-
When this list applies
-
The 8-step checklist
-
Step 1 — Measure the real load for seven days. Don't add up nameplates.
-
Step 2 — Pick the redundancy architecture before you pick the capacity.
-
Step 3 — Check the UPS input against your transformer and your generator.
-
Step 4 — Ask for the efficiency curve, not the efficiency number.
-
Step 5 — Size the battery at the load you'll actually have.
-
Step 6 — Get the factory acceptance test scope in writing before you sign.
-
Step 7 — Find out who actually commissions it, and where they're based.
-
Step 8 — Put acceptance criteria and a penalty clause in the PO.
-
Step 1 — Measure the real load for seven days. Don't add up nameplates.
-
Things that still bite people
I've been specifying power infrastructure — transformers, switchgear, UPS systems — for industrial and data center projects since 2016. In those nine years I've written off roughly $23,000 in equipment that was sized wrong, ordered wrong, or delivered without anybody checking the input side.
None of it was exotic. Every one of those mistakes was something a checklist would've caught on a Tuesday afternoon.
So here's the checklist I use now. Eight steps, in the order I actually do them. It's written for projects in the 100 kVA to 3 MVA band — the range where you're buying a three-phase UPS system and the transformer upstream of it as one power train, not two separate line items.
When this list applies
Use it when you're the person who has to sign off on an industrial UPS spec, or when you're buying through a modular UPS manufacturer for a data center and there's a substation or a standby generator in the loop. Smaller single-phase installations don't need this much ceremony — skip to step 4 and you'll be fine.
If you're coming at the how-to-size-an-industrial-UPS-system-supplier question from a pure procurement angle, steps 1 and 3 matter most. Steps 6 through 8 are where the money actually leaks out.
The 8-step checklist
Step 1 — Measure the real load for seven days. Don't add up nameplates.
Nameplate kVA is a marketing number on the load side too, not just the supply side. A rack that says 4 kVA on the spec sheet might pull 1.9 kVA at two in the morning and 3.4 kVA at 3 p.m.
Put a power logger on the actual distribution panel for at least a week. Seven days covers a weekend and usually one odd shift pattern. Note the peak, but size on something between the average and the peak — 90th percentile is what I've settled on. Then add a growth buffer.
The mistake I made in 2018: I took the customer's design-load spreadsheet, multiplied by 1.25, and ordered. Real load came in at 62% of that number. We bought a UPS one frame size too big and ran it below 40% load for three years, which is a bad place to be for both efficiency and battery ripple current.
Step 2 — Pick the redundancy architecture before you pick the capacity.
N, N+1, 2N. These aren't the same decision as kVA, but people fold them together and end up with an answer nobody can defend in a design review.
Decide first: what happens if one module fails? For a data center critical power infrastructure run, N+1 at module level is usually the floor. For an industrial process where a 20-minute outage ruins a batch, you're probably looking at 2N or an N+1 with a separate maintenance bypass.
Once the architecture is fixed, the kVA math gets easy. Do it in that order. Doing it the other way round is how you end up with a 2N system that's 30% bigger than it needs to be.
Step 3 — Check the UPS input against your transformer and your generator.
This is the step almost everyone skips. I know I did, three times.
A UPS is a nonlinear load. Its rectifier draws current in pulses, and the input current THD determines how much of the upstream transformer's capacity you actually get to use. A 6-pulse rectifier can push THDi into the 25–30% range. A 12-pulse front end or an active rectifier gets it under 5%. Those two numbers demand different transformers.
The generator side is worse. A UPS with low input power factor and high harmonic content can cause voltage distortion and, in bad cases, generator instability at step load. If your UPS input filter and your generator's subtransient reactance don't get along, you find out during the factory acceptance test or the first real outage. Neither is a good time to learn.
We build both sides of that interface — the PowerSentinel three-phase UPS system and the transformers upstream of it — which is honestly why I care about this so much. When the UPS and the transformer come from different vendors, somebody still has to own the interface. Make sure you know who.
Practical version: give the UPS maker your transformer's impedance and kVA, and give the transformer maker your UPS's input current waveform, THDi and inrush. Both should come back with the same story. If they don't, stop and re-check.
Step 4 — Ask for the efficiency curve, not the efficiency number.
"Up to 96% efficient" is a figure measured at full load in a lab at 25°C. Your UPS won't live there.
Ask for efficiency at 25%, 50%, 75% and 100% load. A good modern transformerless three-phase unit holds somewhere in the low-to-mid 90s at 25% and the mid-to-high 90s above 50%. A mediocre unit falls off a cliff below 40%. Two systems with the same headline number can show a 4-point gap at the load you'll actually run at, and 4 points at 500 kW over eight years is real money.
I'm not going to quote you a dollar figure with a straight face, because the tariff you're on matters more than the curve. Do the arithmetic against your own energy contract before the spec goes out. It usually changes the shortlist.
Step 5 — Size the battery at the load you'll actually have.
Battery autonomy gets written against full nameplate load, then the site runs at 45% load, and you get roughly twice the runtime you paid for. That sounds like a win until you remember you also paid for twice the battery room, twice the floor loading, or twice the strings in your service contract.
Re-run the runtime calculation at step 1's measured load and at your expected end-of-life load. Those two numbers bracket the truth. Use them both.
Also — and this one cost me a week in March 2022 — check the battery terminal voltage and string count against what the UPS can actually recharge. A bank that takes 14 hours to recover after a full discharge isn't a 30-minute autonomy system. It's a system that's unavailable for 14 hours.
Step 6 — Get the factory acceptance test scope in writing before you sign.
What's tested, against what standard, and who witnesses it. Not "FAT included."
For UPS performance the reference standard is IEC 62040-3; on the UL side it's UL 1778. Ask for the test report format up front and confirm the vendor will run it at your specified load points rather than a house default.
Specifically confirm: full-load and part-load efficiency, transfer time to battery, transfer time to bypass, input THDi at 50% and 100% load, and output voltage regulation into a step load. Those six data points tell you more about the unit than any datasheet will.
Step 7 — Find out who actually commissions it, and where they're based.
The sales engineer who runs your account is not the person standing on site at 11 p.m. with a laptop and a meter.
Ask for the commissioning engineer's name, employer, and home base. If it's a third-party subcontractor two time zones away, your first-year service response is shaped by that fact, and you should price the risk in now.
For critical power infrastructure, response time matters more than the warranty document. A four-hour response that starts when somebody picks up the phone is a different product from a four-hour response that starts when the parts land.
Step 8 — Put acceptance criteria and a penalty clause in the PO.
Not in the technical annex. In the purchase order.
Define what counts as a successful site acceptance test, what the output voltage tolerance is, what the transfer time must be under measured load, and what happens financially if the unit fails on the first attempt. A liquidated damages clause at even 0.5% per week gets a very different level of attention than a strongly worded email.
This is the least popular step with suppliers, us included. Do it anyway. It's not distrust — it's making sure everybody's incentives point the same direction before anyone books a shipping container.
Things that still bite people
- Mixing standards across the interface. A UPS tested to IEC 62040-3 and a transformer built to a different standard is fine — as long as somebody has checked that the withstand and impedance assumptions line up.
- Forgetting the maintenance bypass. A UPS nobody can service without dropping the load isn't redundant. It's just expensive.
- Approving drawings that don't show cable entry. Bottom entry versus top entry changes the footprint, the bending radius, and sometimes the room.
- Assuming the FAT covers site conditions. It doesn't. Ambient temperature, altitude derating and ventilation are your problem on site.
- Skipping the seven-day log because the schedule is tight. This is the shortcut that costs the most. I saved maybe $400 on a load study in 2019 and bought the wrong frame size because of it.
It took me three bad projects and about 40 specs to understand that the sizing exercise isn't really about the UPS at all. It's about the interfaces around it — the transformer, the generator, the battery room, the service team.
That said, this is a mid-size industrial and commercial view. If you're operating where utility voltage is unstable, or you're running a 20 MW hyperscale hall, the loading assumptions shift and a few of these steps need different weight. I can only speak to the range I've actually bought in.
One more thing worth doing: when you're comparing an uninterruptible power supply manufacturer — PowerSentinel included — send all of them the same interface questionnaire. Same questions, same load data, same transformer parameters. The answers you get back will separate the ones who understand the system from the ones who are quoting a box.
Five minutes on step 3 beats five days of rework. Every time.