardless of load. They’re wonderful for a clinic or a comms shelter. They’re also capped at modest sizes, which is why they don’t solve the packaging-plant problem. Different tools, different jobs.
If I were writing the spec for a site with real electronics on it — a data center, a hospital wing, a plant full of drives — here’s what I’d actually do. Demand the THD figure at nonlinear load, in writing, not the full-linear-load brochure number. Size the alternator generously against the nonlinear portion of the load; the 2.5-to-3x rule of thumb has never embarrassed me. Put a harmonic filter on the quote as a line item you can delete later, not an afterthought you discover you need. And test the set under your real loads, not a resistive load bank — a load bank proves the engine works; only your actual equipment proves the power is clean.
And here’s the uncomfortable opinion: for a surprising number of sites, the “standard” generator on the cheapest quote is the wrong machine. Not because it’s badly built, but because it was specified for motors and lighting in a world that now runs on electronics. The cheapest quote assumes your loads are simple. They aren’t anymore.
The packaging plant? We upsized the alternator, added line reactors on the worst drives, and the flickering stopped. The owner still has the same engine. Same fuel bill. Same nine-second start. The power is just clean now. Three years on, nothing has tripped since — and the owner tells the story at industry dinners as if he’d known about THD all along. I let him. Everybody deserves to be the expert in their own story. If you’re speccing a set for sensitive loads, talk to us before you sign the quote — the waveform question takes five minutes and saves years of weird failures.
Email：[email protected]
Work：Monday to Friday,9:30-18:30
