Let me save you some money. I've handled Caterpillar generator orders for industrial clients since 2017, and I've personally made — and documented — enough significant mistakes to fill a small binder (the invoices are in there too). This is the checklist I wish I'd had before my first installation.
If you're buying a caterpillar-generator for a plant, a data center, or any facility that can't afford downtime, read this before you sign anything. Seven steps, in order, no fluff.
Most buyers focus on price per kilowatt and completely miss the cost of installation — concrete pad, ventilation louvers, fuel lines, exhaust runs, commissioning. In my experience, those add 30–50% to the total on a 750 kW install. So if your budget is $100,000 for the generator itself, plan on closer to $140,000–150,000 by the time it produces power. I've seen a plant manager's face when that invoice arrives. It's not a good look.
In 2019, I specified a 1250 kVA unit for a client that needed maybe 800 kVA. My logic was simple: bigger is safer, right? Wrong.
Caterpillar industrial generator sets carry different ratings depending on the intended duty. ISO 8528-1 lays out the categories:
Run a standby-rated machine at full load every day and you'll discover wet stacking (unburned fuel and carbon accumulating in the exhaust path, which leads to oil dilution and injector issues). It creeps up quietly and then fails at hour 300 of an actual outage when everything depends on that generator. From the outside, a bigger generator looks safer. The reality is an oversized standby unit can fail precisely when you need it.
And one more thing: check the difference between kVA and kW. A generator rated 1000 kVA at 0.8 power factor delivers 800 kW. If your site runs a lot of motors, UPS systems, or other inductive loads, the power factor matters just as much as the nameplate number. Get the kVA wrong and every calculation after it is wrong too.
What I do now: get the load profile first. The nameplate number isn't something I pick; it's something I justify on paper before ordering.
This is the piece almost nobody factors into the initial spec: site conditions.
Generator ratings are based on ISO standard reference conditions — roughly sea level and 25°C ambient. For every 300 meters above that, air density drops and a turbocharged diesel gives up some of its output. Add extreme heat on top and the loss mounts quickly.
In 2021, we commissioned a 3512 at a Phoenix site in July. It performed about 15% below spec because we didn't account for 45°C ambient plus heat radiating from adjacent process equipment. The client was not amused.
What I do now: I pull the derate curves from the manufacturer's datasheet before the purchase order is finalized. If altitude and ambient are against you, the answer is a larger frame or a different turbo configuration — neither of which can be added after delivery.
I once approved a day tank setup that looked perfect on the drawing. It worked fine until we tried to return excess fuel to the main tank and the return line turned out to be undersized. That cost $890 in redo plus a week of delay. $890 in parts, plus the conversation with the plant manager I'd rather forget.
The quieter killer is the fuel itself. Diesel that sits in a tank for months grows bacteria and algae. In 2022, a factory ran on a 3516 during an outage and shut down after nine hours because the fuel was contaminated. Nine hours. The root cause was a fuel polishing habit that didn't exist.
What I do now: fuel polishing goes on the maintenance schedule on day one, not after the emergency. It's a lot cheaper than a 2 AM filter change in the middle of an outage.
A diesel generator rejects a huge amount of heat — roughly a third of the fuel's energy comes out as heat from the radiator and the engine block. That heat has to go somewhere. If the room's louvers are undersized, the radiator just recirculates hot air until the engine trips on high coolant temperature.
Our first rental-power project had a 1250 kVA unit in a plant with louvers that looked correct on paper. The engine ran, the temperature gauge kept climbing, and it shut down after 45 minutes at load. The louver area was wrong for the radiator airflow. The drawing said it would work; physics disagreed.
What I do now: read the heat rejection data in the generator set manual instead of trusting the building spec from 1998. The louvers need to be sized for the radiator's rated airflow, the duct path should avoid sharp 90-degree turns, and the louver blades need motorized actuators that open when the generator starts.
An industrial generator doesn't work alone. It sits inside a larger power system with an automatic transfer switch, circuit breakers, and possibly paralleling gear if you're running multiple units.
A mistake I made on an early data center project: we sized the generator and never checked whether the ATS could switch the neutral. The client needed a separately derived system (in other words, a switched neutral) and we had to retrofit an extra contactor — plus a week of schedule delay.
In another 2020 job, I had two hours to decide whether to accept a delivery with the wrong control wiring. Normally I'd have refused the shipment and re-ordered. But the client's deadlines pushed me to say yes. We spent a week tracing wires on-site that should have been straightforward from day one.
What I do now: get the single-line diagram from the electrical engineer before ordering. If the generator will run in parallel with the utility or with other units, tell the manufacturer early. Paralleling changes control wiring, protection relays, and commissioning in major ways.
NFPA 110, the standard for emergency and standby power systems, requires an acceptance test at rated load before a system goes into service. Plenty of contractors skip it to save a few thousand dollars. I used to think that was a reasonable risk — until a unit passed every no-load check, we connected live load on go-live day, and it tripped within two minutes. A connection on the AVR (automatic voltage regulator) was wrong. No load test, no catch.
This is where a multimeter earns its keep. If you ended up here searching "how to set amp gain with multimeter," you're probably thinking of audio amplifiers — different skill, same tool. On a generator, the job is verifying voltage and current. Check phase-to-phase voltage and phase-to-neutral voltage, then clamp each phase conductor and compare amp draw against the nameplate. If the phase currents are unbalanced, stop and find out why before you bring the generator online.
What I do now: budget for a load bank test in every project. A portable resistive load bank (the rental kind) is the correct way to get a generator to 100% nameplate load safely. The upside was saving $2,000 by skipping it; the risk was discovering the problem at 2 AM during an actual outage. Now it's in the timeline, every time.
Nobody plans to need emergency engine parts. But a Caterpillar standby diesel generator is still a machine, and machines fail from neglect faster than they fail from use.
One of my lowest moments: ordering an oil filter and a set of fuses for a generator that had been down for four days, only to find the parts were on backorder for two weeks. The client's IT manager called me five times a day. The parts invoice was small; the damage to the relationship wasn't.
What I do now: before the purchase order is signed, I verify the parts list for the first two years of maintenance — air, fuel, and oil filters, belts, coolant hoses. They either sit in the stockroom or under a service contract. At $400 for a kit, it's cheap insurance for a site where a power loss stops money from being made.
I'm not going to pretend an industrial diesel generator is the right answer to every power question. It isn't.
Honestly, I'm not sure why some facilities rush through commissioning. My best guess is it feels like an extra cost — when it's actually the one test that tells you whether everything else on this list worked. That said, if you handle Steps 1–4 properly, the load test is usually boring. Boring is good.
Over the years I've handled maybe 40 generator projects. Maybe 38, I'd have to check the CRM. The ones that went sideways all share a pattern: someone skipped a step in this list because it seemed unnecessary or too expensive. The most expensive step is usually the one you think you can skip.